url
stringlengths 33
62
| title
stringlengths 10
115
| author
stringclasses 2
values | content
stringlengths 61
41k
| comments_count
int64 0
1.68k
| scraped_at
float64 1.76B
1.76B
|
|---|---|---|---|---|---|
https://www.phoronix.com/news/AMD-Releases-AOMP-14.0
|
AMD Releases ROCm AOMP 14.0 Compiler - Switches To New "amd-stg-open" Branch
|
Michael Larabel
|
AMD released AOMP 14.0 during SC21 week as the newest version of their LLVM/Clang-based compiler providing OpenMP GPU offload support for Radeon graphics processors.
AOMP 14.0 was released this morning as the newest version of this patched-up version of LLVM/Clang that gets OpenMP GPU offload into good shape with Radeon GPUs and AMD Instinct accelerators.
With AOMP 14.0, they have switched to using a new code branch for their LLVM/Clang base. They have switched over to using "amd-stg-open" as a new staging area for code from AMD that they are making publicly available but not yet ready (or otherwise not appropriate/applicable) for upstreaming into the LLVM monorepo. They explain of amd-stg-open in today's announcement, "The amd-stg-open branch is constantly changing as AMD merges upstream development trunk with its internal open development efforts. Some AMD modifications are experimental and/or under review for the LLVM mono-repo. The AOMP release is a snapshot of amd-stg-open and supporting repositories to build various components."
The amd-stg-open branch itself for those curious about AMD's LLVM modifications can be found via this GitHub repository separate from the AOMP repository.
In addition to AOMP 14.0 shifting to use this branch and incorporating code from upstream LLVM/Clang 14.0 Git, the new release pulls in Radeon ROCm 4.5 components, a new HIP-based build method is used, there is support for unified shared memory with Vega GFX90A hardware, support for the atomic hint clause added to support fast floating-point atomics, and a variety of other improvements.
Those wanting to try out this AOMP 14.0 Radeon OpenMP offloading compiler can find the sources as well as CentOS/RHEL, Ubuntu, and SLES 15 binaries via GitHub.
| 15
| 1,760,719,345.896011
|
https://www.phoronix.com/news/AMD-Zen3-Microcode-Nov-2021
|
AMD Releases Updated Zen 3 CPU Microcode (November 2021)
|
Michael Larabel
|
A new Family 19h microcode binary was merged today into the linux-firmware.git repository that serves as the central source for all of the binary firmware/microcode files for Linux systems.
The updated AMD Family 19h "Zen 3" microcode was committed today to linux-firmware.git. Unfortunately, as usual, there isn't any public change-log to note what has changed with this AMD CPU microcode revision.
This update though does come just days after the AMD-SB-1021 security bulletin. That AMD security bulletin for November 2021 noted a wide number of CVEs (including several "high" severity) issues affecting their EPYC processors. Those CVEs included vulnerabilities in the PSP, SMU, and other CPU-related issues.
AMD is releasing new AGESA for mitigating those issues and should be pushed down soon as BIOS updates by various server/motherboard vendors. I'm waiting to hear back if today's CPU microcode update is related to last week's disclosures or if there is anything else of note with today's revision.
| 6
| 1,760,719,346.772616
|
https://www.phoronix.com/news/Linux-5.16-AMD-SEV-ES-Intra
|
Linux 5.16 To Support AMD SEV/SEV-ES Intra-Host Live Migration
|
Michael Larabel
|
Last week was the main set of Kernel-based Virtual Machine (KVM) changes for Linux 5.16 that introduced RISC-V hypervisor support and AMD PSF control bit support, among other changes. A second set of KVM changes were sent out on Friday that is headlined by having AMD SEV/SEV-ES intra-host migration support.
With this secondary set of KVM updates for Linux 5.16, the mainline kernel can now handle intra-host migration of virtual machines leveraging Secure Encrypted Virtualization (or SEV-ES, the Encrypted State additions introduced with EPYC 7002 Rome). Live migration hasn't been supported due to the complexities and security with Secure Encryption Virtualization while now at least intra-host migration is supported for where the source and destination VM are on the same underlying server (inter-host migration is not).
This AMD SEV intra-host migration required introducing a new KVM guest API and guest kernel support changes for handling SEV live migration and then the SEV/SEV-ES host migration code changes.
The full list of KVM changes sent in yesterday can be found as part of this pull request.
As of Linux 5.16, the SEV-SNP "Secure Nested Paging" additions with EPYC 7003 "Milan" processors still haven't been upstreamed. AMD continues working on upstreaming the SEV-SNP support to the mainline kernel but is still an ongoing matter. Hopefully it won't be too much longer before seeing that SEV-SNP support ready in full for mainline while until then AMD continues to distribute the patches via their own source tree.
| 3
| 1,760,719,347.837795
|
https://www.phoronix.com/news/AMD-P-State-Linux-v3
|
AMD Posts Latest Revision To Its New P-State Linux Driver
|
Michael Larabel
|
In early September AMD posted their new "amd-pstate" CPU frequency scaling driver for Linux that leverages ACPI CPPC data available with Zen 2 and newer processors for making wiser frequency scaling decisions. The goal of AMD P-State is to offer better performance-per-Watt and today they have posted a new revision of this driver.
In late September AMD posted the amd-pstate v2 patches while now in ending out October they have posted amd-pstate v3. With this updated patch series, they still are limiting the driver compatibility to select Zen 3 processors while with time they plan to broaden it up to the rest of the Zen 2 and Zen 3 processors.
New to the amd-pstate v3 series is a patch from Valve to allow using SystemIO for _CPC registers. This patch should help some platforms like the ASUS Zephyrus G15 that use a SystemIO register for the highest performance _CPC element. The v3 patches also have a variety of fixes, updated documentation, cleaning up some of the APIs, clearing up some elements of the sysfs interface, and other updates.
The amd-pstate v3 patches are out for review and testing on the kernel mailing list. Though given the timing of these patches and still needing to undergo a fresh round of review, it's probably unlikely to see this driver land for the upcoming Linux 5.16 merge window especially if it does indeed open up on Sunday rather than being pushed back by one week. That's too bad for this driver, which was developed in cooperation with Valve as part of the Steam Deck and other AMD Linux developments, but hopefully it will get ready for mainline by the 5.17 cycle.
| 13
| 1,760,719,350.749273
|
https://www.phoronix.com/news/AMD-Zen-4-Genoa-Bergamo
|
AMD Shares Early Details Of Zen 4 Genoa, Bergamo
|
Michael Larabel
|
In addition to announcing Milan-X processors at the virtual Accelerated Data Center Premiere event, AMD just provided some new public details concerning next-generation Zen 4 processors.
The key takeaways from today's disclosures around the Zen 4 road-map include:
EPYC "Genoa" processors will feature up to 96 high performance Zen 4 cores with DDR5 and PCIe 5.0 support. Zen 4 as expected uses the TSMC 5nm process.
EPYC "Bergamo" is a high-core count compute engine designed for cloud-native workloads and other environments. Bergamo will feature up to 128 Zen 4 "C" cores. Bergamo will carry all of the same Genoa features like DDR5, PCIe 5.0, CXL 1.1, and more. Bergamo is socket-compatible with Genoa. The "Zen 4 C" is software-compatible with Zen 4 but with enhancements for cloud workloads and power efficiency enhancements.
Look for more information on Zen 4 desktop and server platforms in 2022 and the expected launch later in the year. Bergamo isn't expected until H1'2023.
| 18
| 1,760,719,351.216983
|
https://www.phoronix.com/news/AMD-Smart-Trace-Buffer-Linux
|
AMD "Smart Trace Buffer" Feature Being Wired Up For Linux
|
Michael Larabel
|
AMD patches this month have begun preparing support for a new CPU feature called Smart Trace Buffer (STB).
Earlier this month the initial patch was posted for enabling AMD Smart Trace Buffer support on Linux as an extension to AMD's PMC kernel driver. This morning the STB enablement has already been revised in a second version of the patches.
AMD Smart Trace Buffer is described by the patches as "a debug trace buffer which is used to help isolate failures by analyzing the last feature that a system was running before hitting a failure. This nonintrusive way is always running in the background and trace is stored into the SoC."
It's basically a means of helping to isolate where a system problem may have occurred. Adding the Smart Trace Buffer support to the AMD PMC driver is little more than one hundred lines of code. The Smart Trace Buffer contents can then be read on the system via DebugFS with a new file named stb_read. Enabling it can be done via the amd_pmc.enable_stb=1 module parameter with it at least for now not being enabled by default from the driver reporting side.
Intel's Trace Hub is somewhat along similar lines but appearing to be much more feature-rich than AMD Smart Trace Buffer with a broader scope in offering many more features around system debug. Intel Trace Hub has also been supported under Linux for a while via the intel_th driver.
The v2 patch was posted this morning on the kernel mailing list. Besides the discussion around this patch work, I haven't been able to find any other substantive public information on AMD Smart Trace Buffer. But there was a reference to Smart Trace Buffer from some Cezanne PSP firmware hits, so it's at least been present in AMD SoCs already. From the patch discussion around AMD STB there was also some excitement that this functionality may also help the AMD engineers in debugging some of the outstanding suspend-to-idle issues plaguing some AMD Linux systems.
| 11
| 1,760,719,352.356082
|
https://www.phoronix.com/news/Linux-SME-No-Default-Use
|
Linux To No Longer Enable AMD SME Usage By Default Due To Problems With Some Hardware
|
Michael Larabel
|
Being sent in as a fix for the Linux 5.15 kernel this morning and to be back-ported to existing stable series is a behavior change that the Linux kernel will no longer use AMD Secure Memory Encryption (SME) by default on supported hardware but rather making it now opt-in due to shortcomings of some platforms.
Since the introduction of AMD SME support to the Linux kernel, Secure Memory Encryption has been activated by default when the SME support (AMD_MEM_ENCRYPT) is built into the kernel. That defaulting of "AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT" allowed for Secure Memory Encryption to be used out-of-the-box without needing to specify any extra kernel parameters or the like. Unfortunately, that's led to boot failures on some platforms particularly around IOMMU along with other headaches to work out as well, like some graphics driver issues with not expecting the memory to be encrypted.
The change to not use AMD SME by default stems from this latest mailing list thread over platform problems (in this case, but not limited only to, Raven Ridge) and boot failures possible from trying SME by default. Unfortunately with SME needing to be enabled at an early stage in the kernel boot process, there isn't the possibility at least for now of having enhanced logic for figuring out in a more robust manner when it's possible to enable/disable SME without user interaction.
The patch making the change sums up the current situation:
So with today's x86/urgent pull request going into Linux 5.15 and then back-ported to prior kernels, AMD memory encryption will not default to enabled. Assuming your kernel is built though with the AMD memory encryption code included, it is possible to enjoy Secure Memory Encryption by setting the "mem_encrypt=on" option to have what was previously the default behavior.
| 28
| 1,760,719,352.768824
|
https://www.phoronix.com/news/AMD-Yellow-Carp-Audio-Driver
|
AMD Publishes New Linux Audio Driver Code For Yellow Carp / ACP6x Audio Co-Processor
|
Michael Larabel
|
It was just with Linux 5.15 last month that AMD added a new audio driver for Van Gogh APUs sporting an ACP 5.x audio co-processor (ACP) while now a new driver has been posted for Yellow Carp / Rembrandt with it introducing new ACP 6.x audio co-processor IP.
AMD engineers have been busy the past few months bringing up "Yellow Carp" under Linux that appears at this point to be the Linux codename for the forthcoming Ryzen 6000 series "Rembrandt" APUs. On top of all the graphics driver work and other Linux kernel driver support patches so far, new audio code came out on Monday morning.
The set of 13 patches get Yellow Carp audio support working on Linux. With YC/Rembrandt there is now AMD Audio Co-Processor 6.x (ACP6x) engine support that needs to be enabled. With some 1,440 lines of new code while some of it is based on audio driver code going back to Renoir there is ACP6x support working with DMIC endpoint support.
Hopefully these patches will get reviewed timely and be in good shape for appearing in the 5.16 kernel to allow time for this audio driver support to work its way out prior to Rembrandt APUs appearing.
| 3
| 1,760,719,354.207065
|
https://www.phoronix.com/news/AMD-SEV-SNP-Autumn-2021
|
AMD SEV-SNP Development Continues Towards The Linux Kernel
|
Michael Larabel
|
Since the launch of AMD EPYC 7003 "Milan" processors earlier this year there has been support for SEV-SNP as the latest evolution of Secure Encrypted Virtualization. The mainline Linux kernel still isn't yet supporting SEV Secure Nested Paging from the upstream kernel, but the out-of-tree patches continue to be available for those interested and development work continues in getting that code ready for mainline as well as ironing out other features.
AMD engineers David Kaplan and Brijesh Singh presented at last week's Linux Security Summit hosted by the Linux Foundation. Their presentation was centered on SEV-SNP for delivering the latest AMD VM security/isolation capabilities and the work they've been doing on bringing up the Linux support.
Going back to the March launch of the EPYC 7003 series processors, AMD has been posting Linux patches to enable SEV-SNP albeit still going through the long process of getting the functionality into the mainline Linux kernel so it can be easily enabled by distribution vendor kernels and elsewhere for making SEV-SNP commonplace with latest-generation EPYC servers. They have been going through several rounds of review to get the code squared away so it meets upstream standards and addresses all review comments.
There isn't yet a public recording of the Linux Security Summit session for those that missed it, but the slide deck is available going over SEV-SNP, the current state, and the future. Beyond the basics covered by the current kernel patches, restricted interrupt injection, lazy page validation, live migration, support backing pages from HugeTLB, and vTPM support are among other items planned to implement in future patches.
Beyond the mailing list patches, AMD engineers continue to use this GitHub repository for staging their latest SEV-SNP support patches.
| 0
| 1,760,719,354.332676
|
https://www.phoronix.com/news/AMD-30x-Energy-Efficiency-2025
|
AMD Aims For 30x Energy Efficiency Improvement For AI Training + HPC By 2025
|
Michael Larabel
|
AMD this morning announced a goal of increasing the energy efficiency of EPYC processors running AI training and high performance workloads by 30x... Within the next four years.
AMD went public today with their goal of a 30x energy efficiency improvement by 2025 for AI training and HPC workloads using EPYC processors with AMD Instinct accelerators.
AMD today announced a goal to deliver a 30x increase in energy efficiency for AMD EPYC CPUs and AMD Instinct accelerators in Artificial Intelligence (AI) training and High Performance Computing (HPC) applications running on accelerated compute nodes by 2025. Accomplishing this ambitious goal will require AMD to increase the energy efficiency of a compute node at a rate that is more than 2.5x faster than the aggregate industry-wide improvement made during the last five years.
Given that we are almost to 2022, AMD must feel very confident in their next few generations of high performance computing products... Exciting times ahead! Learn more about their new energy efficiency goal at AMD.com.
| 31
| 1,760,719,355.793834
|
https://www.phoronix.com/news/AMD-Linux-Audio-Guybrush
|
New AMD Linux Audio Driver Patches Posted - Aimed For Enhancing Chromebook Support
|
Michael Larabel
|
In addition to Linux 5.15 adding a new AMD audio driver for "Van Gogh" APUs such as found in the forthcoming Steam Deck, AMD's open-source Linux driver engineers have also been working on other audio improvements -- this time on the Chromebook front.
A set of eight AMD Linux audio patches were sent out today. These audio patches of relevance to Renoir and Cezanne platforms seems to be driven by the Google Chromebook "Guybrush" board but may prove relevant to other hardware moving forward. That Guybrush Zen 3 Chromebook has been known since the end of last year when it began appearing in Chromium OS sources.
The patches sent out today I2S audio support for the AMD ACP (Audio Co-Processor) with a focus on Renoir, a common machine driver module around the AMD ACP given the increasingly complex audio configurations of AMD hardware, Guybrush specific patches, Maxim MAX98360A codec support, and also RT5682-VS codec support in the context of ACP.
It's great seeing the increasing flow of upstream, open-source Linux driver patches by AMD especially on the consumer side given their continuing successes with Chromebooks, Steam Deck, and more. These latest patches given their timing will see mainline with Linux 5.16 at the earliest.
| 4
| 1,760,719,355.975267
|
https://www.phoronix.com/news/AMD-PSF-Linux-5.16-KVM
|
Linux 5.16 To Expose AMD PSF Disable Bit To KVM Guests
|
Michael Larabel
|
While the Linux kernel still hasn't added any formal control yet for AMD Predictive Store Forwarding to disable it short of also toggling Spectre V4 / SSBD, with the Linux 5.16 kernel the AMD PSF bit will now be exposed to KVM guest virtual machines so that they -- either with a patched/future kernel or for other operating systems -- may choose to toggle explicitly disable this AMD CPU feature.
Predictive Store Forwarding is the Zen 3 feature in the name of performance that AMD provided a security analysis of earlier this year and the (small) possibility it could lead to incorrect CPU speculation. More than a half-year later, there still are no indications of any real-world attack on AMD PSF and the functionality remains enabled by default. But efforts to expose a PSF option for disabling it under Linux if so desired (aside from also engaging SSBD) have stalled.
There was initially some disagreement over the naming of the PSF enable/disable control and other semantics around there. Some upstream developers also question the usefulness of a separate PSF control as if administrators are wanting to disable it chances are they also will make use of SSBD that in turn disables it. Basically it might be unnecessary controls and complexity to the kernel.
So while no PSF control for the Linux kernel is happening upstream at the moment, published recently and queued today into the KVM development code for planned introduction in Linux 5.16 is this patch. That patch is just about exposing the Predictive Store Forwarding bit to KVM (the Kernel-based Virtual Machine) so that the guest kernel can make use of it if desired -- such as if having a patched kernel, a future kernel with such a control, or another operating system able to toggle the PSF disable bit. The few lines of code are ready to go for the next kernel cycle.
| 0
| 1,760,719,357.328381
|
https://www.phoronix.com/news/Linux-amd-pstate-v2
|
Updated AMD P-State Driver Published For Linux
|
Michael Larabel
|
Earlier this month AMD published their "amd-pstate" Linux driver that leverages ACPI CPPC data to make more informed CPU frequency scaling decisions with an aim to boost the performance-per-Watt for Zen 3 (and eventually Zen 2) processors on Linux. The second spin of that "amd-pstate" Linux kernel driver is now available for testing.
AMD did a Sunday morning code drop of their new amd-pstate patch series. The "v2" patch series is largely the same as the original patches from earlier this month but with improved code organization and a few other low-level changes. It doesn't appear that there is any fundamental changes with this patch series to affect the overall performance / power efficiency compared to the prior original patches.
Those interested in testing this new Linux amd-pstate driver can find the series on the kernel mailing list. There is also the amd-pstate-dev-v2 via Git for easier building/testing.
I'll be giving this driver revision a go on some more hardware. My initial amd-pstate vs. acpi-cpufreq testing with a few systems showed that there still is room for improvement compared to ACPI CPUFreq, we'll see what happens with more hardware and if the v2 series does happen to have any impact. It's good in any case to see the amd-pstate driver being revved quickly compared to the original AMD ACPI CPPC driver that stalled out immediately after the v1 patches in 2019.
| 5
| 1,760,719,357.394907
|
https://www.phoronix.com/news/AMD-PSF-Control-September
|
Still-Pending AMD PSF Control Patch To Be Retailored For KVM
|
Michael Larabel
|
Of all the great stuff for AMD in Linux 5.15, one of the patches still not having yet been mainlined is the control support around Predictive Store Forwarding (PSF) with Zen 3 processors. It's been six months since AMD published their security whitepaper around PSF while the Linux patch has yet to be mainlined while now it seems will be updated for a reduced focus on KVM usage.
It's been about six months already since AMD published their security analysis of Zen 3's PSF feature that could potentially lead to a side channel attack. Predictive Store Forwarding with the latest-generation Ryzen and EPYC processors allows for speculatively executing instructions based on what it thinks the result of the load will be and while the predictions should be largely accurate, there is the small possibility of incorrect CPU speculation. PSF speculation going awry would be similar to Spectre V4 / SSB.
The AMD analysis from March did note, "AMD believes that for most applications, the security risk of PSF is likely low and where isolation is required, techniques such as address space isolation are preferred over software sandboxing."
AMD did send out Linux kernel patch to optionally allow PSF to be disabled by itself or it's also disabled as part of the Spectre V4 mitigations. For over five months that kernel work has been public, it has undergone code review for possible mainline integration. The semantics of the PSF command-line control naming has resulted in some changes as well as ongoing discussion whether there should be an "auto" mode rather than just the on/off modes and defaulting to PSF enabled.
There doesn't appear to be much of an emphasis by AMD on getting the PSF control support merged quickly given the time it's been since the PSF whitepaper and the patch discussion only reigniting every few weeks or longer. Fortunately, the possibility of PSF being exploited in the real-world appears to remain very low and thus seemingly not a priority for getting the changed merged quickly and with no default change in behavior.
Most recently it's been suggested by upstream just doing away with their current PSF patch and provide updated guidance in the documentation. Since SSBD for Spectre V4 mitigation also disables PSF, interested parties would likely just go that route rather than adding extra kernel controls and complexity.
The latest messaging as of Friday is now that AMD will likely "revisit this later when it seems reasonable to add this in the kernel." A new PSF patch though is expected for at least controlling the PSF feature being exposed to KVM guests and that code is expected soon.
All indications are the the real-world threat around PSF is very low and haven't heard of any increased concern over the past half-year since the original security whitepaper was published. However, more broadly it's not very reassuring to see this kernel patch tossed around for a half-year and due to the lack of a concerted effort potentially just dropped at this stage. Intel has had more vulnerabilities to deal with and often of greater significance, but at least there they have been well organized to ensure that generally on the same day of any new public disclosures they already have kernel patches going into the mainline kernel right away with any new controls/mitigations. This is an ongoing area for AMD to improve upon with their Linux support and hopefully there will be that greater coordination by AMD engineers and the upstream kernel developers so any future issues can be patched more timely.
| 0
| 1,760,719,358.903641
|
https://www.phoronix.com/news/AMD-P-State-XDC2021
|
AMD + Valve Focusing On P-State / CPPC Driver With Schedutil For Better Linux Efficiency
|
Michael Larabel
|
As reported at the start of August, AMD and Valve have been working on Linux CPU performance/frequency scaling improvements with the Steam Deck being one of the leading motivators. As speculated at that time, their work would likely revolve around use of ACPI CPPC found with Zen 2 CPUs and newer. Published last week was that AMD P-State driver for Linux systems indeed now leveraging CPPC information. AMD formally presented this new driver yesterday at XDC2021.
This effort largely comes down to what was anticipated that AMD is working on this new "AMD P-State" driver for leveraging ACPI Collaborative Processor Performance Controls (CPPC) for making more informed CPU frequency scaling / performance state decisions. CPPC is supported by AMD CPUs with Zen 2 and newer though initially this driver is limited to a subset of Zen 3 processors until properly vetted. Additionally, AMD (and Valve) are focused on making use of the Schedutil governor that leverages the kernel's scheduler utilization data for trying to make more accurate decisions too. Schedutil is already the default with ACPI CPUFreq on many Linux distribution kernels when not using Intel's P-State driver and is generally in great shape. Upstream has also been pushing for AMD to embrace Schedutil going back to when they were originally dabbling with ACPI CPPC support in 2019 for Zen 2.
AMD's Ray Huang presented on Friday at the X.Org Developers' Conference (XDC2021) around their work. Below is the presentation for those interested, but long story short as is well known ACPI CPUFreq is less than ideal and AMD P-State is being worked on to improve the situation for newer AMD CPUs, similar to Intel with their P-State driver.
AMD's figures are showing the new P-State driver delivering measurable performance/efficiency improvements over CPUFreq for different workloads. My own testing has been somewhat mixed with the hardware I've been testing... In some cases, yes, the perf-per-Watt is better but the overall raw performance is measurably lower in some cases. I'll have up my article next week.
The AMD P-State driver remains under development and not yet mainlined, so won't be coming until at least Linux 5.16 at the earliest.
There is also this slide deck to go along with the presentation.
| 28
| 1,760,719,359.184149
|
https://www.phoronix.com/news/Linux-5.15-AMD
|
Linux 5.15 Is A Very Exciting Kernel For AMD
|
Michael Larabel
|
While working on my usual Linux kernel feature overview that summarizes the many articles over the past two weeks outlining all of the new features and changes merged, one area that particularly stands out for Linux 5.15 are all of AMD's upstream contributions that happened to make it in this kernel. There is a lot of new enablement on the AMD side -- both for CPUs and Radeon graphics -- but also improving existing hardware support.
Linux 5.15 is particularly exciting for AMD users, just not for CPUs and Radeon GPUs but also on the Ryzen client side and EPYC server front too. Linux 5.15 has a lot to offer with AMD improvements, which is great to see and comes with ramping up their Linux hires.
All of the notable AMD changes I am aware of that are new to Linux 5.15 include:
- The AMD Van Gogh APU audio driver for the AMD ACP 5.x audio co-processor was merged. This is good news for the Steam Deck and other Van Gogh designs.
- An important change/fix in the AMD s2idle code concerning suspend/resume behavior. Thanks to being able to coordinate internally at AMD, it was discovered the existing s2idle behavior didn't match that of Windows and now with this fix select HP ENVY laptops and other laptops should hopefully see more reliable suspend/resume behavior.
- AMD Zen 3 APU temperature monitoring support is finally in place... Unfortunately it took until months after launch and the basic patch was left up to the community this time. In any case, with Linux 5.15+ the new Zen 3 APUs can enjoy working temperature monitoring under Linux.
- Yellow Carp APU temperature monitoring support. Making this rather exciting is that the Yellow Carp APUs have yet to launch! This is presumably the Ryzen 6000 series mobile SoCs. With being used to seeing AMD late in submitting new k10temp additions after-launch, it's great finally having AMD submit the basic temperature monitoring support ahead of launch...
- Optimized C3 entry handling for when AMD CPUs are entering the C3 power sleep state. The optimization is that the kernel doesn't need to flush the caches before entering C3 due to the possible performance impact for other cores that are sharing caches. Intel CPUs have long enjoyed this behavior under Linux and even Zhaoxin and Centaur CPUs have had this in place for Linux while the AMD optimization is only now in place.
On the AMD EPYC/server side specifically there is:
- The AMD PTDMA driver has finally landed with the patches for that undergoing public review since 2019. That Pass-Through DMA support is for high bandwidth memory-to-memory and I/O copies for NTB devices.
- The AMD SB-RMI driver is also new. This is for the sideband remote management interface for AMD EPYC server platforms. This driver living within the HWMON area provides for reading the CPU socket power consumption, power limit, and maximum power limit plus setting/limiting the maximum power limit of a CPU socket. This SB-RMI interface coordinates with the BMC on server platforms with this seemingly another addition with AMD's growing OpenBMC support focus.
- The AMD KVM code now supports 5-level paging for handling greater memory capacities. Thus it's looking like AMD EPYC "Genoa" processors will be supporting 5-level paging.
On the AMD Radeon graphics side:
- A number of new RDNA2 PCI IDs were added pointing to a possible RDNA2 graphics card refresh or other rumors around Radeon RX 7000 series being a mix of existing and new GPUs.
- AMD Cyan Skillfish support for a Navi 1x APU design yet to be launched.
- AMDGPU is now leveraging VKMS for improving its virtual display code.
- There are also fixes to AMDGPU power management, HDMI FreeSync, PCI Express link handling, and many other fixes.
Basically, a lot is going on in the AMD Linux space at the moment. Linux 5.15-rc1 should be out this evening while the stable kernel release will be out about two months from now.For as good as all the changes in Linux 5.15 are, they will need to keep up and continue working early on new hardware support particularly from the CPU side if they wish to match Intel's stellar open-source upstream reputation around early hardware support and new features. So we'll see how their Linux improvements pan out over the longer term in the kernel cycles ahead.
| 40
| 1,760,719,360.385029
|
https://www.phoronix.com/news/AMD-PSTATE-2021
|
AMD Posts New "AMD-PSTATE" CPUFreq Driver Leveraging CPPC For Better Perf-Per-Watt
|
Michael Larabel
|
At last! AMD has posted the Linux kernel driver patches for their new "AMD-PSTATE" driver! This driver with modern AMD Zen CPUs (initially limited to Zen 3) to achieve greater performance per Watt / power efficiency on Linux than the conventional ACPI CPUFreq driver.
The new AMD-PSTATE driver is akin to Intel's P-State driver long used by Intel CPUs as better catering to their hardware than the generic ACPI CPUFreq driver. AMD-PSTATE leverages ACPI Collaborative Processor Performance Controls (CPPC) for making more informed performance state decisions.
ACPI CPPC has been supported since Zen 2 processors but the initial AMD-PSTATE driver is limited to just Zen 3 processors. AMD says they will extend their coverage with time, which would mean going back to supporting Zen 2 processors too.
It was back in July 2019 that AMD originally posted "amd_cpufreq" as a CPPC-based driver right as they were launching the Zen 2 processors. However, that patch set was abandoned and never made it to mainline. Over the past two years I routinely asked AMD about the CPPC Linux support to which they commented on the lack of resources, but great to see this new AMD CPUFreq driver finally materializing.
It's not entirely unexpected though. Last month I wrote about AMD and Valve working on a new CPU performance scaling design. In that prior article basically laid out that it would likely be the long-awaited CPPC-based approach and now this new driver patch series is delivering on just that.
Further pointing to the Valve / Steam Deck connection is that the initial AMD P-State patches are tested on AMD Cezanne APUs. AMD posted a few tests with their patch series showing nice gains out of this AMD-PSTATE driver compared to ACPI CPUFreq.
The code was posted today across 19 patches. Needless to say, I am currently building a new kernel with these patches and will be delivering a number of AMD Ryzen 5000 series and EPYC 7003 series benchmarks over the coming days looking at the performance and power efficiency of this new driver. Benchmarks across mobile, desktop, and servers with Zen 3 starting now. It's a long time with more than two years since the original AMD CPPC patches were posted, but great to see it come about and this time around hopefully has sufficient momentum to get worked on punctually and mainlined for benefiting the Steam Deck and the growing number of AMD Linux users at large. Stay tuned!
| 68
| 1,760,719,360.712658
|
https://www.phoronix.com/news/AMD-BRS-Branch-Sampling
|
AMD Branch Sampling Support Being Worked On For Linux
|
Michael Larabel
|
A new kernel patch series was posted this morning working on AMD Branch Sampling "BRS" support.
AMD Family 19h "Zen 3" processors support this branch sampling feature for collecting details on branches taken during code execution. The latest AMD Zen 3 processors do contain improvements to BRS functionality. AMD BRS is detailed within their processor programming reference guides that are public but until now with this new patch series sent out by a Google engineer there hasn't been any Linux support/integration.
This AMD BRS bring-up for Linux is integrating the functionality into Linux's wonderful perf subsystem. But long story short why this AMD Branch Sampling support is news worthy is that the functionality can be used for AutoFDO-style optimizations on AMD processors with the compiler leveraging that collected hardware data to make more informed and accurate optimization decisions. With Google engineers working on this BRS bring-up, that appears to be their motivation is to in turn leverage it for compiler optimizations.
AMD's exposed feature set for "perf" on Linux has traditionally been lacking compared to Intel and the exposed performance counters, but it's great that in recent times there has been catching up. Once this AMD BRS support is squared away, it will be fun to see what compiler optimization work will come in leveraging this sampling data.
The set of 13 patches getting Linux's perf code wired up for AMD Zen BRS support can be found on the kernel mailing list.
| 3
| 1,760,719,361.571408
|
https://www.phoronix.com/news/AMD-PTDMA-For-Linux-5.15
|
AMD PTDMA Driver Landing For Linux 5.15 After Two Years In The Works
|
Michael Larabel
|
Going back to September 2019 was work on the AMD PTDMA driver for supporting this controller found on modern AMD processors for high bandwidth memory-to-memory and I/O copy operations. With the Linux 5.15 cycle the AMD PTDMA driver is finally being merged to the mainline kernel.
AMD EPYC processors feature multiple PTDMA device instances. While the PTDMA controllers have been found in AMD processors already, the upstreaming process has taken two years and gone through more than ten rounds of code review to address outstanding items and other suggestions for improvements.
The AMD PassThru DMA driver was last revised in August and after v11 was published was deemed ready to go. This driver interfaces with the Linux kernel's DMA subsystem.
The AMD PTDMA for high bandwidth memory-to-memory and I/O copies aren't good for general purpose peripheral DMA but cater to AMD Non-Transparent Bridge (NTB) device usage.
This driver, simply called "ptdma", comes in at a little more than one thousand lines of code. It was sent in yesterday as part of the dmaengine updates for Linux 5.15.
| 4
| 1,760,719,362.608007
|
https://www.phoronix.com/news/AMD-S2idle-Fix-Linux-5.15
|
AMD Has An Important Suspend/Resume Fix With Linux 5.15
|
Michael Larabel
|
Since last year AMD has been working to get its s2idle / suspend-to-idle S0ix sleep state code in order for supporting this lowest power platform idle state on newer AMD laptops and there has also been other AMD suspend/resume improvements in recent times. Now with the Linux 5.15 kernel cycle is an important fix for the AMD s2idle code.
Stemming from a user reporting incorrect resume from suspend with an HP ENVY X360, Linux 5.15 has a fix for it and other laptops facing a similar problem. The original issue was resuming from suspend would yield the power LED not working, some keys like brightness controls not working, and the cooling fan not spinning up even under load. Yeah, that's quite a poor Linux laptop experience.
Sent in to Linux 5.15 today as part of more ACPI updates is changing around the suspend-to-idle support code to meet the platform firmware expectations on some AMD systems.
While debugging the problem of the HP ENVY X360 laptop going awry after resume, AMD checked with their internal team and discovered that under Windows both the AMD and Microsoft methods need to be called. Previously just the Microsoft method was called when present on Linux systems. Many OEM systems though don't implement anything for the AMD methods but some do as in the case of this HP laptop and so the s2idle code now will be calling both the Microsoft and AMD methods when present. This at least gets the HP ENVY X360 to behave now on resume and we'll see soon enough what other hardware similarly benefits from this change.
This AMD s2idle patch for Linux 5.15 was written by Mario Limonciello, a recent and prominent hire to AMD's Linux client team.
| 24
| 1,760,719,363.160542
|
https://www.phoronix.com/news/Dell-Mario-On-AMD-Linux-Team
|
A Prominent, Longtime Dell Linux Engineer Recently Joined AMD's Linux Team
|
Michael Larabel
|
Here should hopefully be a great indication about AMD's Linux efforts moving forward with one of their recent and exciting hires at the company.
As noted across various Phoronix articles over the past year, AMD has been ramping up their Linux staff given their market successes on both client and server. This has been much desired considering nearly a decade ago they let go many of their Linux developers and closed their Operating System Research Center. AMD's Linux support for recent CPU (and GPU) launches has been good, but Intel continues to serve as the "gold standard" of Linux support when it comes to generally providing punctual pre-launch support for new hardware, generally ensuring their new hardware features are supported under Linux, making sure their forthcoming microarchitectures are well supported by the prominent open-source compilers ahead of time, etc. Intel's large open-source engineering pool has allowed this generally very good pre-launch hardware support. In addition, that large talent pool has led Intel to contributing significantly to various non-Intel-specific improvements to the Linux kernel and other areas.
While the AMD processor support is generally in good shape at launch on Linux, their Linux engineers have been under-staffed since the unfortunate OSRC closure in 2012. This has led to non-critical items like temperature and power monitoring often not coming to Linux until post-launch by the community even when it's often just been new IDs needing to be added, Google and other companies contributing hardware features normally carried out by the hardware vendor, compiler support for new microarchitectures not being published until around launch-time and thus longer until in a stable compiler, various quirks, and other areas that could be improved upon.
Thus I was rather excited to notice on Friday (albeit belated) that Mario Limonciello is among their recent Linux hires at the company. Mario is a longtime Linux engineer at Dell since ~2008 that was involved with "Project Sputnik" and their other Linux-enablement work at Dell. In particular, heavily involved with Dell's efforts around Ubuntu preloads across a range of consumer laptops/desktops over the years. He's also been involved with FWUPD firmware updating on Linux and other efforts and submitting a number of upstream Linux kernel patches.
Prior to joining Dell, he was also an Ubuntu contributor (and Mythbuntu developer, pictured above - left, back in 2007 at the former Ubuntu Live conference) and also involved with the ATI fglrx install scripts back in the day and more.
With his background and significant Linux client work at Dell, it's great to see he is now at AMD working on their Linux efforts.
It turns out he joined AMD earlier this year. His apparent joining of the company came weeks after the jobs mentioned in this prior article about AMD is hiring more Linux engineers. What is significant about that is that those job postings were for Linux on the client side. Those postings also noted "a new organization" being built at AMD around Linux on the client side.
Further pointing to the AMD Linux client work is how I noticed he was at AMD... Mario's latest kernel patches. Now at AMD, his latest patch series is bringing up Yellow Carp for the k10temp temperature monitoring driver! Pre-launch temperature monitoring support for a client APU! Yellow Carp is possibly the next-gen Rembrandt APUs. Seeing the CPU temperature monitoring driver support pre-launch for Yellow Carp / Rembrandt while it may seem silly to Windows users that take it for granted, seeing this happen on the client-side is significant. Heck as mentioned in the Ryzen 7 5700G Linux testing just this week, with that already-shipping desktop APU that k10temp support isn't even coming until the next kernel cycle (5.15).
Since joining AMD in April, Mario has also been working on some fixes for Renoir/Cezanne and other bits as well. Hopefully this is just the start of more AMD Linux client improvements. AMD has also been hiring to work on the Linux scheduler, memory management, and other areas with some of those positions not yet filled.
| 31
| 1,760,719,364.226456
|
https://www.phoronix.com/news/AMD-C3-Optimize-Linux-5.15
|
Optimized C3 Entry Handling For AMD CPUs Queued For Linux 5.15
|
Michael Larabel
|
The work I initially wrote about last week for AMD optimizing their C3 entry handling to avoid an unnecessary cache flush will now be picked up for the upcoming Linux 5.15 kernel cycle.
As explained in the prior article, this optimization around entry to the ACPI C3 sleep state is already applied for Intel, Zhaoxin, and Centaur processors on Linux. Unfortunately, it's just AMD CPUs are late to the party with the kernel code not having marked them as safe for this optimization until an AMD engineer submitted a patch last week.
When a CPU core enters the C3 power sleep state, the default behavior to ensure cache coherence is to flush the cache even if that cache is shared with other CPU cores not currently sleeping. Thus AMD's Deepak Sharma submitted a patch last week to allow the optimization unconditionally for all AMD CPUs, "AMD CPU which support C3 shares cache. Its not necessary to flush the caches in software before entering C3. This will cause performance drop for the cores which share some caches. ARB_DIS is not used with current AMD C state implementation. So set related flags correctly."
That patch has now been picked up as part of the ACPI material being collected for the Linux 5.15 merge window. That next cycle is expected to kick off next week pending the release of Linux 5.14 this weekend and in turn will debut as stable this autumn. This basic optimization is overdue, but at least the few lines of code patch is now on the way to being merged.
| 1
| 1,760,719,364.638543
|
https://www.phoronix.com/news/AMD-Mem-Device-Public
|
AMD Posts Latest Linux Patches For Supporting The Frontier Supercomputer
|
Michael Larabel
|
AMD engineers continue working on preparing the Linux kernel for the Frontier supercomputer.
Much of the Frontier bring-up for the Linux kernel over the past number of months has been around supporting the coherent interconnect between AMD EPYC CPUs and the Instinct "Aldebaran" GPUs/accelerators with allowing CPUs coherent access to the GPU memory. The latest patch series out today for the Linux kernel is again focused on this GPU device memory handling.
The new patches introduce the notion of "MEMORY_DEVICE_PUBLIC" as memory that is mapped for CPU access but within the GPU device memory and supporting migration to/from those areas. Back in May we wrote about the earlier work on that coherent GPU device memory support with migration support.
The 14 "v1" patches introduce the MEMORY_DEVICE_PUBLIC type and plumbed into the kernel's memory management code and the necessary changes to the AMDKFD kernel driver. This MEMORY_DEVICE_PUBLIC code works with their recently-merged heterogeneous memory management (HMM) support and Shared Virtual Memory (SVM) capability recently added to AMDKFD.
Long story short, the Linux kernel bring-up for Frontier remains ongoing. The Frontier ORNL site is still showing a 2021 delivery for this first US exascale supercomputer.
| 1
| 1,760,719,365.952391
|
https://www.phoronix.com/news/AMD-SEV-SNP-v5
|
AMD Publishes Latest SEV-SNP Guest + Hypervisor Support For Linux
|
Michael Larabel
|
AMD has published their fifth revision of SEV-SNP support for the KVM hypervisor and guest VM support for this Secure Encrypted Virtualization Secure Nested Paging functionality found with new EPYC 7003 series server processors.
SEV-SNP is the latest iteration of Secure Encrypted Virtualization. SEV-SNP provides additional memory integrity protections around replay protection, data corruption, memory aliasing, and memory re-mapping. There are also other hardware protections with SEV-SNP as outlined in the SEV comparison table:
While SEV-SNP support is found in EPYC 7003 series processors since their March debut, the Linux kernel bring-up for this latest functionality is still a work-in-progress. Shortly after the EPYC "Milan" announcement in March, the Linux kernel patches began and have now worked from their initial request-for-comments form to now a five rounds of code review for preparing the KVM hypervisor code and guest support for the latest SEV functionality.
On the hypervisor side are 45 patches that provide the basic building blocks for handling of SEV-SNP virtual machines. There still is though more security enhancements with SEV-SNP that aren't supported by this code, such as interrupt protection. Further feature work on SEV-SNP is expected after this initial base support lands.
With this v5 series there are many updates stemming from the prior rounds of code review, the ability to enforce a minimum SEV-SNP firmware version, and other low-level code improvements.
There are also the 38 patches providing the SEV-SNP guest support. That gets the initial support in place but like on the hypervisor side there still will be future feature work. As one example of future work, the guest support for now is also only doing pre-validation of memory pages rather than the on-demand/lazy validation -- Intel Linux engineers are currently working on similar code in this area around Linux "unaccepted memory" handling for that lazy/on-demand validation, which for their case is for TDX.
Like on the hypervisor side, the v5 guest patches are primarily for addressing comments raised during the prior code review.
It's getting very close to the Linux 5.15 merge window kicking off and not immediately clear if this code will prove to be ready for the next cycle whether these v5 patches have all issues addressed, but chances are given the timing the code will likely not be until at least Linux 5.16. Thus stretching it out from potentially appearing in autumn 2021 Linux distributions. We certainly hope the code will reach mainline in the next cycle or two so that this AMD SEV-SNP support can appear out-of-the-box in the default kernel of Ubuntu 22.04 LTS in the spring.
For out-of-tree kernel use, besides the patches on the kernel mailing list itself, AMD does continue to maintain the sev-snp-devel branch of AMDESE/AMDSEV on GitHub that some organizations are utilizing for SEV-SNP protections right now with EPYC Linux servers. Moving forward, hopefully AMD will manage to deliver more punctual upstream, mainlined kernel support for such features -- especially with AMD hiring more Linux engineers, including in the area of virtualization.
| 0
| 1,760,719,366.222753
|
https://www.phoronix.com/news/AMD-Linux-Skip-C3-Cache-Flush
|
AMD To Optimize C3 Entry On Linux By Finally Skipping The Cache Flush
|
Michael Larabel
|
A minor optimization was posted by an AMD engineer on Wednesday for the Linux kernel.
The optimization posted is around the ACPI C3 power state handling on Linux for AMD processors. Right now when a CPU core enters the C3 power sleep state, its cache is flushed even though the cache may be shared with CPU cores that are not in the sleep state. As the AMD patch notes, "this will cause performance drop for the cores which share some caches."
Thus the new patch to optimize the C3 entry on AMD CPUs is now having that cache flushing skipped as it's not necessary for their processors and can avoid potentially hurting the performance for other cores with shared cache.
That flushing on C3 sleep is the default behavior with the kernel for cache coherence purposes. Up to now within the Linux kernel Intel CPUs unconditionally already avoided that cache flush on C3 entry. All Zhaoxin CPUs had this basic optimization too on Linux as well as newer Centaur CPUs (depending upon microcode version) were skipping their cache flushing on C3 entry, but now AMD has joined the party too.
| 6
| 1,760,719,367.410254
|
https://www.phoronix.com/news/Radeon-Software-Linux-21.30
|
Radeon Software for Linux 21.30 Brings Support For Ubuntu 20.04.3 LTS, RX 6600 XT
|
Michael Larabel
|
With the Radeon RX 6600 XT review embargo lifted and the cards set to hit retail channels this week, the Radeon Software for Linux 21.30 driver has been released in now officially supporting the RX 6600 series.
Today's packaged driver release succeeds Radeon Software for Linux 21.20 from back in June. The Radeon Software for Linux 21.30 driver supports the current RX 6600/6700/6800/6900 series line-up while continuing to go back to supporting prior generations through the Radeon Rx 289/290/290X Hawaii GPUs (GCN 1.2). While AMD dropped pre-Polaris support from their mainline Windows driver, that isn't the case here with the upstream AMDGPU Linux driver continuing to support GCN 1.2 GPUs -- and even GCN 1.0/1.1 too unofficially when switching over from the Radeon DRM driver via module options.
Besides having RX 6600 XT support, the Radeon Software for Linux 21.30 driver only lists Ubuntu 20.04.3 LTS support as the other change for this quarterly update. Ubuntu 20.04.3 LTS is due out later this month and brings the hardware enablement (HWE) stack from Ubuntu 21.04 to this long-term support distribution.
In addition to supporting Ubuntu 20.04.3 LTS, this packaged Linux driver also has support for Ubuntu 18.04.5, RHEL/CentOS 7.9, RHEL/CentOS 8.4, and SUSE SLED/SLES 15 SP3.
This updated Radeon Software for Linux 21.30 driver can be downloaded from AMD.com. I'll have up some fresh RDNA2 upstream vs. packaged driver benchmarks this week.
In case you missed it, be sure to check out today's Radeon RX 6600 XT Linux review for benchmarks on this new mid-range graphics card.
| 4
| 1,760,719,368.168031
|
https://www.phoronix.com/news/AMD-5-Level-Paging-KVM
|
AMD Preparing 5-Level Paging Linux Support For Future CPUs
|
Michael Larabel
|
Future AMD CPUs -- potentially AMD EPYC 7004 "Genoa" -- will be supporting 5-level paging.
5-level paging is for increasing the virtual and physical address space on x86_64 systems. With 5-level paging the virtual address space goes from a 256 TiB maximum to 128 PiB while the physical address space threshold goes from 64 TiB to 4 PiB. This 5-level paging support is important for today's increasing powerful and memory intensive servers. The downside to 5-level paging is page table walks taking longer due to that extra level, but in practice that cost should be small especially with software optimizations made since Intel originally began working on 5LP support years ago.
Intel worked on the 5-level paging Linux kernel support going back five years now while on the hardware side it's just with their latest Intel Xeon Scalable 3rd Gen "Ice Lake" processors where they support 5-level paging. That support has been around since the days of Linux 4.14 while in Linux 5.5 5-level paging support is enabled by default on supported CPUs and all ironed out on more recent kernels.
While Intel's Xeon Ice Lake processors support 5-level paging, there hasn't been much (any?) public communication from AMD around their 5-level paging plans for their EPYC server processors. But out now is a new patch series in preparing the AMD SVM KVM code for 5-level page tables. Kernel changes to this AMD SVM specific code within the Kernel-based Virtual Machine is needed for handling 5-level page tables.
One of the patches does note, "AMD future CPUs will require a 5-level NPT if host CR4.LA57 is set." Setting the CR4 register bit LA57 is the same as Intel for enabling 5-level paging with supported processors. Thus the AMD 5-level paging for future (EPYC) processors is likely to be quite similar to Intel and leverage all of Intel's existing Linux kernel support. This is also confirmed by the AMD 5-level paging Linux kernel patches all being specific to just the AMD SVM code -- in this case there are some KVM changes needed to the AMD code for proper 5-level page table handling for guest VMs.
Given the timing of these patches and AMD's current practices of not usually sending out hardware enablement related CPU patches too far in advance for the Linux kernel or related open-source components (unlike Intel's established open-source track record over the past two decades), it's likely a safe bet to assume next-gen EPYC "Genoa" processors next year will be supporting 5-level paging.
| 10
| 1,760,719,369.222108
|
https://www.phoronix.com/news/Linux-AMD-sev_secret
|
Linux "sev_secret" Patches For Tapping Confidential Computing Secret Areas In AMD SEV
|
Michael Larabel
|
The latest AMD SEV work happening to the Linux kernel for benefiting EPYC servers with virtualization is the new "sev_secret" module for allowing guests to access confidential computing secret areas.
AMD Secure Encrypted Virtualization does allow guest VM owners to inject "secrets" into the virtual machines without the host or hypervisor being able to read those secrets. At present though the Linux kernel doesn't allow accessing of these secrets from within guest virtual machines.
Thanks to IBM engineers, support for accessing the confidential computing secret areas within AMD SEV guests is coming in the form of the new "sev_secret" kernel module. The sev_secret module handles copying of the secrets fron the EFI memory to kernel-reserved memory and then allows exposing those secrets within the VM via SecurityFS.
One of the example use-cases for this secret injection usage to VMs is for having guest VMs perform operations on encrypted files and the decryption key being passed to the VM using this mechanism. In doing so, the host/hypervisor doesn't have access to said key and with SEV the guest's memory is also encrypted.
This patch series sent out today by IBM engineer Dov Murik succeeds a former "request for comments" on the matter.
| 1
| 1,760,719,369.671583
|
https://www.phoronix.com/news/AMD-PTDMA-Linux-v11
|
AMD PTDMA Driver Revised Ahead Of Its Possible Inclusion For Linux 5.15
|
Michael Larabel
|
One of the AMD patch series that has been in the works for more than one year is the PTDMA driver providing pass-through DMA engine support on Linux. The driver is now up to its eleventh revision but the mainlining might happen soon.
The AMD PTDMA Linux driver effort dates back to September 2019 for enabling their PTDMA controller in performing high bandwidth memory-to-memory and I/O copy operations. Modern AMD CPUs support multiple PTDMA controllers, the PTDMA driver hooks into the kernel's direct memory access (DMA) subsystem and is intended to be used with AMD Non-Transparent Bridge (NTB) devices but not for general purpose peripheral DMA.
The prior AMD PTDMA v10 patches were sent back around the time of the Linux 5.14 merge window while the DMA subsystem maintainer Vinod Koul had commented the patches were still in his review queue when prompted by the AMD engineer involved over the status of getting the patches picked up.
This week the v11 patches were sent out with various code improvements. Given there is still a few weeks until the Linux 5.15 merge window, we'll see now if the work is in a state ready for mainlining.
| 1
| 1,760,719,370.739022
|
https://www.phoronix.com/news/AMD-SB-RMI-Linux-Sensor
|
AMD SB-RMI Driver Coming For Linux 5.15
|
Michael Larabel
|
AMD continues pushing new code out for Linux in better exposing their platform's capabilities in the open-source world. The latest AMD driver work now queued via "-next" branches for introduction this autumn in Linux 5.15 is SB-RMI sensor support.
AMD SB-RMI is the Side-Band Remote Management Interface for out-of-band communication between the AMD SoC/CPU and the baseboard management controller (BMC) via the Advanced Platform Management Link (APML / SBI). Queued via the hwmon-next Git branch is an initial AMD SB-RMI sensor driver for Linux.
This SB-RMI driver (simply named SBRMI) allows for reading the CPU socket power consumption, power limit, and maximum power limit. Additionally, the driver allows setting/limiting the maximum power limit of a CPU socket. Though given SB-RMI, it's important to reiterate this is just for server platforms with a supported BMC configuration and just not any AMD platform / desktop. Capable dual socket servers are supported by the driver.
The SB-RMI driver exposes the current socket power, the read/write current socket cap, and maximum cap via the hwmon sysfs interface. The code is ready to go once the Linux 5.15 merge window opens up in the coming weeks.
It's nice seeing progress being made by AMD engineers on exposing these additional capabilities under Linux. For EPYC processors there is currently RAPL support while Zen voltage/current support was previously dropped from Linux with the mainline k10temp driver over lack of public AMD documentation. There was also the AMD energy monitoring driver that was removed from the Linux kernel earlier this year over a disagreement on handling of the CPU metrics being restricted to root-only for security purposes. Hopefully there is progress made on those fronts particularly for desktop/mobile users moving forward.
This SB-RMI work does appear to be motivated at least in part over AMD's growing OpenBMC support just like the SB-TSI sensor support for Linux that landed a few kernels ago.
| 12
| 1,760,719,371.350239
|
https://www.phoronix.com/news/AMD-Valve-New-CPU-Freq
|
AMD + Valve Working On New Linux CPU Performance Scaling Design
|
Michael Larabel
|
Along with other optimizations to benefit the Steam Deck, AMD and Valve have been jointly working on CPU frequency/power scaling improvements to enhance the Steam Play gaming experience on modern AMD platforms running Linux.
It's no secret that the ACPI CPUFreq driver code has at times been less than ideal on recent AMD processors with delivering less than expected performance/behavior with being slow to ramp up to a higher performance state or otherwise coming up short of disabling the power management functionality outright. AMD hasn't traditionally worked on the Linux CPU frequency scaling code as much as Intel does to their P-State scaling driver and other areas of power management at large.
AMD is ramping up efforts in these areas including around the Linux scheduler given their recent hiring spree while it now looks like thanks to the Steam Deck there is renewed interest in better optimizing the CPU frequency scaling under Linux.
AMD and Valve have been working to improve the performance/power efficiency for modern AMD platforms running on Steam Play (Proton / Wine) and have spearheaded "[The ACPI CPUFreq driver] was not very performance/power efficiency for modern AMD platforms...a new CPU performance scaling design for AMD platform which has better performance per watt scaling on such as 3D game like Horizon Zero Dawn with VKD3D-Proton on Steam."
AMD will be presenting more about this effort next month at XDC. It's quite possible this new effort is focused on ACPI CPPC support with the previously proposed AMD_CPUFreq. Back when Zen 2 launched in 2019, AMD did post patches for their new CPUFreq driver that leveraged ACPI Collaborative Processor Performance Controls but the driver was never mainlined nor any further iterations of the patches posted. When inquiring about that work a few times since then, AMD has always said it's been basically due to resource constraints that it wasn't a focus at that time. Upstream kernel developers also voiced their preference to seeing AMD work to improve the generic ACPI CPPC CPUFreq driver code rather than having another vendor-specific solution. It's also possible AMD has been working on better improvements around the now-default Schedutil governor for scheduler utilization data in making CPU frequency scaling decisions.
In any case, we are excited to see the AMD+Valve power/performance scaling improvements come to Linux. The Steam Deck leverages a Zen 2 based custom APU. If it's indeed leveraging ACPI CPPC, these Linux AMD platform improvements should benefit all Zen 2 hardware and newer (Zen 1 and prior lacking CPPC).
Hopefully AMD and Valve continue cooperating about further Linux optimizations both for on the CPU and graphics side. Already on the graphics side in recent years Valve has been employing multiple developers to improve Mesa with a particular focus on the RADV Vulkan driver and their work on the ACO compiler back-end and much more for significantly boosting the open-source Radeon advantage for Linux gamers.
| 26
| 1,760,719,372.59714
|
https://www.phoronix.com/news/AMDVLK-2021.Q3.2
|
AMDVLK 2021.Q3.2 Released With New Extensions, Implicit External Sync For All GPUs
|
Michael Larabel
|
AMD engineers today published their latest code drop of the AMDVLK official open-source Radeon Vulkan driver for Linux systems.
AMDVLK 2021.Q3.2 is this new release and is headlined by the Vulkan extensions of VK_EXT_extended_dynamic_state2 and VK_KHR_copy_commands2 now being supported. VK_EXT_extended_dynamic_state2 was introduced with Vulkan 1.2.176 earlier this year and is an update to the extension introduced last year around providing additional dynamic state for reducing the amount of pipeline state objects being compiled and binded. The VK_EXT_extended_dynamic_state2 has various clarifications and improvements over the original extended dynamic state handling extension. VK_KHR_copy_commands2 meanwhile provides extensible versions of the Vulkan buffer and image copy commands. This extension allows for using extensible structures for extension-specific information not allowed in the original copy commands extension.
AMDVLK 2021.Q3.2 also enables implicit external synchronization for all GPUs and has crash and corruption fixes for some DXVK-based games.
Source downloads and RHEL/CentOS and Ubuntu binaries of AMDVLK 2021.Q3.2 can be found via GitHub.
| 10
| 1,760,719,372.820699
|
https://www.phoronix.com/news/AMD-Barcelo-Coreboot-Starts
|
Coreboot Starts Seeing Bits For AMD Barcelo
|
Michael Larabel
|
Early work is underway on Coreboot for AMD's Barcelo as the successor to Lucienne.
Merged on Saturday were the first bits of new code referencing AMD's Barcelo landing in Coreboot. The Barcelo code so far is building off the existing Cezanne Coreboot code while the bits added this weekend were for a new GPU PCI ID and vBIOS ID remapping. More code is needed moving forward but this shows they are already working towards Barcelo bring-up for Coreboot with these initial baby steps.
Barcelo is expected next year as the successor to the Lucienne (Zen 2 models) Ryzen 5000 series parts. Barcelo is a combination of Zen 3 CPU cores and still Radeon Vega 7 graphics. Barcelo is a step below Rembrandt that has Zen 3 cores with Navi 2 graphics and other notable improvements.
It's good to see more AMD platform Coreboot work happening early, likely driven in large part for Google Chromebooks. In addition to the job postings from a few months ago, AMD does currently have more Coreboot-related job openings for Coreboot BIOS engineering, emulation systems design engineer, BIOS firmware engineer, and more.
| 9
| 1,760,719,374.370374
|
https://www.phoronix.com/news/AMD-AOCC-3.1-Released
|
AMD AOCC 3.1 Compiler Released - Rebased On LLVM 12.0
|
Michael Larabel
|
AMD earlier this week quietly published a new version of its AOCC code compiler that is now rebased against the upstream LLVM/Clang 12.0 compiler state.
AMD Optimizing C/C++ Compiler 3.0 was released back in March alongside the EPYC 7003 "Milan" launch. AOCC 3.1 is now available as the latest incremental improvement to this LLVM/Clang downstream that focuses on carrying various out-of-tree patches optimizing the open-source compiler for AMD's Zen microarchitecture family, making Flang suitable for compiling more Fortran code-bases, and other enhancements when building code for AMD CPUs.
With AOCC 3.1 the main change is re-basing its LLVM code-base and associated sub-projects against the state of upstream LLVM 12.0 from April. Thus AOCC is now shipping against the latest stable code-base for LLVM while it will be succeeded by the upstream LLVM 13.0 release due out in September.
AOCC 3.1 also adds vector / inline / unroll related pragma directives in the Flang Fortran compiler and expanded its OpenMP 4.5 coverage with Fortran. Those are the only changes officially listed for the AOCC 3.1 release, which remains available as a binary-only compiler release.
AOCC 3.1 downloads and more information are available from developer.amd.com. I'll have up some new AOCC 3.1 compiler benchmarks on Zen 3 shortly.
| 6
| 1,760,719,374.580142
|
https://www.phoronix.com/news/AMD-Debug-S0ix-PMC
|
AMD Pushes Improvements To Help Debug S0ix Power States On Linux
|
Michael Larabel
|
AMD's PMC Linux driver with Linux 5.15 is expected to offer more debugging information for diagnosing S0ix power states behavior to analyze if an AMD SoC is hitting or not the desired low-power states.
S0ix idle standby power states are valuable for energy savings and thus important to ensure such states are properly being achieved when portions of the SoC can power-down. While Intel's S0ix handling on Linux has become fairly mature, on AMD APU/SoC systems there still are some cases of issues around these sub-states to ACPI's S0.
Queued in the x86 platform driver's "for-next" branch this week were a set of patches from AMD for enhancing their amd-pmc driver. Notable on the S0ix debugging front are exposing SMU metrics via DebugFS so S0ix debug statistics are easily obtainable from user-space. The time in various states is reported, the hint count, S0i3 cycle count, and other information.
A follow up patch added additional logging around S0ix counters for ensuring the S0ix states are actually being reached.
Beyond debugging improvements, the AMD-PMC driver improvements for Linux 5.15 also add a new ACPI ID that will be used by future AMD PMC controllers. That added ACPI ID is identified with "AMD_CPU_ID_YC" and comes after AMD_CPU_ID_CZN "Cezanne". This presumably is in reference then to Yellow Carp.
Ultimately it's good to see improved debuggability around S0ix come for AMD platforms in ensuring proper power management for AMD laptops as well as other AMD-powered devices like Steam Deck.
| 4
| 1,760,719,375.93599
|
https://www.phoronix.com/news/AMD-FidelityFX-Super-Res-Source
|
AMD Posts FidelityFX Super Resolution Source Code
|
Michael Larabel
|
After AMD posted FidelityFX Super Resolution last month with various initial launch titles, the source code to this NVIDIA DLSS alternative is now publicly available.
AMD today posted the FidelityFX Super Resolution code today under an MIT license. This alternative to NVIDIA Deep Learning Super Sampling is being supported by an increasing number of game engines and now with the code publicly available open-source game engines and more can more easily adopt this Vulkan (and Direct3D) upscaling technology.
The source code posted today is the full sources and with a CMake build system for the sources. However, it is catering to a Windows workflow with Windows 10 and Visual Studio being listed as being required. However, as shown with Dota 2 on Linux with FSR and support within Steam Play, FSR does work on Linux for Vulkan software.
Those interested in the FidelityFX Super Resolution source code can find it on GitHub.
| 23
| 1,760,719,376.05959
|
https://www.phoronix.com/news/AMD-Zen-3-APU-Temp-hwmon
|
AMD Zen 3 APU Temperature Monitoring Narrowly Misses Linux 5.14
|
Michael Larabel
|
Landing into "hwmon-next" just after the Linux 5.14-rc1 tagging that marks the formal end to feature work for the current cycle was the k10temp driver adding support for AMD Zen 3 APU temperature monitoring under Linux.
The k10temp driver already has support for Zen 3 desktop CPUs and even Zen 3 SP3/Threadripper CPUs with the necessary model IDs in place, but unfortunately the Zen 3 Ryzen APU model ID was missing until the commit on late Sunday.
The change is rather trivial and namely just adding the Family 19h Model 50 bits in place for allowing this k10temp thermal driver to work for Zen 3 APUs.
Zen 3 APU temperature monitoring makes it to hwmon-next.
The patch is small enough that it could be submitted as a "fix" so early in the Linux 5.14 cycle, but as of writing it's only in "hwmon-next" and not marked for stable or submitted as a pull request for 5.14 mainline. Thus this temperature reporting might not appear until Linux 5.15. The patch for k10temp was posted back in May but from the hwmon side was held up by needing this patch with the IDs that worked its way in through the TIP branch.
This k10temp addition didn't come from AMD engineers but rather an end-user after presumably getting his hands on one of the Zen 3 APUs so far to market and finding the Linux temperature reporting not working. It's a pity such a mundane patch didn't work its way into the Linux kernel months ago - even before launch - for such a basic feature but for areas like k10temp it generally hasn't been a focus by AMD and left up to the community.
| 18
| 1,760,719,377.367142
|
https://www.phoronix.com/news/Linux-More-AMD-IOMMU-Opt
|
More AMD IOMMU Optimization Work Is On The Way For Linux
|
Michael Larabel
|
Thanks to one of VMware's Linux engineers there are improvements pending to the AMD IOMMU support code to help with performance.
Earlier this year VMware's Nadav Amit landed a patch to make use of AMD hardware IOMMU functionality to avoid full address-space invalidation by handling page-specific invalidations when needing to flush multiple pages.
The intent was there for making use of this AMD CPU feature for page-specific invalidations in the IOMMU code but it didn't end up taking full advantage. Nadav Amit has been working on a follow-up patch series to further its utilization. New patches make further use of page-selective flushes on AMD CPUs including for unmap operations, avoiding using flush-queue on vIOMMUs, relaxed flushes when gathering, syncing once on scatter-gather map operations, and breaking flushes to naturally aligned ranges on vIOMMU.
Basically, these newer 7 patches make greater use of the selective flushing in the AMD code for more efficient IOMMU handling. Now that the Linux 5.14 merge window is over, the patches have been re-based to 5.14-rc1 while hopefully we will see the work ready for 5.15 in a few months.
| 3
| 1,760,719,377.539466
|
https://www.phoronix.com/news/AMD-VanGogh-Linux-Audio
|
AMD VanGogh APUs Get New Audio Driver For Linux
|
Michael Larabel
|
Since last year AMD has been working on VanGogh APU support for Linux initially with their graphics driver support and that has spread to other areas. It also turns out now that with VanGogh APUs will be a new Linux audio driver.
A new ASoC driver has been posted by AMD this week for the audio co-processor (ACP) on VanGogh. There is nearly two thousand lines of new code for this new "vangogh" driver providing support for the AMD ACP5x (ACP 5.x) audio block.
Details on the new ACP 5.x audio block with Van Gogh are light besides noting that there are now two I2S controllers supported.
The set of 12 patches bringing up this VanGogh audio driver for Linux can be found on the kernel mailing list.
VanGogh APUs are rumored to launch later this year as APUs comprised of Zen 2 CPU cores but featuring RDNA2 (Navi 2) graphics and rumored support for LPDDR5 memory.
| 7
| 1,760,719,378.82039
|
https://www.phoronix.com/news/AMD-SEV-SNP-Linux-v4
|
AMD SEV-SNP Support Revised For Linux + Updated Hyper-V Isolation VM Code
|
Michael Larabel
|
AMD engineers and their partners continue work towards upstreaming Secure Encrypted Virtualization's Secure Nested Paging (SEV-SNP) support for the mainline Linux kernel.
AMD SEV-SNP debuted this year with EPYC 7003 "Milan" processors. SEV-SNP offers additional hardware features for EPYC's virtualization capabilities. With SEV-SNP there is additional memory integrity protections around replay protection, data corruption, memory aliasing, and memory re-mapping. There are also other hardware protections with SEV-SNP as outlined in the comparison below.
AMD has offered out-of-tree SEV-SNP Linux support for early Milan customers while the mainline upbringing is still in progress but at least moving forward in the right direction. Wednesday brought the latest guest support patches that are 36 patches and now up to their fourth revision. This new version adds support to use PSP-filtered CPUID handling, support for the extended guest request, documentation updates, and other low-level changes. This SEV-SNP guest support for VMs is some 3k new lines of code.
Yesterday also saw the latest SEV-SNP hypervisor patches -- 40 patches at 3.5k lines of new code. This KVM-focused hypervisor support for SEV-SNP exposes some new ioctls, adds support for the extended guest message requests, and other changes resulting from earlier code review feedback. Still to be completed on the hypervisor support side is making use of SNP's interrupt security.
Separately, Microsoft sent out their latest patches for the Linux kernel around Hyper-V Isolation VM support. This Hyper-V Isolation VM support is around virtualization-based security and making use of AMD SEV-SNP..
These patches are still undergoing review and it's already late into the 5.14 kernel merge window, so the earliest we'll see these patches picked up would be the 5.15 kernel later in the year. Hopefully by that time this AMD SEV-SNP support will be ready for upstream. AMD began posting the code for the kernel back in March, shortly after the EPYC 7003 debut -- hopefully moving forward they will be more like Intel in regards to upstreaming code prior to product launch to generally ensure splendid out-of-the-box Linux server support on launch day. Or at least ensuring the code is in better shape for upstreaming as soon as the launch occurs rather than still going through rounds of public review months later. In any case the EPYC 7003 series support is in good standing aside from this tardy functionality and the performance continues to be mighty impressive.
Learn more about AMD SEV-SNP in this whitepaper. Those not minding the out-of-tree code can find deployment steps and various sources via AMDESE's AMDSEV GitHub.
| 0
| 1,760,719,379.12462
|
https://www.phoronix.com/news/AMD-Linux-H1-2021
|
AMD's Linux Strides In H1'21 From FreeSync HDMI To PyTorch ROCm
|
Michael Larabel
|
As part of our various Q2'21 and H1'21 Linux/open-source recaps, here is a look back at the most popular AMD Linux/open-source news so far this calendar year.
AMD's Linux graphics team continues making strides on current and future open-source GPU driver support, ROCm continues maturing, AMD continues hiring more Linux engineers on the CPU side of the house, and AMD's latest Zen 3 processors continue performing very well on Linux.
The most popular AMD Linux/open-source related news for H1'2021 on Phoronix included:
AMD Is Currently Hiring More Linux Engineers It looks like thanks to AMD's increasing sales and continuing successes in the enterprise space with more HPC wins and the like, AMD is hiring more Linux engineers. AMD currently has several interesting job openings on the Linux front. Chrome 89 vs. Firefox 86 Performance Benchmarks On AMD Ryzen + Ubuntu Linux Given this week's launch of Chrome 89 and the recent Firefox 86 debut, here are some quick benchmarks for those curious about the current performance when using Ubuntu Linux with a AMD Ryzen 9 5900X and Radeon graphics. dav1d 0.8 Released With More Optimizations - More AMD Performance Dav1d 0.8 was released this weekend (and subsequently 0.8.1 too) as the latest major release for this CPU-based AV1 decoder hosted by the VideoLAN project. Dav1d continues to be about offering the best AV1 decode speed and with the v0.8 series are even faster results -- so here are some of our initial data points as well from some weekend benchmarking. Radeon Vulkan Driver Adds Option Of Rendering Less For ~30% Greater Performance If your current Vulkan-based Radeon Linux gaming performance isn't cutting it and a new GPU is out of your budget or you have been unable to find a desired GPU upgrade in stock, the Mesa RADV driver has added an option likely of interest to you... Well, at least moving forward with this feature being limited to RDNA2 GPUs for now. AMD Ryzen 5000 Temperature Monitoring Support Sent In For Linux 5.12 Due to an unfortunate misalignment of the Ryzen 5000 series launch and the Linux kernel cycles, CPU temperature monitoring for Ryzen 5000 (Zen 3) desktop CPUs isn't landing until now with the Linux 5.12 kernel cycle. The 11 Most Interesting Features For Linux 5.11 - Lots For AMD + Intel This Cycle Linux 5.11 stable is expected to be released on Sunday barring any second thoughts by Linus Torvalds that could lead to an eighth weekly release candidate that would in turn push the official release back by one week. In any case, Linux 5.11 will be formally out soon and it's an exciting one on the feature front. Quake II RTX Performance For AMD Radeon 6000 Series vs. NVIDIA On Linux Last month with the Radeon Software for Linux 21.10 driver there was finally Vulkan ray-tracing support added to that proprietary Vulkan driver component, the first time that Vulkan ray-tracing has been available on Linux for any AMD Radeon 6000 series graphics card across the multiple driver options. Last month I posted some initial Vulkan ray-tracing AMD vs. NVIDIA Linux benchmarks while questions were raised how well the driver performs with NVIDIA's Quake II RTX port. Here are some initial benchmarks for those wondering. AMD Proposing Redesign For How Linux GPU Drivers Work - Explicit Fences Everywhere Well known open-source AMD Linux graphics driver developer Marek Olšák published an initial proposal this week as "a redesign of how Linux graphics drivers work." AMD Publishes Security Analysis Of Zen 3 "PSF" That Could Possibly Lead To A Side-Channel Attack AMD published a security whitepaper this week looking at their Predictive Store Forwarding (PSF) feature that is new to Zen 3 series processors. AMD is going to allow customers to disable this performance feature as they think it may be vulnerable to a Spectre-like attack. AMD FreeSync HDMI Patch Appearing For Their Open-Source Linux Driver While the AMD Linux graphics driver for some time has been supporting FreeSync over DisplayPort connections, FreeSync displays connected via HDMI have not been supported. But now we are finally seeing the start of patches at least as far as HDMI pre-v2.1 support is concerned. Linux 5.11 Released With Intel Integer Scaling, AMD Performance Boost, RTX 30 KMS What better way for open-source enthusiasts to celebrate Valentine's Day than with the stable release of the Linux 5.11 kernel... Linus Torvalds even changed the kernel codename for the occasion to being the "Valentine's Day Edition" kernel. AMD Publishes Initial Open-Source Linux Driver Code For "Beige Goby" AMD has published initial open-source Linux graphics driver code for a new GPU dubbed Beige Goby. System76 Introduces Thelio Mira Computers - Powered By AMD Ryzen 5000 Series System76 today introduced the Thelio Mira as their nwest desktop computer offering that is sized between the Thelio and Thelio Major while still packing quite a bit of compute potential. Linux 5.10.20 Released - Fixes The Erroneous Record-Breaking AMD Clock Frequencies Recent kernel point releases have reported erroneous maximum frequencies on AMD Zen 2 / Zen 3 CPUs in the area of 6GHz+ while now with the latest stable releases that is being fixed. AMD Energy Monitoring Driver Slated To Be Removed From The Linux Kernel As a surprise and big disappointment, the "amd_energy" driver that exposes AMD EPYC server CPU energy monitoring metrics under Linux for being able to calculate the per-core and package power consumption and more is now set to be removed from the mainline Linux kernel. The AMD Zen 2 / Zen 3 Performance Fix For Linux 5.11 Has Landed Just in time for the expected Linux 5.11 stable release on Sunday, the AMD frequency invariance performance regression I've been noting and writing about since Christmas day has been resolved with the previously covered fix having been merged today. Linux 5.12 Corruption, GNOME 40, AMD Milan, Rust In Linux-Next Made For An Exciting March There was a lot to get excited about over the course of the past month when it comes to our open-source and Linux area of focus. Mesa's Radeon "RADV" Vulkan Driver Makes First Steps Towards Ray-Tracing There still is much work left to be completed but Mesa's Radeon Vulkan driver "RADV" has made its first baby steps towards ray-tracing support with Radeon RX 6000 "RDNA2" series hardware. PyTorch 1.8 Released With AMD ROCm Binaries PyTorch 1.8 was released on Thursday as the newest version of this widely-used machine learning library. Exciting many will be easier AMD Radeon ROCm support with Python wheels now provided for that Radeon Open eCosystem support. QEMU 6.0 On The Way With LTO Support, AMD SEV-ES Guests, Multi-Process Experiment This week marked the hard feature freeze for QEMU 6.0 along with the tagging of QEMU 6.0-rc0. The QEMU 6.0 release should happen around the end of April for this important piece of the open-source Linux virtualization stack. The most popular AMD/Radeon Linux hardware reviews and benchmark articles so far this year included:
AMD Radeon RX 6800 vs. NVIDIA RTX 30 Linux Performance Heating Up Given the open-source Radeon driver progress for RDNA2 over the past three months since the Radeon RX 6800 series were launched, here is a look at how the Radeon RX 6800 series and NVIDIA GeForce RTX 30 series is currently competing on Linux when using the latest Linux drivers from the respective vendors. AMD Radeon RX 6700 XT Linux Performance At the start of March AMD announced the Radeon RX 6700 XT as their new RDNA2 graphics card starting out at $479 USD. Tomorrow the RX 6700 XT is going on sale while today marks the embargo lift on reviews. We have been testing the Radeon RX 6700 XT over the past two weeks and have up our initial Linux support experience and gaming benchmark results to share. Linux 5.11 Is Now Looking Great For AMD Zen 2 / Zen 3 Performance Not only is the AMD "CPU frequency invariance regression" from that new support with the in-development Linux 5.11 kernel on course to address the performance shortcomings I outlined last month, but with the patched kernel for a number of workloads the performance is now ahead of where it was at with Linux 5.10. Benchmarking AMD Ryzen 5 5500U Linux Performance With A $450 Lenovo Laptop The AMD Ryzen 5 5500U with six cores / twelve threads within a Lenovo laptop at $449 USD is quite a steal. This is also my first time benchmarking the AMD Ryzen 5 5500U after waiting months on Ryzen 5000 series laptop availability. Here are some initial benchmarks of the Ryzen 5 5500U under Ubuntu 21.04 Linux against various other Intel/AMD laptops. Windows 10 Build 21370 vs. Ubuntu 21.04 Linux On AMD Ryzen 5900X Last month when carrying out tests of Windows 10 vs. Linux on the Intel Core i9 11900K "Rocket Lake" processor we were very surprised to see Windows 10 frankly performing so well compared to Ubuntu and picking up more wins than usual. That unexpectedly strong showing for Windows 10 might be due to Intel's P-State behavior with Rocket Lake or other power management tuning or there the lack of on Linux at this time. But it led me to wondering if the latest Windows 10 updates spelled out anything different on the AMD Ryzen side... So here are some benchmarks of the latest Microsoft Windows 10 against Ubuntu 21.04 on the same AMD Ryzen 9 5900X system. Windows 10 vs. Ubuntu 21.04 On The AMD Ryzen Threadripper 3990X Earlier this month were benchmarks looking at Windows 10 vs. Ubuntu 21.04 on an AMD Ryzen 9 5900X desktop to which Ubuntu came out roughly 8% faster than the Microsoft OS on average. But what about the difference for HEDT systems? Given the more radical performance difference we have seen in the past with Windows vs. Linux for Threadripper systems, here are some recently conducted benchmarks on that front with the 64-core Threadripper 3990X. NVIDIA RTX 30 Series vs. AMD Radeon Linux Gaming Performance For April 2021 With recently receiving the rest of the NVIDIA GeForce RTX 30 "Ampere" line-up we had no access to previously for testing, the past few weeks were busy with testing/re-testing these new graphics cards as well as prior GeForce RTX 20 series hardware and relevant AMD Radeon graphics cards for offering a current look at the 1440p and 4K Linux gaming performance. GCC 10 vs. GCC 11 Compiler Performance On AMD Zen 3 After recently looking at the early LLVM Clang 12 compiler performance on the AMD Ryzen 9 5950X, in today's benchmarking is a look at how the GCC 11 compiler performance is looking in its near final state compared to GCC 10 under a variety of build CFLAGS/CXXFLAGS configurations on the AMD Zen 3 desktop. Updated Portal 2 Vulkan Rendering Code Yielding Great Radeon Results Valve on Monday rolled out a new Portal 2 build that improves its new Vulkan renderer support. For those interested here are some fresh benchmarks of Portal 2 with OpenGL and Vulkan on the open-source AMD Radeon Linux drivers. Radeon RX 6800 Series Linux Performance Nearly Three Months After Launch Given the daily progress and changes made to the open-source AMDGPU Linux kernel driver and the Mesa drivers providing the open-source OpenGL (RadeonSI) and Vulkan (RADV) support, here is a look at how the Radeon RX 6800 series performance is currently for the latest Linux graphics driver code compared to the performance seen back on the November launch day for the Radeon RX 6800 and RX 6800 XT graphics cards.
| 5
| 1,760,719,380.481035
|
https://www.phoronix.com/news/Linux-AMD-EDAC-HBM2-Aldebaran
|
Linux Prepares For AMD Servers With Aldebaran GPU Nodes Sporting HBM2
|
Michael Larabel
|
The latest public code patches on the mailing list today are preparing for newer AMD heterogeneous servers that will have Aldebaran GPU nodes connected via xGMI links to the CPU(s) and the GPU dies in turn having HBM2 memory.
These new heterogeneous AMD system details were revealed today as part of a set of patches prepping the AMD64 EDAC (Error Detection And Correction) kernel driver code for non-CPU nodes. The AMD64 EDAC driver is for traditionally dealing with and correcting system DRAM ECC errors while now being extended to GPU node memory accessible from the CPUs via the xGMI high-speed interconnect.
The public patches note that there will be systems with GPU nodes connected via xGMI links and the GPU dies have HBM2 memory. The patches go on to confirm those nodes as being Aldebaran, the codename for a next-gen AMD CDNA GPU/accelerator that saw initial kernel driver support in Linux 5.13 and continues seeing more open-source driver work around it. Aldebaran is the apparent successor to MI100 "Arcturus" and thus presumably will debut as something along the lines of the AMD Instinct MI200.
These patches published a short time ago note that Aldebaran has two dies (further confirming Aldebaran as an MCM design) with each having four unified memory controllers (UMCs). Each unified memory controller manages eight memory channels that each are connected to 2GB of HBM2 (or HBM2E) memory.
The seven patches posted prepare the EDAC memory driver for the notion of connected non-CPU nodes, recognizing the HBM Gen2 memory type, address translation on Data Fabric version 3.5, and related plumbing. Getting this Linux support squared away timely is being driven by the dominance of Linux in the HPC space and especially with AMD's increasing supercomputer design wins. Most notably Aldebaran and in turn this Linux code is likely what we are to see within the upcoming Frontier exascale supercomputer where it has been mentioned already to have the coherent interconnect between the EPYC CPUs and Radeon Instinct GPUs.
Given the timing of these patches with the Linux 5.14 merge window already open, these amd64_edac additions will likely land for Linux 5.15 unless drawn out by an extended review process.
| 1
| 1,760,719,380.660455
|
https://www.phoronix.com/news/Linux-EPYC-Breakout-KVM
|
Linux Formerly Affected By A KVM SVM Guest-To-Host Breakout Code Vulnerability
|
Michael Larabel
|
Linux's KVM virtualization component previously could allow a virtual machine guest relying on AMD SVM virtualization to breakout into the host. This bug persisted in the Linux kernel from late 2020 to March 2021 before being addressed and is the first known issue of such a guest-to-host breakout that didn't also depend upon bugs within user-space components.
Google's Project Zero initiative today published a case study on this guest-to-host breakout issue. This vulnerability doesn't stem from an issue with the AMD EPYC processors but rather a bug that was within the Kernel-based Virtual Machine (KVM) AMD backend code around its SVM nested virtualization handling.
The case study in full can be read on Google's blog and is an interesting technical read for anyone interested in the matter in great detail.
Linux 5.10 and 5.11 were originally affected by this vulnerability tracked as CVE-2021-29657 before being fixed late in the Linux 5.12 cycle and fixes being back-ported to point releases on those prior series. This is troublesome considering most major cloud providers rely on KVM, but at least the bug was short lived within the kernel and just a software bug while in this case one of the fortunate times public/cloud infrastructure doesn't shift too quickly to new kernel versions.
The post concludes with, "This blog post describes a KVM-only VM escape made possible by a small bug in KVM’s AMD-specific code for supporting nested virtualization. Luckily, the feature that made this bug exploitable was only included in two kernel versions (v5.10, v5.11) before the issue was spotted, reducing the real-life impact of the vulnerability to a minimum. The bug and its exploit still serve as a demonstration that highly exploitable security vulnerabilities can still exist in the very core of a virtualization engine, which is almost certainly a small and well audited codebase. While the attack surface of a hypervisor such as KVM is relatively small from a pure LoC perspective, its low level nature, close interaction with hardware and pure complexity makes it very hard to avoid security-critical bugs."
The Google engineers involved call for greater investing in detection capabilities and better isolation of VM hosts. Fortunately, while this software vulnerability was not introduced by AMD engineers, AMD has been hiring more Linux software engineers in recent months and they continue posting new openings, including in the area of virtualization, as they continue improving their Linux software support overall.
| 2
| 1,760,719,382.363854
|
https://www.phoronix.com/news/AMD-SFH-Next-Gen-Patches
|
AMD SFH Linux Driver Updated For "Next Gen" Ryzen Laptops
|
Michael Larabel
|
There's the next chapter to the unfortunately rather sad state of the AMD Sensor Fusion Hub (SFH) driver support under Linux.
Since 2018 with AMD Ryzen laptops there has been the Sensor Fusion Hub for various accelerometer/gyroscopic sensor functionality, among other uses and akin to Intel's Sensor Hub. It wasn't though until January 2020 that AMD published their SFH driver for Linux.
But while that code was published at the start of last year, the SFH driver wasn't merged until late 2020 for the Linux 5.11 kernel due to requiring multiple revisions before it was ready for mainline acceptance. Granted, SFH isn't critical functionality for most users, but still unfortunate it took nearly three years to be mainlined.
Now the next chapter are the SFH updates slated for Linux 5.14 with its merge window opening in the next week or two. The HID "for-next" Git branch has pulled in the AMD SFH updates for this next kernel.
Notable with the SFH driver is preparing it for multi-generation support with the initial Linux driver only supporting the first-generation SFH platforms. That was followed up by extending it for "AMD next generation SoC's, but in this case it's next generation for what the driver supports but unfortunately current and prior generation for what's actually to market.
The Linux support is a bit late here.
"Extend ALS support for AMD next generation SoC's like Renoir, Cezanne." AMD Renoir was launched back in January 2020 (prior to the initial AMD SFH Linux driver being even published) as their Zen 2 + Vega APU successor to Picasso. Cezanne at least are this year's current-generation Zen 3 APUs up through the Ryzen 9 5980HX. So in this case the "next generation" while new to the Linux driver is quite far behind but at least in-step this also now supports Cezanne.
The SFH work for Linux 5.14 is rounded out by HPD sensor support for Human Presence Detection on supported platforms.
Linux 5.14 should be out in time to power autumn 2021 Linux distributions. At least SFH isn't really necessary for most Linux users and all core functionality on Renoir has been in good shape. I'm still working on getting my hands on a Cezanne laptop for Linux testing to report more there. On that Cezanne front it's still been difficult finding laptops in-stock in the US that aren't powered by NVIDIA RTX graphics (given most Linux users preferring Radeon graphics) or $2k+ for the new Ryzen PRO ThinkPads. But hopefully soon will find an interesting model available for Linux testing. Anyhow, at least AMD is making more progress in making more Linux hires and planning for enhanced Linux support moving ahead.
| 8
| 1,760,719,382.449294
|
https://www.phoronix.com/news/AMD-PSF-Control-Still-Waiting
|
AMD PSF Control Support Still Awaiting The Mainline Linux Kernel
|
Michael Larabel
|
It's been three months since AMD published a security whitepaper outlining the possibility of a side channel attack with PSF. The Predictive Store Forwarding functionality is new to AMD Zen 3 (Ryzen 5000 / EPYC 7003 series) processors and as part of their security analysis they are allowing users the ability to opt-out of using this feature in the name of greater security but the feature still hasn't been picked up for the mainline Linux kernel.
While the security whitepaper mentioned Linux patches for allowing PSF to be disabled, it wasn't until days after that when the PSF control patches were published.
Fortunately, the risk of any real-world impact from Predictive Store Forwarding leading to a side-channel attack is low and disabling PSF appears to have a minimal performance hit at least when toggling the bit back in April, but now mid-way through summer this feature still hasn't been picked up by the mainline kernel tree.
Since the original patches in April, the AMD PSF control patch for Linux is currently up to its sixth revision. Over the past three months there have been multiple revisions ranging from relabeling it from a "mitigation" to feature "control" (though seemingly since relabeled again back as a mitigation) to code refactoring and other low-level changes as a result of upstream review.
The control knob for disabling AMD PSF has also been renamed from the original revision and what was reported in the whitepaper. Once the patch lands, the way to disable PSF for Zen 3 and later (assuming no further changes) is by using predictive_store_fwd_disable=on to disable the Predictive Store Forwarding feature.
From the v6 revision concerns were again over the "predictive_store_fwd_disable=on" option being a double negative to disable the functionality and some possible moving around of the code. But the last ping by the AMD engineer last week looking for anymore feedback on the PSF control/mitigation has so far not turned up any responses.
Thus for now the work is just on the kernel mailing list and hasn't been picked up by any "-next" branches. We'll see if there is any activity in the next few days considering the Linux 5.14 merge window is opening up in the next week or two while being days ahead of the three month mark since this security analysis was made public. This looks to be an area where AMD has room to further improve upon considering Intel for their well known security mitigations has setup the processes, clout among the upstream kernel community, and their large pool of Linux kernel engineers for being able to usher in such work normally on same-day / embargo lift days, granted usually for more pressing issues.
In any case, we'll see if the PSF control lands for Linux 5.14 and back-ported to prior stable series. Once it happens, I'll be back through with a fresh round of benchmarks.
| 3
| 1,760,719,383.707771
|
https://www.phoronix.com/news/AMD-SEV-Local-Migration
|
AMD SEV/SEV-ES Local Migration Support Patches For Linux
|
Michael Larabel
|
Google engineers have prepared a set of Linux kernel patches allowing for AMD Secure Encrypted Virtualization (SEV) / SEV-ES encrypted state to allow for local migration support of these encrypted virtual machines on the same host.
Local migration of VMs allow for moving the guest to a new user-space VMM within the same host such as when upgrading/changing its resources or other alterations to the virtual machine but short of remote migration to a different host.
With Secure Encrypted Virtualization on EPYC processors isolating the VMs from the hypervisor with encrypted memory and SEV-ES also protecting the CPU register state, VM migration support requires special handling. In particular, the proper and secure passing of SEV metadata from the old to new VMM.
With this kernel patch series submitted by Google, the just under 500 lines of new code allow for SEV/SEV-ES local migration support with the Kernel-based Virtual Machine (KVM). Meanwhile on their latest SEV front with EPYC 7003 series processors, the SEV-SNP upstreaming is still ongoing but hopefully that too will be settled soon.
This isn't the first time we have seen Google engineers submitting patches to the Linux kernel for AMD EPYC server CPU features. They have in the past worked on other features like the RAPL support for EPYC and more. For some of these non-critical features they have been at times a bit tardy to arrive for mainline until tackled by the likes of Google, but with AMD continuing to hire more Linux engineers it looks like they are working to provide much more robust support moving forward.
| 1
| 1,760,719,383.974109
|
https://www.phoronix.com/news/AMD-PTDMA-Linux-Driver-v10
|
AMD Continues Working To Mainline Their PTDMA Driver For Linux
|
Michael Larabel
|
Published all the way back in September 2019 was a Linux driver for supporting the Pass-Through DMA controller for EPYC processors. The PTDMA hardware allows for high bandwidth memory-to-memory and I/O copy operations. Now mid-way through 2021 that AMD PTDMA Linux driver remains in the works and is up to its tenth driver revision while waiting to see if it's now ready for mainline or further changes are still deemed necessary.
This AMD PTDMA controller and driver is optimized for use with AMD Non-Transparent Bridge (NTB) devices and not general purpose DMA. NTB is used for in connecting multiple separate memory systems to the same PCI Express fabric. The PTDMA driver supports the 0x1498 (PCI device ID) controller found within EPYC processors since 7002 "Rome".
Following the initial patch series in September 2019, the AMD PTDMA Linux driver quickly went through several rounds of review over the following months. Later revisions continued but took more time between patch submissions while now as of today is up to its tenth revision. This latest revision has various low-level code improvements and clean-ups. The v10 PTDMA patches can be found on the kernel mailing list.
We'll see in the days/weeks ahead if any other issues with the new driver submission are raised or if it's now deemed in good enough shape for finally mainlining in the Linux kernel. Hopefully with AMD continuing to hire more Linux engineers, issues like these with belated drivers will become a thing of the past.
| 0
| 1,760,719,385.202513
|
https://www.phoronix.com/news/OnLogic-TM800-AMD-Client
|
OnLogic Introduces New AMD Ryzen Powered Industrial Grade Mini-ITX Thin Client
|
Michael Larabel
|
Often times when checking out thin clients and industrial PCs / IoT units from OnLogic and other industrial PC vendors, they tend to be Intel powered and within the forums the first question is often about any AMD equivalent... For those wondering about any new AMD-powered thin clients / mini-ITX systems within an aluminum chassis, OnLogic has introduced the TM800.
The OnLogic TM800 is an industrial-grade AMD Ryzen powered Mini-ITX thin client that via the Radeon Vega 8 graphics can drive four displays simultaneously.
The OnLogic TM800 is based around an AMD Ryzen V1605B SoC with quad-core processo and having Vega 8 graphics. The device sports four DisplayPort outputs, dual Gigabit Ethernet, various USB2/USB3 ports, a full size mPCIe socket, and more. The V1605B is quad-core with SMT for eight threads, 2.0GHz base frequency, and 3.6GHz maximum frequency while having a TDP between 12 and 25 Watts.
This industrial thin client powered by the AMD Ryzen V1605B starts out at $934 USD for its base price. More details on the new unit at OnLogic.com.
| 17
| 1,760,719,385.429445
|
https://www.phoronix.com/news/AMD-Kernel-Engineers-June-2021
|
AMD Is Hiring More Linux Engineers For The Scheduler, Memory Management, Net I/O
|
Michael Larabel
|
It looks like AMD's rising marketshare in the data center is paying off as AMD is hiring more Linux kernel engineers.
On top of hiring more Linux engineers earlier this year as part of a client-focused push, it's been brought to my attention they are now looking to hire several more Linux kernel engineers. This time around they appear to be focused on Linux kernel work in the server / EPYC space but some of that work does also carry over to benefit AMD desktop/mobile efforts as well.
AMD's latest round of job postings pertaining to Linux are based out of a mix of Austin, Texas and Bangalore, India. Among their many Linux job openings at the moment:
Linux Kernel - Scheduler Development Lead - Great to see AMD working on kernel scheduler improvements with an emphasis on performance. Hopefully this will also carry over to CPUFreq / CPPC with Schedutil and related work there. There is apparently a "small scheduler focused team" building up by AMD in India with a focus on optimizations/features for EPYC.
Linux Kernel - Memory Management Development/Performance Lead - Again, great seeing AMD working more on low-level Linux kernel infrastructure and more performance optimizations.
Networking I/O Lead - In Austin, it's great to see AMD hiring for multiple networking I/O related Linux positions to focus on performance and low-latency I/O and scalability.
Linux Virtualization Performance Lead - One of several positions around Linux virtualization for EPYC with a focus on KVM/QEMU.
See more of AMD's open Linux-related positions via jobs.amd.com.
While they have worked to ensure sufficient launch-day Linux support across their product portfolio, they have rather neglected any core infrastructure improvements for Linux the past number of years since they shutdown the OSRC. So it's great to see them seemingly getting more talent to work on upstream core kernel improvements beyond just new product enablement and working to tune their hardware for optimal performance/functionality on Linux. This is one of the areas where Intel has long held an advantage over other vendors is with their vast teams of experienced open-source/Linux engineers working not only on hardware bring-up but also improving the Linux kernel and related components around power management, scheduler enhancements, and more along with ensuring other key software is well tuned for their hardware.
| 44
| 1,760,719,386.722534
|
https://www.phoronix.com/news/AMD-SmartShift-Linux-Still-Come
|
AMD Continues Working On SmartShift Support For Linux
|
Michael Larabel
|
Announced last year by AMD was SmartShift Technology for laptops with both AMD CPUs and GPUs to allow dynamically shifting the power budget between the CPU/GPU depending upon the current workload. AMD promoted SmartShift as delivering up to 14% extra performance and now this technology is being worked on for their Linux driver.
AMD describes SmartShift as "AMD SmartShift enabled notebooks feature a hardware boosting interface between the Processor and Graphics with machine learning algorithms to automatically boost performance where workloads demand it. This interface links the common Infinity Control Fabric blocks together so that the CPU and GPU react quickly to different workloads."
SmartShift appeared in select all-AMD notebooks last year before saying the roll-out was paused until 2021. At Computex Taipei 2021 they talked again about SmartShift briefly and we are also seeing the latest Linux kernel patches posted for enabling SmartShift on Linux.
AMD this week for Computex Taipei began re-talking about SmartShift after announcing it last year. (Coincidentally?) new Linux patches have also been appearing. Published on Sunday was the fifth time a patch for AMDGPU was submitted for enabling SmartShift with the discrete Radeon notebook GPU when part of a supported system.
Today was also another patch sent out again for exposing the SmartShift power-share information to user-space via sysfs. This exposes the APU and dGPU power sharing information to user-space for monitoring the behavior and verify that it is in fact working on a given system. The sysfs information will indicate how much of the APU power headroom is being used by the discrete GPU.
A follow-up patch adds a writable sysfs attribute for controlling the SmartShift bias level. From -100 to 100 is the interface for setting -100 to prefer maximum APU performance while 100 is for preferring maximum GPU performance with the default 0 level being balanced performance.
So it seems the SmartShift support for Linux is getting squared away in time for the next round of AMD Ryzen + Radeon laptops coming to market. Presumably this time around there will be more prevalent SmartShift support given their newly-announced AMD Advantage program and more time to bake the SmartShift functionality since last year's initial debut.
These SmartShift Linux patches haven't yet been queued into any development trees but if deemed ready soon enough could still make it for the 5.14 kernel otherwise postponed to later in the year.
| 12
| 1,760,719,386.774501
|
https://www.phoronix.com/news/AMD-PSF-Linux-v6
|
AMD Publishes Latest Linux Patch To Toggle Predictive Store Forwarding (PSF)
|
Michael Larabel
|
It's been a month and a half since AMD published a security analysis of their new Zen 3 "Predictive Store Forwarding" feature that while helping performance could theoretically lead to a new side-channel attack. While they published a Linux patch to allow disabling PSF if desired for increased security, to this day they remain in the works and have yet to be mainlined.
Days after that security white-paper was published AMD began posting the Linux patches to disable PSF optionally for users concerned around the possible security implications of Predictive Store Forwarding. As of Monday, there have now been six rounds of these patches for simply offering up the PSF user control.
The revisions consolidated the work, renamed PSF to a "control" rather than a "mitigation", fixed up some MSR handling, and now has been revved the sixth time. With this latest revision it shifts around the code, simplifies the PSF "mitigation" (but I thought a few rounds ago was changed to not be called a mitigation...), and changes the kernel parameter to be called predictive_store_fwd_disable= with accepted values of on/off (yes, so "on" means to disable).
With the latest patch, On kernels with the latest patch, booting with predictive_store_fwd_disable=on will disable Predictive Store Forwarding for AMD Zen 3 processors (Ryzen 5000 and EPYC 7003 processors).
Given the timing of this patch being past the Linux 5.13 merge window, it remains to be seen if they will try to mainline it this cycle as a security fix or rather just punt it off for Linux 5.14 as a feature during the next merge window, assuming no further rounds of this patch are needed for providing this PSF control that would further delay its mainlining.
| 5
| 1,760,719,388.207882
|
https://www.phoronix.com/news/AMD-VMA-Changes-Frontier
|
AMD Preparing More Linux Code For The Frontier Supercomputer
|
Michael Larabel
|
Frontier as the first US exascale supercomputer being commissioned by Oak Ridge National Laboratory and the Department of Energy. while being powered by AMD CPUs/GPUs is in the process of seeing more Linux kernel changes for bringing up the new platform.
Frontier is being powered by AMD EPYC and Radeon Instinct accelerators. While set to be delivered in 2021, the Linux software support continues to be worked on for making this supercomputer a reality. In particular, the latest code sent out is working on coherent handling of GPU memory with this supercomputer supporting a coherent interconnect between the CPUs and GPUs. The latest patch series out of AMD is proposing changes to the memory management code around device zone page migration and ultimately for handling page migration and coherent CPU access to video memory.
Building off the recent Heterogeneous Memory Management (HMM) Shared Virtual Memory (SVM) code that is now queued for Linux 5.14, this new patch series is [RFC PATCH 0/5] Support DEVICE_GENERIC memory in migrate_vma_*. Longtime AMD Linux engineer Felix Kuehling sums up the situation as:
AMD is building a system architecture for the Frontier supercomputer with a coherent interconnect between CPUs and GPUs. This hardware architecture allows the CPUs to coherently access GPU device memory. We have hardware in our labs and we are working with our partner HPE on the BIOS, firmware and software for delivery to the DOE.
The system BIOS advertises the GPU device memory (aka VRAM) as SPM (special purpose memory) in the UEFI system address map. The amdgpu driver looks it up with lookup_resource and registers it with devmap as MEMORY_DEVICE_GENERIC using devm_memremap_pages.
Now we're trying to migrate data to and from that memory using the migrate_vma_* helpers so we can support page-based migration in our unified memory allocations, while also supporting CPU access to those pages.
This patch series makes a few changes to make MEMORY_DEVICE_GENERIC pages behave correctly in the migrate_vma_* helpers. We are looking for feedback about this approach. If we're close, what's needed to make our patches acceptable upstream? If we're not close, any suggestions how else to achieve what we are trying to do (i.e. page migration and coherent CPU access to VRAM)?
| 12
| 1,760,719,388.551731
|
https://www.phoronix.com/news/AMD-Beige-Goby-Linux
|
AMD Publishes Initial Open-Source Linux Driver Code For "Beige Goby"
|
Michael Larabel
|
AMD has published initial open-source Linux graphics driver code for a new GPU dubbed Beige Goby.
Beige Goby is the latest Linux-specific codename being used for new hardware enablement in their graphics drivers... Following Dimgrey Cavefish, Sienna Cichlid, Navy Flounder, and others, the newest following their color and fish theme is the Beige Goby.
Beige Goby is another RDNA 2 graphics processor... Perhaps the Radeon RX 6600 series seeing some rumors around recently (edit: or some other lower-tier RX 600 series variant). Most of the Beige Goby enablemnt in the AMDGPU Linux driver is following the Dimgrey Cavefish and Sienna Cichlid code paths.
The initial Beige Goby AMDGPU Linux kernel driver patches can be found via this patch series but not particularly exciting itself with largely following the existing GFX10 code paths and a lot of auto-generated additions. There are fifty thousand lines of new code, but the vast majority of that is the automated header definitions.
Based on the timing of these patches, we'll likely see the work land for Linux 5.14.
| 15
| 1,760,719,390.003118
|
https://www.phoronix.com/news/SEV-Linux-5.13-Cleaning
|
AMD SEV Code Gets Reorganized In Linux 5.13 Ahead Of SEV-SNP Landing
|
Michael Larabel
|
While past the Linux 5.13 merge window, some reorganizing/cleaning to the AMD Secure Encrypted Virtualization (SEV) code was merged to mainline today to make it easier for when the SEV-SNP and other feature code is submitted for Linux 5.14 or later.
While post merge window activity is normally limited to bug/regression fixes for the Linux kernel, some reorganization to the SEV code was merged today ahead of Linux 5.13-rc2. The basis for merging the code now is that it will vastly improve the experience when it comes time to landing new feature code moving forward. Most notably, the SEV-SNP code is out under review and those patches will depend upon this reorganized SEV kernel code.
Today's x86 merge further explained, "the three SEV commits are not really urgent material. But we figured since getting them in now will avoid a huge amount of conflicts between future SEV changes touching tip, the kvm and probably other trees, sending them to you now would be best. The idea is that the tip, kvm etc branches for 5.14 will all base ontop of -rc2 and thus everything will be peachy. What is more, those changes are purely mechanical and defines movement so they should be fine to go now (famous last words)."
Whether SEV-SNP will be ready in time though for mainlining in Linux 5.14 or wait for a later version remains to be determined. SEV Secure Nested Paging is new to AMD EPYC 7003 "Milan" processors for adding integrity protections -- all the technical details are laid out in this whitepaper.
| 0
| 1,760,719,391.131575
|
https://www.phoronix.com/news/Linux-AMD-MCE-Rework-DF3
|
AMD Refactors MCE Driver Code, Prepares For Future While Finally Adding DF3/Rome Support
|
Michael Larabel
|
AMD has published a set of patches refactoring their MCE kernel driver, making various machine check architecture (MCA) address translation updates in preparing for "future systems" while at the same time finally introducing Data Fabric 3 support for EPYC 7002 "Rome" processors and newer.
Published on Friday were a set of 25 patches to improve the AMD MCA address translation code within their machine-check exception driver. The patch cover letter notes that "the reference code was recently refactored in preparation for updates for future systems."
The vast majority of the patch work is refactoring of the existing code. The patch series doesn't introduce any of the new support for those "future systems" yet, but it does finally add address translation support for Data Fabric version 3 servers. Data Fabric version 3 is for EPYC 7002 "Rome" and later. That support is arguably long overdue with EPYC 7003 series having already debuted last quarter and Rome being out since 2019, but in any case it's now here in patch form and hopefully this future system support will come out timely and land presumably before next year's EPYC Genoa launch. With Data Fabric version 3 there are new memory interleaving modes now supported and a number of bit fields have been altered compared to prior generations, which need to be compensated for in the driver.
These AMD Linux patches are out for review on the kernel mailing list. It's too late for seeing in 5.13 but could be tidied up in time for 5.14 later this summer.
| 0
| 1,760,719,391.562511
|
https://www.phoronix.com/news/AMD-Zen-3-Scheduler-Model-LLVM
|
AMD Zen 3 Scheduler Model Finally Added To LLVM/Clang
|
Michael Larabel
|
While last minute AMD Zen 3 "znver3" improvements managed to make it for GCC 11 that was recently released, the recent debut of LLVM 12.0 wasn't so lucky on the Zen 3 support front. There was the very basic enablement that landed in LLVM 12 but now the more complete support isn't expected until LLVM 13 this autumn.
The initial "-march=znver3" support made it into LLVM 12 but the Zen 3 tuned scheduler model hadn't landed even though the initial scheduler model updates were posted for review in January. It was only this weekend that the Zen 3 scheduler model has now landed within LLVM Git for LLVM 13.0 that will debut as stable in September~October or slightly before that if making a LLVM 12.0.1 release with it back-ported.
This Zen 3 scheduler model was built from scratch relying upon LLVM's llvm-mca machine code analyzer for generating realistic instruction costs, etc. The new scheduler model has seen limited testing/benchmarking so far but should help some workloads with RawSpeed being one noted workload so far that this addition has helped.
This commit adds the 14.3k lines of new code providing the current Zen 3 model. It's too bad though that this didn't land months ago in ensuring good out-of-the-box compiler support at launch at least for AMD EPYC 7003 series or ideally even ahead of the Ryzen 5000 series launch last year. Intel for their part added Icelake-Server to GCC and LLVM/Clang back in 2018 and remains well known for their timely open-source enablement well ahead of launch.
LLVM 13.0 should be out in the September or October timeframe (the LLVM 13 release calendar has yet to be published but their second release of the year usually occurs then, depending upon how it plays out with blocker bugs, etc) so we'll see what more AMD optimizations might land still ahead of this next open-source compiler release.
| 8
| 1,760,719,392.708738
|
https://www.phoronix.com/news/AMD-ROCm-CRIU-Prototype
|
AMD Begins Prototyping CRIU Support For ROCm Compute
|
Michael Larabel
|
As part of AMD's growing HPC focus and maturing of their Radeon Open eCosystem GPU compute stack, they ended out this week by making public a prototype implementation of CRIU support for ROCm.
AMD's Radeon open-source graphics software developers are working on Checkpoint/Restore In Userspace (CRIU) handling for ROCm. CRIU allows the ability to freeze a running process and archiving it to disk that can then be thawed/restored later on. This user-space-based solution is, of course, much more tricky when it comes to handling processes interacting with the GPU.
Overnight an initial set of patches were posted for the AMD Radeon graphic's "AMDKFD" kernel code for supporting CRIU. These 17 patches with more than two thousand lines of new kernel code is still in a "request for comments" / prototyping stage.
Ultimately they are working towards being able to upstream this checkpoint/restore support in the AMDKFD driver that will be usable to the ROCm stack.so ROCm applications can be CRIU'ed. The new kernel ioctl for the new capabilities is still not finalized yet, so it may be a while before this support is squared away.
In any case for those interested in CRIU around AMD Radeon compute workloads, see this patch series for more details.
| 5
| 1,760,719,392.978421
|
https://www.phoronix.com/news/Linux-5.13-AMD-Energy-Removed
|
AMD Energy Driver Booted From The Linux 5.13 Kernel
|
Michael Larabel
|
While a lot of new features and improvements have been accumulating for the Linux 5.13 kernel with the ongoing merge window, one of the unfortunate aspects of this new kernel is that the AMD Zen CPU energy driver "amd_energy" is indeed being removed.
It's not being removed as some superior driver is being introduced but rather a disagreement between the upstream kernel maintainer(s) and AMD over the handling of the exposed energy sensors. The situation was laid out earlier this month in AMD Energy Monitoring Driver Slated To Be Removed From The Linux Kernel. Long story short, since last year the AMD Energy sensor information has been limited to root due to the PLATYPUS security vulnerability. HWMON maintainer Guenter Roeck proposed slightly limiting and randomizing the sensor data so it couldn't be used for nefarious purposes but still accurate enough for genuine use-cases and no longer needing to be root-only access. However, AMD engineers didn't like that approach.
With the hardware monitoring subsystem maintainer not wanting the information to be restricted to root-only and AMD not wanting the limiting/randomization approach, Guenter went ahead and removed the driver. That's unfortunate for then losing out on AMD CPU energy data for EPYC processors on Linux 5.13 whether root or not.
Guenter wrote in this week's HWMON pull, "The most notable change is the removal of the amd_energy driver. It was rendered all but unusable by making its attributes privileged-only to work around a security issue. A suggested remedy was rejected by AMD, so the only real solution was to remove the driver. For the future, we'll have to make sure that no privileged-access-only drivers are accepted into the hwmon subsystem in the first place. The hwmon ABI document was updated accordingly."
Hopefully some new solution or compromise can be reached so AMD energy monitoring support can be restored in the mainline kernel.
Aside from stripping out the AMD_Energy driver, the hardware monitoring updates for Linux 5.13 does bring a new driver for FSP-3Y power supplies and a new driver for NZXT Kraken water cooling products. The NZXT Kraken X42 / X52 / X62 / X72 now have sensor monitoring support with the mainline kernel for these all-in-one liquid cooling setups.
| 37
| 1,760,719,394.16673
|
https://www.phoronix.com/news/No-More-AMD-Energy
|
AMD Energy Monitoring Driver Slated To Be Removed From The Linux Kernel
|
Michael Larabel
|
As a surprise and big disappointment, the "amd_energy" driver that exposes AMD EPYC server CPU energy monitoring metrics under Linux for being able to calculate the per-core and package power consumption and more is now set to be removed from the mainline Linux kernel.
The removal of this driver sadly isn't for a case like it's being replaced by some superior solution but rather a disagreement in the exposing of the energy data.
Last year as a result of the PLATYPUS power attack Linux restricted access to such data to root/privileged users.
Linux hardware monitoring "HWMON" subsystem maintainer Guenter Roeck recently came up with an approach to allow for non-root monitoring by slightly limiting and randomizing the amd_energy access so it still should be useful for monitoring but not accurate enough for carry out a PLATYPUS style attack on AMD processors.
That amd_energy change was queued up in hwmon-next and then slated for appearing in the upcoming Linux 5.13 cycle. But now it turns out that AMD is against this change and as a result this driver is slated for removal from the mainline tree.
Roeck decided to "remove amd_energy driver." He reasoned in the commit:
Commit 60268b0e8258 ("hwmon: (amd_energy) modify the visibility of the counters") restricted visibility of AMD energy counters to work around a side-channel attack using energy data to determine which instructions are executed. The attack is described in 'PLATYPUS: Software-based Power Side-Channel Attacks on x86'. It relies on quick and accurate energy readings.
This change made the counters provided by the amd_energy driver effectively unusable for non-provileged users. However, unprivileged read access is the whole point of hardware monitoring attributes.
An attempt to remedy the situation by limiting and randomizing access to chip registers was rejected by AMD. Since the driver is for all practical purposes unusable, remove it. As a heavy user of amd_energy myself when testing AMD EPYC processors, it's sad to see this happen and the driver outright now set to be removed with the next kernel cycle as opposed to just keeping the current (unfortunate) behavior of root-only access. Or, alternatively, allowing the unrestricted but limited/slightly-randomized access behavior via a module option. It's also a shame that at first the amd_energy driver did work on AMD Ryzen processors only to be limited after the fact to EPYC processors.
So assuming no last minute change of course, Linux 5.13 is set to remove this driver and we're back to the situation of not being able to make use of the AMD energy counters on Linux with a mainline kernel -- short of using out-of-tree solutions like Zenpower where your mileage may vary, etc.
| 43
| 1,760,719,394.428299
|
https://www.phoronix.com/news/Linux-5.13-No-Root-AMD-Energy
|
Linux 5.13 Will Stop Restricting CPU Power Metrics Access For AMD Energy Driver
|
Michael Larabel
|
Following the PLATYPUS discovery last year that CPU energy information could be used for possible side channel attacks, the Intel RAPL counters were not only restricted to root but the "amd_energy" driver for exposing CPU energy information on supported Zen series CPUs was also dialed back to root-only in the name of security. Linux 5.13 is introducing a new mechanism so AMD CPUs will be able to still read the energy counters even if not operating as root.
Linux hardware monitoring subsystem (HWMON) maintainer Guenter Roeck authored a change to the AMD_Energy driver and went ahead and committed it on Friday to his hwmon-next tree ahead of the Linux 5.13 merge window opening up later in April. The restricting of energy counters to root privileges as a result of the side-channel attack is "annoying" and so he slightly modified the behavior of the driver. As PLATYPUS relies on quick and accurate energy readings, the approach is to make it.... slightly less accurate.
PLATYPUS was made public last November as using Intel RAPL for a side channel attack. This in turn led the Linux kernel to restricting access to CPU energy counters too root... Quite inconvenient to access compared to other sensors. What the amd_energy driver is now set to do is to cache energy values for a short and randomized period of time. It's only cached for a very short period of time (the random cache time is calculated as "jiffies + HZ + get_random_int() % HZ") but enough that it should make the energy readings unreliable for a PLATYPUS style side channel attack but still should be accurate enough for normal use-cases of wanting to monitor the real-time AND CPU energy metrics while now restoring the ability to do so as a non-root user.
The amd_energy driver remains principally focused on AMD EPYC processors but hopefully in time the AMD Ryzen processors will be restored to similar level of support with this mainline kernel driver that exposes its metrics via sysfs.
This patch is the one introducing the change in hwmon-next.
| 9
| 1,760,719,395.55908
|
https://www.phoronix.com/news/AMD-PSF-Mitigation-Linux
|
AMD Sends Out Linux Kernel Patches To Allow Disabling Predictive Store Forwarding (PSF)
|
Michael Larabel
|
AMD last week published a security whitepaper on Zen 3's Predictive Store Forwarding (PSF) functionality introduced with Ryzen 5000 series and EPYC 7003 series processors. In the whitepaper they mentioned Linux patches were published for allowing this feature to be disabled if concerned about the security risk, well, today those patches were made public.
Hitting the Linux kernel mailing list today were the five patches for mitigating Predictive Store Forwarding if desired. With a patched kernel, PSF remains on by default but can be disabled via the Spectre V4 mitigation control or by setting the nopsfd kernel parameter boot option to just force off this feature. Again, this is only relevant for Zen 3 (and presumably future) CPUs.
AMD doesn't believe there is much real-world security risk but for those concerned the option is available. My initial PSF benchmarking found minimal performance impact when disabling it. Ongoing tests have shown no measurable difference outside of some long-running database workloads where even there was just 1~2% or less. So overall the exposure seems to be quite minimal even if you want to disable it.
The AMD PSF patches out now for public review can be found on the kernel mailing list.
| 1
| 1,760,719,395.969574
|
https://www.phoronix.com/news/AMD-PSF-Security-Analysis
|
AMD Publishes Security Analysis Of Zen 3 "PSF" That Could Possibly Lead To A Side-Channel Attack
|
Michael Larabel
|
AMD published a security whitepaper this week looking at their Predictive Store Forwarding (PSF) feature that is new to Zen 3 series processors. AMD is going to allow customers to disable this performance feature as they think it may be vulnerable to a Spectre-like attack.
Zen 3's Predictive Store Forwarding aims to enhance performance by trying to predict dependencies between loads and stores. PSF can speculatively execute instructions based on what it thinks the result of the load will be and while the predictions should be largely accurate, there is the possibility of incorrect CPU speculation.
AMD researchers believe the impact of bad PSF speculation is similar to that of Spectre V4 (Speculative Store Bypass) and particularly of concern for untrusted code being executed within a sandbox/isolation.
AMD is not aware of any code that would be considered vulnerable to PSF behavior and that the risk with PSF is "likely low" but did provide guidance on disabling the Predictive Store Forwarding behavior. When the Spectre V4 / SSB mitigation is in place, PSF is disabled or there is also a new bit that can be set for Zen 3 CPUs to specifically disable the Predictive Store Forwarding behavior.
Predictive Store Forwarding can be disabled on a per-thread basis. AMD's whitepaper says they have proposed Linux patches that would allow disabling of Predictive Store Forwarding using the Zen 3 "PSFD" bit as well as new kernel command line options of psfd/nopsfd. However, as of writing those patches do not appear to be public.
I've been scouring a variety of sources today and those Linux patches do not appear to have been published yet. Once they are published, I'll be running some benchmarks looking at the performance impact if just disabling this Predictive Store Forwarding functionality in the name of increased security. By default and for most users, AMD is comfortable with recommending it be left on.
See more within this AMD whitepaper and stay tuned for PSF benchmarks.
| 22
| 1,760,719,397.021261
|
https://www.phoronix.com/news/Znver3-GCC-10-Backport
|
AMD Zen 3 Tuning Backported To The GCC 10 Compiler
|
Michael Larabel
|
In the past few weeks since the introduction of the EPYC 7003 "Milan" processors there has finally been AMD Zen 3 "Znver3" tuning work that's been hurried into the GCC 11 compiler code-base ahead of its stable release in the coming weeks. That initial Zen 3 tuning work has also now been back-ported to the GCC 10 branch ahead of its next point release.
The past few weeks have seen several commits to GCC Git by SUSE's Jan Hubicka for correcting/completing the Znver3 targeting being introduced in GCC 11, which should be out later this month or early next month depending upon how the rest of the cycle plays out.
For those that aren't quick to move to major new versions of the GNU Compiler Collection, the GCC 10 branch saw on Wednesday the Znver3 tuning work back-ported.
On releases/gcc-10 is the initial Zen 3 tuning work as well as fixing for some of the latencies. Separately is also a fix to avoid the compiler crashing on -march=amd stemming as a regression from when the AMD Family 19h support was first added.
This Zen 3 tuning thus will be found in the upcoming GCC 10.3 point release. Granted, most Linux distributions don't move to new GCC point releases so easily mid-stream, so its relevance still a ways out unless you are on a rolling-release distribution, relying on a gcc-snapshot type package or third-party builds, or rolling your own compiler. Ideally moving forward we can see more punctual AMD compiler enablement for both GCC and Clang. Even before this Zen 3 tuning support was published, months prior Intel already began volleying their Alder Lake and Sapphire Rapids compiler patches, which isn't unusual as the company is often contributing their compiler patches one or two years at least prior to the processors shipping in understanding the long and fragile release cycles of these open-source compilers. Thus by the time of Intel launching new CPU models, the default compiler on the major Linux distributions tend to already have mature out-of-the-box support in place, rather than months down the road.
For those not wanting to roll your own GCC compiler build of GCC 10 or 11 for the Znver3 tuning, AMD in March did release their LLVM Clang based AOCC 3.0 compiler optimized for EPYC 7003 series.
| 7
| 1,760,719,397.503492
|
https://www.phoronix.com/news/AMD-Linux-Q1-2021-Highlights
|
The Most Popular AMD Linux/Open-Source News For Q1'2021
|
Michael Larabel
|
With AMD's busy Q1 of introducing the Radeon RX 6700 XT graphics card, introducing the Ryzen 5000 mobile series, the AMD EPYC 7003 "Milan" series launch, and continuing to advance their open-source/Linux software efforts, it was another busy quarter. Here is a look back from the Linux/open-source perspective of what interested readers the most.
Not counting our Linux hardware reviews or featured benchmark articles, below is a look at the most viewed AMD/Radeon stories of the quarter out of more than one hundred original articles. AMD is excitingly ramping up their Linux staff, much kernel progress is being made, Linux-friendly vendors like System76 have been expanding their AMD powered offerings, ROCm progress, more Coreboot talk, and other work achieved.
It will be interesting to see what Q2 has in store... I am still checking all the major online retailers daily in still trying to get my hands on Ryzen 5000 series laptops for Linux testing, so stay tuned there as hopefully will find some in stock during Q2. There will also be more EPYC 7003 series Linux benchmarks and other Radeon Linux driver testing planned.
AMD Is Currently Hiring More Linux Engineers It looks like thanks to AMD's increasing sales and continuing successes in the enterprise space with more HPC wins and the like, AMD is hiring more Linux engineers. AMD currently has several interesting job openings on the Linux front. Chrome 89 vs. Firefox 86 Performance Benchmarks On AMD Ryzen + Ubuntu Linux Given this week's launch of Chrome 89 and the recent Firefox 86 debut, here are some quick benchmarks for those curious about the current performance when using Ubuntu Linux with a AMD Ryzen 9 5900X and Radeon graphics. dav1d 0.8 Released With More Optimizations - More AMD Performance Dav1d 0.8 was released this weekend (and subsequently 0.8.1 too) as the latest major release for this CPU-based AV1 decoder hosted by the VideoLAN project. Dav1d continues to be about offering the best AV1 decode speed and with the v0.8 series are even faster results -- so here are some of our initial data points as well from some weekend benchmarking. The 11 Most Interesting Features For Linux 5.11 - Lots For AMD + Intel This Cycle Linux 5.11 stable is expected to be released on Sunday barring any second thoughts by Linus Torvalds that could lead to an eighth weekly release candidate that would in turn push the official release back by one week. In any case, Linux 5.11 will be formally out soon and it's an exciting one on the feature front. System76 Introduces Thelio Mira Computers - Powered By AMD Ryzen 5000 Series System76 today introduced the Thelio Mira as their nwest desktop computer offering that is sized between the Thelio and Thelio Major while still packing quite a bit of compute potential. AMD FreeSync HDMI Patch Appearing For Their Open-Source Linux Driver While the AMD Linux graphics driver for some time has been supporting FreeSync over DisplayPort connections, FreeSync displays connected via HDMI have not been supported. But now we are finally seeing the start of patches at least as far as HDMI pre-v2.1 support is concerned. Linux 5.11 Released With Intel Integer Scaling, AMD Performance Boost, RTX 30 KMS What better way for open-source enthusiasts to celebrate Valentine's Day than with the stable release of the Linux 5.11 kernel... Linus Torvalds even changed the kernel codename for the occasion to being the "Valentine's Day Edition" kernel. Linux 5.10.20 Released - Fixes The Erroneous Record-Breaking AMD Clock Frequencies Recent kernel point releases have reported erroneous maximum frequencies on AMD Zen 2 / Zen 3 CPUs in the area of 6GHz+ while now with the latest stable releases that is being fixed. The AMD Zen 2 / Zen 3 Performance Fix For Linux 5.11 Has Landed Just in time for the expected Linux 5.11 stable release on Sunday, the AMD frequency invariance performance regression I've been noting and writing about since Christmas day has been resolved with the previously covered fix having been merged today. PyTorch 1.8 Released With AMD ROCm Binaries PyTorch 1.8 was released on Thursday as the newest version of this widely-used machine learning library. Exciting many will be easier AMD Radeon ROCm support with Python wheels now provided for that Radeon Open eCosystem support. AMD Has A Very Exciting Announcement Next Week AMD has announced that next week on 15 March they will be hosting a digital launch event for the EPYC 7003 "Milan" processors. QEMU 6.0 On The Way With LTO Support, AMD SEV-ES Guests, Multi-Process Experiment This week marked the hard feature freeze for QEMU 6.0 along with the tagging of QEMU 6.0-rc0. The QEMU 6.0 release should happen around the end of April for this important piece of the open-source Linux virtualization stack. The Latest Open-Source AMD Firmware / Coreboot Happenings In Early 2021 While AMD has been crushing it when it comes to Linux performance and generally delivering good launch-day support, the one area many Linux/open-source advocates have been eager and hopeful to see change is around Coreboot support and ideally open-source firmware support such as by re-opening AGESA. Both inside and outside of AMD there continues being work in this direction. AMD Ryzen 5000 Temperature Monitoring Support Sent In For Linux 5.12 Due to an unfortunate misalignment of the Ryzen 5000 series launch and the Linux kernel cycles, CPU temperature monitoring for Ryzen 5000 (Zen 3) desktop CPUs isn't landing until now with the Linux 5.12 kernel cycle. A Fix Has Been Proposed For The Slower AMD Performance On Linux 5.11 With the in-development Linux 5.11 kernel there are many great features and improvements especially for AMD users with some new drivers and other pleasant enhancements. But as I outlined back on Christmas day: Linux 5.11 Is Regressing Hard For AMD Performance With Schedutil. Fortunately, a fix is now en route to the Linux 5.11 kernel for fixing that performance regression affecting AMD Zen 2/3 desktops and servers. AMD Sends In More "New Stuff" For Radeon Graphics With Linux 5.12 Sent in last week were many AMD graphics driver updates slated for Linux 5.12 including the likes of Radeon RX 6000 series OverDrive support. This week marks another batch of AMDGPU kernel driver changes being submitted to DRM-Next ahead of the Linux 5.12 cycle. AMD + Older Intel CPUs To See Much Faster AES-NI XTS Crypto Performance On Linux 5.12 AMD processors along with older Intel processors will enjoy much faster AES-NI XTS crypto performance with the Linux 5.12 kernel this spring. AMD Announces The Radeon RX 6700 XT For $479 AMD today announced the Radeon RX 6700 XT as the newest RDNA 2 graphics card to begin shipping later this month. System76 Pangolin Laptop Launches - Powered By AMD Ryzen 4000 Series Going back to last December System76 had been teasing a new Pangolin laptop that would be AMD powered. Finally their new laptop has launched with Ryzen 4000 series mobile processors and making use of the integrated Radeon graphics. AMD Clarifies ROCm Compute Support For GUI Applications AMD recently added a notice to the ROCm repository reinforcing their focus on headless, non-GUI workloads while now that statement is being sort of retracted and they have clarified their support intentions around this open-source Radeon Open eCosystem driver stack.
| 0
| 1,760,719,398.69408
|
https://www.phoronix.com/news/AMD-Zen-3-Inst-Fixes-GCC11
|
New AMD Zen 3 Fixes Published For The GCC 11 Compiler
|
Michael Larabel
|
Last week there were a few round of Zen 3 compiler patches published and quickly merged into the GCC 11 compiler code-base ahead of its imminent release, This week there is some new activity albeit fixes for this new "Znver3" target.
Last week saw several patches for working to tune the Znver3 GCC 11 support with correct latencies for more instructions and other optimizing. Today SUSE's Jan Beulich merged a number of GCC x86-64 fixes, including specifically for the Zen 3 support.
The code is about fixing Zen 3 instructions:
For INVLPGB the operand count was wrong (besides %edx there's also %ecx which is an input to the insn). In this case I see little sense in retaining the bogus 2-operand template. Plus swapping of the operands wasn't properly suppressed for Intel syntax.
For PVALIDATE, RMPADJUST, and RMPUPDATE bogus single operand templates were specified. These get retained, as the address operand is the only one really needed to expressed non-default address size, but only for compatibility reasons. Proper multi-operand insn get introduced and the testcases get adjusted / extended accordingly.
While at it also drop the redundant definition of __amd64__ - we already have x86_64 defined (or not) to distinguish 64-bit and non-64-bit cases. We'll see how much more tuning and/or fixes for AMD Zen 3 happens ahead of the GCC 11.1 stable release in roughly one month's time.
As I've said many times before, it's too bad that this Znver3 compiler support wasn't worked out much sooner ahead of any Ryzen 5000 or EPYC 7003 series processors shipping rather than now so late in the GCC 11 cycle on its stage four leg of development. If the Zen 3 compiler support was worked out long ago, there wouldn't be this mad dash and would have allowed for the new compiler target to bake and receive plenty of early testing by enthusiastic Linux users for much broader test coverage. Intel on this front generally publishes their initial compiler support at least a year before the CPUs launch if not two or three years in many cases.
| 17
| 1,760,719,399.055082
|
https://www.phoronix.com/news/GCC-11-Znver3-Round-3
|
Zen 3 GCC Tuning Continues With More Correct Latencies Rather Than "Random Numbers"
|
Michael Larabel
|
On Monday, the AMD EPYC 7003 "Milan" launch day, we finally got to see some serious tuning begin for the Zen 3 "Znver3" CPU target in the GCC compiler after that initial code landed at the end of last year. Yesterday a second Zen 3 tuning patch was published and then today a third tuning patch has made it out.
This third Znver3 tuning patch out today is again from SUSE's Jan Hubicka. He sent it out on the mailing list and right away merged it as a "fix" for the GCC 11 compiler release that will debut as stable in the next month or so as GCC 11.1
This third patch is a small change but is updating the costs of integer divides to match the actual latencies found with Zen 3 processors.
While this change may lead to disabling vectorization in some cases, Hubicka noted with the patch, "in general it is better to have actual latencies than random numbers."
The idiv costs were lowered from the existing values of 16/22/30/45/45 now to 9/10/12/17/17 so the compiler is able to make better decisions for the generated instructions with regards to the instructions cost (latency). Those previous costs aren't entirely "random numbers" as they were carried forward from the Znver2 cost table for Zen 2 but not accurate for AMD's latest microarchitecture. Whether the Znver2 costs are accurate for Zen 2 is another matter.
It will be interesting to see how much more AMD Zen 3 tuning happens in time for the GCC 11.1 stable release. But even with GCC 11.1 being just over a month or so away from release, this updated GNU compiler won't be found out-of-the-box on Ubuntu until 21.10 this autumn while one of the earliest users of it will be Fedora 34. Thus my desire for seeing more timely compiler support out of AMD for the open-source GCC and LLVM/Clang compilers continues as one of the areas where Intel tends to be months or years ahead of schedule. Had this Znver3 tuning been all squared away ahead of time, Linux enthusiasts (especially those on the likes of Arch and Gentoo) could have been banging on the code for a while already with Ryzen 5000 series hardware to help ensure it's fit and optimal ahead of hitting AMD HPC customers and other prominent deployments - the same case as if AOCC 3.0 had been available or as a public beta closer to the Ryzen 5000 series launch.
| 6
| 1,760,719,400.557423
|
https://www.phoronix.com/news/AMD-SEV-SNP-Linux-RFC
|
AMD Sends Out New Linux Code For SEV-SNP With EPYC 7003 Series
|
Michael Larabel
|
In addition to AMD EPYC 7003 "Milan" processors offering fantastic performance, another important highlight for these new Zen 3 server processors is SEV-SNP for upping the Secure Encrypted Virtualization capabilities. AMD has been offering SEV "Secure Nested Paging" patches via a GitHub repository while now they are working towards mainlining this feature for the Linux kernel.
AMD Secure Encrypted Virtualization's Secure Nested Paging builds upon SEV/SEV-ES to offer integrity protections, including against malicious hypervisor attacks. This AMD whitepaper spells out SEV-SNP in more detail for those interested in all of the finer details of this feature round on EPYC 7003 series processors.
Last year we saw AMD working on some early bits around SEV-SNP for the Linux kernel while finally today, one week after the EPYC 7003 series was publicly announced, we are seeing more of the Linux kernel patches work their way out on the kernel mailing list for review and eventual inclusion into the upstream kernel.
Hitting the kernel mailing list minutes ago were SEV-SNP guest support and SEV-SNP hypervisor support. Both of these patch series were sent out as an initial "request for comments" - indicating further revisions are likely necessary before being ready for mainlining into the upstream Linux kernel.
The SEV-SNP guest support currently amounts to 13 patches and implements the initial blocks for being able to boot SEV-SNP VMs. However, this code isn't yet feature complete for maximum security with interrupt security being one of the items not yet implemented. Also still to be completed is CPUID filtering, the ability to query the attestation report, lazy validation, and more.
The SEV-SNP hypervisor support also still has yet to implement the interrupt security and query attestation portions. The hypervisor support is some 30 patches at the moment.
Given they are "RFC" patches and a big ticket security feature, it will likely be some time before these SEV-SNP patches are ready for mainline. The Linux 5.13 merge window is already coming up quickly in just a few weeks now, so presumably this work won't be ready until at least Linux 5.14, but stay tuned to Phoronix for other updates as to the status of this open-source virtualization work. In any case hopefully won't take too long and we'll see what enterprise Linux distribution vendors may pick up these patches early for offering SEV-SNP functionality on EPYC 7003 series servers. It would be nice if the SEV-SNP kernel support was already mainlined and ready for day-one on EPYC Milan without resorting to out-of-tree patches (i.e. more like Intel's very timely pre-launch Linux support) with only rare exceptions, but at least progress is being made by AMD to offer more punctual Linux feature support and they are slowly but surely moving in that right direction.
| 1
| 1,760,719,400.565693
|
https://www.phoronix.com/news/AMD-AOCC-3.0-Compiler
|
AMD AOCC 3.0 Released As Zen 3 Optimized LLVM Clang 12 Based Compiler
|
Michael Larabel
|
With today's AMD EPYC 7003 "Milan" launch there is also the public availability of AOCC 3.0 as their LLVM/Clang downstream now carrying patches for optimized Zen 3 support.
AOCC 3.0 is AMD's downstream of LLVM/Clang focused on offering leading AMD CPU support. With AOCC 3.0 they have re-based against the in-development LLVM 12 trunk code while have enablined additional tuning/optimizations around the EPYC 7003 series. AOCC 3.0 also continues to carry patches/configuration in tuning for AMD's Math Library (AMDLibM), FLANG is enabled for Fortran support, OpenMP is enabled by default, and other tweaks in aiming to deliver the fastest AMD Linux binaries.
AOCC 3.0 is available today as a generic x86_64 Linux binary as well as packages for Debian/Ubuntu and RHEL/CentOS based systems. Ubuntu 20.04 LTS, SUSE Linux Enterprise Server 15, and CentOS/RHEL 8.
AOCC remains closed-source so all of AMD's tuning/optimizations are unfortunately not clear. Upstream LLVM/Clang 12 still has the basic "znver3" enablement, similar to GCC -- up until this morning when GCC receiving some initial tuning for Znver3.
I'll be having benchmarks of AOCC 3.0 shortly on Ryzen 5000 series and EPYC 7003 as well as of the latest GCC and Clang Znver3 performance.
Those wanting to download AOCC 3.0 can grab it from developer.amd.com.
| 5
| 1,760,719,402.015415
|
https://www.phoronix.com/news/GCC-Znver3-Tuning-Patch
|
GCC Compiler Sees New Patch With Tuning For AMD Zen 3 (Znver3)
|
Michael Larabel
|
GCC is in the process of seeing better support for AMD Zen 3 processors with its znver3 target.
Back in December the GCC 11 compiler picked up initial support for AMD Zen 3 (znver3). That support was just enabling the newly-supported CPU instructions found with this latest-generation Zen microarchitecture. Unfortunately missing at that time was any tuning for allowing the compiler to make more informed decisions over instruction scheduling with the instruction costs, etc. So it's been carrying the same data as Zen 2 that is also largely carried over from the original Zen compiler code.
Now today, a few weeks prior to the GCC 11 stable release, a new Zen 3 compiler patch has surfaced. Longtime GCC developer Jan Hubicka of SUSE has provided a "tuning part 1" patch for the GCC Znver3 code.
Hubicka tuned the Znver3 target based on benchmarks but overall he characterized the tuning as a smooth upgrade from Zen 2. The tuning includes adjusting some instructions that now have shorter latencies on Zen 3 CPUs, gather instructions are a lot faster, and FMADD was optimized with Zen 3.
That's all dandy and great to see but it's noted there are some performance regressions remaining and the instruction scheduler could still use some love with it still treating Zen as an in-order CPU, among other areas left for further tuning of the Znver3 optimized code path.
This initial tuning patch can be found on the GCC mailing list.
It's great seeing this patch surface and will hopefully be in good shape and land for the upcoming GCC 11.1 stable even though we are well into stage four (regression fixing) stage.
Timely compiler enablement is still an area where AMD could benefit from improvements. The initial/basic Znver3 support didn't land in trunk until December, after Ryzen 5000 series processors were already sharing. While EPYC 7003 is now imminent, GCC 11.1 won't be out as stable until April and in terms of Linux distribution adoption it won't appear in the likes of Ubuntu until its 21.10 release in October. And as noted the tuning for Zen 3 is still ongoing so will be carried forward to future GCC 11 point releases / GCC 12. If AMD had their Znver3 support out months ago, the eager enthusiasts on the likes of Arch Linux and Gentoo could have already been pounding on this code with Ryzen processors for a while now, helping to flesh out any issues or areas for improvement prior to new EPYC processors hitting their all important HPC and data center customers.
Intel meanwhile sent out their Sapphire Rapids and Alder Lake enablement last summer and continue to work on feature bring-up for areas like AMX. Intel's "icelake-server" target has also been mainlined for more than two years already and in released versions -- for years now Intel has been reliably getting their GCC and LLVM/Clang support upstream around a year or two ahead of launch, even when CPUs arrive on schedule. One of Intel's strengths has been with their very punctual and predictable Linux/open-source support well ahead of launch and ensuring the timing works out for generally having widespread support at launch. Hopefully this will be improved upon moving forward on the AMD side given their growing Linux customer base in areas like HPC.
Hopefully LLVM Clang picks up similar Znver3 tuning soon, but there the timing is even less opportune as LLVM 12.0 is releasing literally any day now. LLVM/Clang 12 has the basic Znver3 enablement while any further tuning will come either for a LLVM 12.0.1 point release in a few months and/or LLVM 13 in the September timeframe.
I'll be running some fresh AMD Zen 3 compiler benchmarks shortly with this latest GCC patch.
| 3
| 1,760,719,402.361887
|
https://www.phoronix.com/news/AMD-Next-Week-Milan
|
AMD Has A Very Exciting Announcement Next Week
|
Michael Larabel
|
AMD has announced that next week on 15 March they will be hosting a digital launch event for the EPYC 7003 "Milan" processors.
On 15 March at 11AM EST they will be unveiling the 3rd Gen AMD EPYC "Milan" processors. Today's announcement is quite bare while there has already been much speculation and alleged leaks regarding EPYC Milan.
With the EPYC 7002 "Rome" processors still dominating against Intel's best, it will be very interesting to see how EPYC 7003 performs ahead of Intel "Ice Lake" Xeon processors. On the desktop side, Ryzen 5000 Zen 3 processors continue impressing on Linux that make us all the more excited for the EPYC 7003 series.
For those interested in high performance computing, you'll definitely want to watch the event next week at AMD.com. That's all for now but stay tuned for an exciting 15 March.
| 26
| 1,760,719,403.558653
|
https://www.phoronix.com/news/AES-NI-CTS-Linux-5.12-AMD
|
AES-NI XTS Crypto Performance Looking Good For AMD With Linux 5.12 Fix
|
Michael Larabel
|
Of the performance-related changes with Linux 5.12 worth noting is faster AES-NI XTS performance for systems relying upon return trampolines "Retpolines" as part of the CPU's Spectre V2 mitigations. On the Intel side this primarily impacts older CPUs where Retpolines is still used while on the AMD side through Zen 3 the Retpolines is still relied upon, which as shown by these benchmarks is now much better off for AMD Ryzen AES XTS performance as measured by Cryptsetup.
As reported last year, AES-NI regressed heavily under Retpolines and seemingly went unnoticed for the better part of three years. Now with Linux 5.12 the AES-NI kernel module code has been reworked so it doesn't face such overhead in Retpolines-enabled environments and in turn really helps out with performance.
I previously ran some benchmarks while now for getting an idea as to the impact with Linux 5.12 mainline, I carried out some fresh cryptsetup benchmarks with two AMD systems of Linux 5.11 stable versus Linux 5.12 Git at the end of the merge window.
The AMD Ryzen 5 (Zen 2) laptop is seeing much better AES-XTS performance:
Again, such change is just expected for AES-XTS with the AES-NI kernel module change for Linux 5.12 on Retpoline-enabled systems. It's not a novel improvement but rather addressing a Retpoline-induced performance regression that went unnoticed until recently.
And now for some Ryzen 9 5950X benchmarks with Cryptsetup on Linux 5.11 vs. 5.12 Git in the default kernel configuration where Retpolines remains in use for Zen 3:
The AES-NI work is among many new/improved features with Linux 5.12 and will debut as stable in about two months time. It's also quite likely this AES-NI work as a "regression fix" will be back-ported to stable kernel series too.
| 5
| 1,760,719,404.090851
|
https://www.phoronix.com/news/LUMI-Preparing-For-AMD-HPC
|
The State Of ROCm For HPC In Early 2021 With CUDA Porting Via HIP, Rewriting With OpenMP
|
Michael Larabel
|
Earlier this month at the virtual FOSDEM 2021 conference was an interesting presentation on how European developers are preparing for AMD-powered supercomputers and beginning to figure out the best approaches for converting existing NVIDIA CUDA GPU code to run on Radeon GPUs as well as whether writing new GPU-focused code with OpenMP device offload is worthwhile.
Georgios Markomanolis who is the lead HPC scientist at CSC was the one presenting at this month's FOSDEM virtual conference on preparing for supercomputers with AMD GPUs. Their focus at the moment is on the LUMI EuroHPC supercomputer that is expected to become operational later this year. LUMI is aiming for 550+ Peta FLOPS of peak performance with being powered by AMD EPYC "Milan" processors and AMD Instinct GPUs (post MI100, perhaps the new "GFX90A").
While awaiting the supercomputer, the HPC researchers in Europe that are involved with the LUMI consortium have already been busy analyzing the Radeon Open eCosystem (ROCm) and the available methods for exploiting the GPU performance in porting existing CUDA codebases over as well as the best practices when writing new code.
For converting CUDA code over for AMD GPU execution, the focus is obviously on using AMD's open-source HIP heterogeneous interface. With the "Hipify" Clang is how source-based translations can be achieved in large part from CUDA or there is also Hipify-Perl for the text-based search/replace in migrating from CUDA to HIP. From the HIP-based approaches they have been seeing good results with roughly 2% overhead.
With Fortran code, Hipfort is necessary as as interface library for the GPU kernel and more manual porting compared to the automatic translation. But with one test case at least under their HIP version they found it to be 30% faster, but part of that at least may also come down to compiler stack differences, as noted.
LUMI researchers are also exploring AMD's OpenMP device offloading support that was recently upstreamed in LLVM and continues being developed in the downstream "AOMP" project. So far they have found AOMP to have some performance issues but are expecting it to be improved by the time LUMI is deployed. They expect HIP will ultimately perform better than using OpenMP offloading but may use OpenMP for Fortran or other complicated codebases.
Those curious about the ROCm/HIP experiences so far with European researchers preparing for AMD-powered LUMI can see Georgios Markomanolis' PDF slide deck from his FOSDEM presentation as well as the WebM/VP9 and MP4 video recordings.
| 116
| 1,760,719,405.08286
|
https://www.phoronix.com/news/Zen-3-Desktop-CPU-k10temp
|
AMD Ryzen 5000 Temperature Monitoring Support Sent In For Linux 5.12
|
Michael Larabel
|
Due to an unfortunate misalignment of the Ryzen 5000 series launch and the Linux kernel cycles, CPU temperature monitoring for Ryzen 5000 (Zen 3) desktop CPUs isn't landing until now with the Linux 5.12 kernel cycle.
The hardware monitoring (HWMON) subsystem updates were sent in this morning for the just-opened Linux 5.12 merge window. There is now Zen 3 desktop CPU temperature monitoring support within the existing k10temp driver. The patch just needed to add the IDs for the Zen 3 parts to the k10temp driver while the rest is unchanged. The two line patch didn't come from AMD but a community developer and has been verified to work on the Ryzen 5800X / 5900X / 5950X processors to provide correct CPU temperature monitoring.
There isn't any CPU energy/power monitor yet with the AMD_Energy driver continuing to be just restricted to EPYC server processors. Linux 5.11 also dropped the voltage/current reporting from Zen processors with the k10temp driver over lack of documentation and as a result various issues and inaccuracies.
This is one of the areas where AMD hiring more Linux engineers could benefit from some more timely improvements.
Outside of the Ryzen Zen 3 desktop CPU temperature monitoring support, Family 19h Model 30h (EPYC 7002 presumably) is matched for the AMD_Energy driver. There is also now support for sensor monitoring on ASRock motherboards with the NCT6683 driver.
The full list of Linux 5.12 HWMON changes can be found via the pull request.
| 63
| 1,760,719,405.657539
|
https://www.phoronix.com/news/AMD-Hiring-More-Linux-2021
|
AMD Is Currently Hiring More Linux Engineers
|
Michael Larabel
|
It looks like thanks to AMD's increasing sales and continuing successes in the enterprise space with more HPC wins and the like, AMD is hiring more Linux engineers. AMD currently has several interesting job openings on the Linux front.
While AMD has been delivering reliable Linux support with their recent launches, there is room for improvement in areas like more timely compiler support for new processors, better alignment of their new hardware enablement for getting the code not only upstreamed but into distributions for launch-day, and similar areas. Based on recent job postings, it looks like AMD is working to make such strides.
Here is a look at some of the new and currently active Linux-related job openings at AMD:
Manager Linux Kernel Development - AMD is looking for a manager to see over all their Linux kernel and virtualization development team. The manager will oversee the engineers working on designing/implementing/testing the Linux features for AMD CPUs. I am not aware of any "Linux manager" currently at AMD so hopefully that increased coordination can lead to more punctual hardware enablement and striking a rhythm like Intel is known for with their Linux hardware support.
Linux Technical Lead - This admittedly sounds like a very fun one and even piqued my personal interest. (Heck partially seems along the lines of what I often end up doing externally for the community with relentlessly trawling of Git repositories and development mailing lists in pointing out new AMD support code and commentary and pointing out base support requirements at launch and versions/patches for more optimized support. Plus benchmarking.) The client engineering Linux lead workload will include coordinating with OEM partners and the tier one Linux distributions around new features and components. This also includes working with those Linux distribution vendors and partners on optimized platform feature support and similar scenarios. Basically, seeing that their partners and distributions stay up-to-date on the AMD Linux support. Hopefully such a lead at AMD will indeed help ensure necessary patches or version requirements are in place for seeing good launch-day Linux support with the major distributions.
AMD is also hiring for Linux engineering and Linux systems architect, among other traditional software/hardware engineering roles.
Several of these new job descriptions do begin with, "step up into a new organization built to engage more strategically and deeply with the technical teams of our commercial customers." Interestingly, I only see that opening line on their current Linux job postings. When asking AMD if there is a "new (Linux) organization" at AMD, the comment was there is no organization to announce but this is part of the overall expansion at AMD. So for now it's back to dreaming about a new unit akin to the defunct AMD Operating System Research Center that previously drove their Linux support or Intel's former Open-Source Technology Center.
In any case, great seeing AMD ramping up their Linux talent. See this Linux jobs search for more of the open positions at AMD, which does include some 2021 internships as well.
Related to our interests, there are also a number of firmware engineer job openings mentioning Coreboot, including some noting "this role enables Coreboot technology on AMD products" -- which may just be for Chromebooks or could be for future server support (some of the posts do mention Chromebooks but not all).
What areas would you like to see AMD improve with regard to their Linux support? Share your thoughts by commenting on this article in the forums.
| 94
| 1,760,719,406.62338
|
https://www.phoronix.com/news/Open-Source-AMD-FW-2021
|
The Latest Open-Source AMD Firmware / Coreboot Happenings In Early 2021
|
Michael Larabel
|
While AMD has been crushing it when it comes to Linux performance and generally delivering good launch-day support, the one area many Linux/open-source advocates have been eager and hopeful to see change is around Coreboot support and ideally open-source firmware support such as by re-opening AGESA. Both inside and outside of AMD there continues being work in this direction.
Piotr Król and Michał Żygowski of consulting firm 3MDEB presented at this weekend's FOSDEM Online 2021 about the latest happenings as of so far this year on the open-source AMD firmware front.
Most of the outside visible work by AMD around any open-source firmware / Coreboot support has been on the Chromebook front, thanks to seeing AMD powered Chromebooks and Google mandating the use of Coreboot. Thus as noted before there has been AMD Picasso work for Coreboot while still sticking to the binary AGESA and not much in the way of motherboard coverage outside of Chromebook devices.
Besides the Picasso work, there is support for AMD Cezanne and Majolica support under review for supporting the AMD Ryzen 5000 series mobile with Coreboot. Again, seemingly for upcoming Chromebook devices.
One remark made during the presentation was that, "despite there are no @amd.com patches going upstream, team contributing to Coreboot is officially hired by AMD" and "congratulations to our OSF friends who joined AMD, we already seethings moving faster."
Also exciting from 3MDEB, "There is a chance that AMD servers will also get [open-source firmware] support." As for that there is Project X as open-source Coreboot/Oreboot for AMD Zen. Prominent Coreboot developer Ron Minnich of Google has been working on that fully open-source Rust-based Oreboot support initially for AMD EPYC with the Rome reference board. But the hope is to ultimately support at least a more affordable developer board and see what happens but for now is a minimal implementation. Minnich's OSFC presentation from last year is embedded below for more details.
Another area where AMD is making progress on open-source firmware albeit in the server space is on supporting OpenBMC.
Those wanting to learn more about 3MDEB's thoughts and expertise on the open-source AMD firmware happenings can see this slide deck (PDF) from FOSDEM 2021. Long story short, there is work happening albeit primarily in the server and Chromebook space at this point but we hold out for more surprises down the road.
| 26
| 1,760,719,407.171048
|
https://www.phoronix.com/news/Linux-5.12-More-AMDGPU
|
More AMDGPU Patches For Linux 5.12 Point To First Sign Towards PCIe 5.0, FreeSync HDMI
|
Michael Larabel
|
Sent in today to DRM-Next ahead of the Linux 5.12 kernel cycle were "fixes" but some interesting items worth noting in this batch.
Primary feature development is over for the DRM graphics drivers of new code they want in for Linux 5.12 since we are now late into the 5.11 release phase, but included as part of today's fixes pull request were a few more notable items.
Already in DRM-Next for this AMD Radeon open-source graphics driver with Linux 5.12 is various power management improvements, Radeon RX 6000 series OverDrive overclocking, FP16 enabled for more hardware, and other mostly lower-level changes.
Of this latest pull request focused on fixes, some last minute items worth pointing out include:
- FreeSync now being enabled for "A+A" configurations. This is for newer AMD APUs where they can scanout directly from the GTT, thus avoiding an extra bounce buffer in vRAM. With this pull request, FreeSync support is now enabled for these newer APUs.
- Restoring of the "set_fan_speed_percent" interface after it was disabled a few months ago for issues with some graphics cards.
- PCI Express 5.0 support is now a recognized PCIe capability. There isn't any new functionality wired up besides recognizing PCI Express Gen5 as a possible link speed. AMD has been supporting PCI Express 4.0 since the Radeon RX 5000 series while PCI Express 5.0 isn't expected to appear with AMD Radeon graphics cards until at least 2022. For punctual open-source support we'll hopefully see work on next-gen GPU support published before year's end and that would be when any interesting bits are revealed. This is just a tiny patch denoting work ultimately in that direction. Heck, elsewhere in the kernel is already patchwork beginning to recognize PCIe 6.0. PCI Express 5.0 provides 32 GT/s transfer rates.
- Enabling of user pointer "userptr" support when the AMDKFD compute driver is enabled.
- Various power management changes/fixes for VanGogh APUs.
- Support for DID 2.0 DSC pass-through. This is for some display panels that have active converters where currently prior to this support could just lead to black screens without this DID 2.0 parsing.
- 0x73A1 as a previously missing AMD RDNA 2 "Sienna Cichlid" device ID.
- As noted last week the AMD FreeSync HDMI patch is now queued. This is for the AMD specific FreeSync implementation over HDMI and not the standardized version found in HDMI 2.1+.
- A workaround for some DisplayPort to VGA dongles. Currently such dongles may end up only supporting 1440 x 900 when connected to a 1920 x 1080 display. The workaround corrects this behavior for some legacy dongles so the DP-VGA adapter can at least run at full HD (1080p).
Pretty good collection of fixes and other work queuing up for Linux 5.12! The Linux 5.12 merge window will formally get underway later this month. Linux 5.12 stable should be out in May. The full list of patches for this latest fixes pull request can be found on dri-devel.
| 20
| 1,760,719,408.433596
|
https://www.phoronix.com/news/Linux-511-CPUFreq-Inv-Fix-Patch
|
AMD Regression On Linux 5.11 Being Addressed By New CPUFreq Patches
|
Michael Larabel
|
The AMD "frequency invariance" saga with Linux 5.11 continues... While there was a patch to address the previously noted performance regression caused by the introduction of frequency invariance and seen when using the Schedutil governor, a new CPUFreq-side patch series has been proposed instead -- both of which are addressing the performance issue with this new kernel for AMD Zen 2 / Zen 3 systems.
It started out prior to Christmas when I noticed Linux 5.11 regressing on AMD systems when tested out of the box. On Christmas I outlined the findings of the AMD performance drops on Linux 5.11 and had narrowed it down to the frequency invariance support added this cycle. When using the default Schedutil governor, it was easy to now encounter slower performance relative to Linux 5.10 and prior.
Later in January SUSE engineer Giovanni Gherdovich, who worked on AMD frequency invariance implementation, proposed a fix for the regression. Testing confirmed the AMD performance now in much better shape with the patched Linux 5.11 and in some cases even better off than with Linux 5.10.
So all looked good with this proposed patch and then it was just a matter of waiting for it to be mainlined... Well, this week when inquiring about it being mainlined, Linux power management subsystem maintainer Rafael Wysocki of Intel ended up noting some areas for improvement with it... In turn, Rafael ended up writing a new patch even without being able to test it on AMD hardware. Rather than modifying Schedutil, his approach was to improve CPUFreq.
He noted with the prototype patch, "What the patch below does is to add an extra entry to the frequency table for each CPU to represent the maximum "boost" frequency, so as to cause that frequency to be used as cpuinfo.max_freq. The reason why I think it is better to extend the frequency tables instead of simply increasing the frequency for the "P0" entry is because the latter may cause "turbo" frequency to be asked for less often."
Testing so far of that new CPUFreq is quite positive. I have been hammering that patched kernel now on multiple AMD EPYC servers and Ryzen laptops/desktops for the past day and will have more details in the hours ahead for the developers. But long story short, this new patch does also take care of the issue of the regressed performance on Linux 5.11 and replaces the prior patch from last month.
With those results looking good, Rafael has spun up the patch series formally to address this issue. Here is the patch summary of the changed CPUFreq behavior in addressing the problem:
The source of the problem is that the maximum performance level taken for computing the arch_max_freq_ratio value used in the x86 scale-invariance code is higher than the one corresponding to the cpuinfo.max_freq value coming from the acpi_cpufreq driver.
This effectively causes the scale-invariant utilization to fall below 100% even if the CPU runs at cpuinfo.max_freq or slightly faster, so the schedutil governor selects a frequency below cpuinfo.max_freq then. That frequency corresponds to a frequency table entry below the maximum performance level necessary to get to the "boost" range of CPU frequencies.
However, if the cpuinfo.max_freq value coming from acpi_cpufreq was higher, the schedutil governor would select higher frequencies which in turn would allow acpi_cpufreq to set more adequate performance levels and to get to the "boost" range of CPU frequencies more often.
This issue affects any systems where acpi_cpufreq is used and the "boost" (or "turbo") frequencies are enabled, not just AMD EPYC. Moreover, commit db865272d9c4 ("cpufreq: Avoid configuring old governors as default with intel_pstate") from the 5.10 development cycle made it extremely easy to default to schedutil even if the preferred driver is acpi_cpufreq as long as intel_pstate is built too, because the mere presence of the latter effectively removes the ondemand governor from the defaults. Distro kernels are likely to include both intel_pstate and acpi_cpufreq on x86, so their users who cannot use intel_pstate or choose to use acpi_cpufreq may easily be affectecd by this issue.
To address this issue, extend the frequency table constructed by acpi_cpufreq for each CPU to cover the entire range of available frequencies (including the "boost" ones) if CPPC is available and indicates that "boost" (or "turbo") frequencies are enabled. That causes cpuinfo.max_freq to become the maximum "boost" frequency of the given CPU (instead of the maximum frequency returned by the ACPI _PSS object that corresponds to the "nominal" performance level).
Long story short, this AMD regression on Linux 5.11 is still pending but hopefully fixed in mainline within the coming days. While testing is ongoing, with all the data I am seeing so far does confirm this new patch series works out well. It's getting late in the Linux 5.11 cycle but this regression fix is ultimately still expected to land in time. For those curious, I'll have out some new benchmark numbers in the next day or two from this testing. For now, back to benchmarking.UPDATE (5 Feb 2021): Benchmarks and more details on the new patch.
| 3
| 1,760,719,409.189853
|
https://www.phoronix.com/news/Linux-5.11-AMD-Inv-Freq-Patch
|
A Fix Has Been Proposed For The Slower AMD Performance On Linux 5.11
|
Michael Larabel
|
With the in-development Linux 5.11 kernel there are many great features and improvements especially for AMD users with some new drivers and other pleasant enhancements. But as I outlined back on Christmas day: Linux 5.11 Is Regressing Hard For AMD Performance With Schedutil. Fortunately, a fix is now en route to the Linux 5.11 kernel for fixing that performance regression affecting AMD Zen 2/3 desktops and servers.
As outlined in that original article after bisecting the sizable performance regressions and in follow-up tests, AMD hardware performing slower on Linux 5.11 came down to the CPU frequency invariance support introduced this cycle and is utilized by the "Schedutil" CPU frequency scaling governor. With Schedutil often being the default for AMD systems on newer versions of the Linux kernel, this regression on Linux 5.11 compared to prior kernel releases has been unfortunate.
Giovanni Gherdovich of SUSE who worked out the AMD CPU frequency invariance support in cooperation with AMD engineers has now come up with a fix. From his analysis of the situation, "The problem happens on CPU-bound workloads spanning a large number of cores. In this case schedutil won't select the maximum P-State. Actually, it's likely that it will select the minimum one. A CPU-bound workload puts the machine in a state generally called "over-utilization": an increase in CPU speed doesn't result in an increase of capacity. The fraction of time tasks spend on CPU becomes constant regardless of clock frequency (the tasks eat whatever we throw at them), and the PELT invariant util goes up and down with the frequency (i.e. it's not invariant anymore)."
Giovanni was indeed able to reproduce the significant performance hits, such as with one test case, "See how the 128 threads case is almost 40% worse than baseline in v5.11-rc4."
The patch to fixing this Linux 5.11 Git performance regression is now out on the Linux kernel mailing list.
I'll be firing off a large round of benchmarks on multiple AMD systems tomorrow to confirm this regression is indeed sorted out and the AMD Ryzen / EPYC performance looking in good order for Linux 5.11 that will debut as stable next month. Will be back through with confirmation in the next day or two.UPDATE: Linux 5.11 Is Now Looking Great For AMD Zen 2 / Zen 3 Performance
| 6
| 1,760,719,409.702653
|
https://www.phoronix.com/news/AMD-CES-2021-Ryzen-5000-Mobile
|
AMD Announces Ryzen 5000 Series Mobile Processors, Previews EPYC Milan
|
Michael Larabel
|
Following Intel's product announcements yesterday, AMD CEO Dr. Lisa Su providing her virtual keynote this morning for the virtual CES 2021. Here are the highlights for how AMD is aiming to make 2021 even more exciting than their prior stellar year.
Highlights from the CES 2021 keynote are below. Sadly, some areas of the presentation are light on technical details, but looking forward to learning more and getting our hands on the new hardware for Linux testing when the time comes. Highlights include:
- The Ryzen 5000 mobile series processors, a.k.a. Zen 3 mobile, will appear in the weeks ahead. The new Ryzen mobile processors feature up to eight physical cores. Given the successes of the Ryzen 4000 series mobile processors and improving Linux support, very excited for them and intend to pickup a Ryzen 5000 series notebook for Linux testing/benchmarking.
- The Ryzen 7 5800U mobile processor is 7~44% faster for various tasks over the Intel Core i7 1185G7 Tiger Lake depending upon workload.
- The new Ryzen "HX" mobile processors are unlocked and allow for overclocking. The Ryzen 9 5900HX has 8c/16t and boosts up to 4.6GHz.
- Ryzen 5000 series laptops will start shipping in February.
- RDNA 2 graphics are coming to notebooks... Confirming more RDNA 2 desktop graphics are also coming to market in H1.
- AMD "previewed" 3rd Gen EPYC Milan. EPYC Milan in one 32-core 2P test was 68% faster than the comparable Intel Xeon Scalable. EPYC Milan is still on track for launching later in Q1.
Below is the virtual keynote in full.
| 46
| 1,760,719,411.063879
|
https://www.phoronix.com/news/AMD-More-Zen-3-LLVM
|
AMD Publishes More Zen 3 Compiler Support Patches For LLVM
|
Michael Larabel
|
AMD is back on track publishing more Zen 3 compiler support patches for the LLVM compiler stack.
Last month AMD engineers began posting their "znver3" support for LLVM that was largely replicating the existing Znver2 (Zen 2) target and exposing the newly-enabled instructions. Fortunately, now through the holidays, further work is coming for LLVM and presumably GCC in time.
Under review today is the Zen 3 scheduler descriptions for LLVM.
Merged last week meanwhile was this addition for exposing the TLBSYNC, INVLPGB, and SNP instructions found with Zen 3 processors.
The Zen 3 progress for LLVM can be tracked via reviews.llvm.org.
This work should ultimately make it into LLVM 12.0, which will be out around March -- though unfortunate it's taken until months after the Ryzen 5000 series premiered as the first Zen 3 processors until seeing optimized support in the compilers for those wishing to enjoy the performance benefits of -march=znver3. As the Zen 3 support continues to mature in both the GCC and LLVM Clang compilers there will be more optimization benchmarks on Phoronix but for now there are the Ryzen 9 5950X compiler benchmarks with various optimization settings from a few days ago.
| 2
| 1,760,719,411.08845
|
https://www.phoronix.com/news/AMD-EPYC-Linux-5.11-Varies
|
Linux 5.11 Is Looking Like A Wild And Bumpy Ride On AMD CPUs So Far
|
Michael Larabel
|
A few days ago I noted nice AMD EPYC performance improvements with PostgreSQL when running on Linux 5.11 compared to prior kernels. I've confirmed that for even more AMD EPYC servers now that the PostgreSQL uplift is there, but other workloads are unfortunately regressing for both Ryzen and EPYC. Here's the start of an exciting Christmas benchmarking adventure looking at this change with Linux 5.11...
The aforelinked article noticed some of the very healthy PostgreSQL database server performance improvements to find with Linux 5.11. Those prior servers tested were EPYC Rome 2P but I have also now confirmed with EPYC Rome 1P seeing gains too on completely different systems. Here is a look at the EPYC 7702 1P performance between Linux 5.10 and the current Linux 5.11 development code:
Some very significant and reproducible improvements for AMD EPYC 7002 series PostgreSQL performance on Linux 5.11 now across all the EPYC hardware I have tried so far. Some of the throughput and latency improvements are quite significant. So all is well? Yes and no. Unfortunately, with this ongoing Linux 5.11 testing I am also seeing some performance regressions hitting on Linux 5.11:
What the heck?! A number of the workloads regressing are mostly user-space applications not interacting with the kernel much or even much in the way of I/O... But that ended up jiving with my hypothesis in regards to the Linux 5.11 improvement for PostgreSQL. It's about the CPU frequency invariance support in Linux 5.11 for AMD Zen 2 and newer.
This comparison was done with a stock kernel each time and using the default CPUFreq scaling governor: Schedutil as has been the case since the change a few cycles ago from ondemand. With Linux 5.11 comes AMD frequency invariance support for Zen 2 CPUs and newer. The frequency invariance support being utilized by the scheduler utilization governor could explain the PostgreSQL improvement in making better decisions there but also worse decisions in some workloads such as those noted above.
So for my Christmas benchmarking adventure this year is looking closer at the Linux 5.11 CPUFreq performance on AMD EPYC relative to Linux 5.10. I now have benchmarks going for the schedutil, performance, and ondemand governors both for Linux 5.10 and 5.11 to see how the performance changes for confirming the theory that this change is what's responsible for this shift in AMD performance. Additionally, the Phoronix Test Suite is monitoring the peak CPU frequency, CPU temperature, and CPU power consumption on a per-test basis for all the tests being run across the different kernels and governors. These numbers should be complete within the next couple of days for seeing how Linux 5.11 ultimately is looking for AMD EPYC.
At least from my close monitoring of the changes each merge window, the AMD frequency invariance support is the main alteration in Linux 5.11 that comes to mind that could explain the performance changes especially for the workloads that are regressing.
While that large CPUFreq governor comparison with thermal/power/frequency impact is ongoing, here is a look at the EPYC 7702 1P results between Linux 5.10 and 5.11 for dozens of different workloads:
There is a fair amount of change with Linux 5.11. There were 148 benchmarks ran for that initial Linux 5.10 vs. 5.11 testing on the EPYC 7702 - see all the data points in full via this OpenBenchmarking.org result file. The follow-up much larger look at the Linux 5.11 performance is in the works for the next few days.
But what about on consumer Ryzen CPUs? I have Christmas benchmarking underway there too... On the Ryzen 9 5900X box here are some preliminary side-by-side numbers:
The Ryzen 9 5900X was seeing many workloads now running slower on Linux 5.11 than 5.10... Including PostgreSQL in that case should you be wanting to run a SOHO database server on consumer hardware or the like. Those initial Ryzen 9 5900X data points via this OpenBenchmarking.org result file.
Given such broad changes on the Ryzen front as well with Linux 5.11, CPUFreq behaving less than optimally seems like the leading contender...
Yes, Linux 5.11 doing poorly does seem to be an "AMD thing" at this point. I fired up some benchmarks on the Dell XPS Tiger Lake notebook with various tests and there Linux 5.11 is overall an improvement -- especially with I/O workloads given the IO_uring improvement to find and more.
Those numbers via this result file and while some differing workloads given the differing focus, no big scary drops seen there yet.
Long story short, lots of AMD Ryzen and EPYC benchmarks now running this Christmas in further investigating this matter... Those wanting to support this timely benchmark investigation can join Phoronix Premium this weekend or at the very least turn off your damn ad-blocker(s). Stay tuned.
UPDATE: See Linux 5.11 Is Regressing Hard For AMD Performance With Schedutil
| 5
| 1,760,719,412.503731
|
https://www.phoronix.com/news/AMD-2020-Highlights
|
AMD Wowed Linux Users In 2020 With Their Fantastic Zen 3 CPUs, Timely New Open-Source GPU Support
|
Michael Larabel
|
AMD this year not only delivered the very powerful Ryzen 5000 "Zen 3" desktop processors and initial Radeon RX 6000 "Big Navi" graphics cards but for exciting Linux users was the timely open-source support for the new GPUs well in advance (and also already preparing for some 2021 graphics products) as well as more timely support around Zen 3 thermal support and other additions. AMD's open-source timing is still improving although not yet optimal but all in it was a hell of a year for AMD Linux users.
AMD's hardware launches this year were fantastic with the Ryzen 5000 series continue to impress on Linux with their great performance and the Radeon RX 6000 series also delivering the best GPU performance for an open-source driver to date. With new enough open-source components, the new products were smooth sailing from day one -- assuming you were able to procure the hardware. But leaving room for improvement looking ahead is they could be even more punctual in their kernel enablement work to ease users especially enterprise folks wanting to stick to LTS kernel versions. There are also other matters like the Zen 3 compiler support patches only appearing after launch and to date still lacking any updated costs table, scheduler model, or other optimizations to really cater to the updated processors. Intel still leads when it comes to having their open-source support from compiler toolchains and kernel support out well ahead of product launches but AMD has been on an improving trajectory.
Looking ahead to 2021, I am damn excited for the next-generation AND Ryzen mobile processors, next-gen Threadripper, and certainly for EPYC 7003 "Milan"... Stay tuned for benchmarks on both when the time comes. Plus plenty of other interesting Zen 3 and RDNA 2 benchmarks still coming with the evolving open-source/Linux state.
The most viewed AMD open-source/Linux news articles for 2020 on Phoronix included:
Linus Torvalds Switches To AMD Ryzen Threadripper After 15 Years Of Intel Systems An interesting anecdote shared in today's Linux 5.7-rc7 announcement is word that Linux and Git creator Linus Torvalds switched his main rig over to an AMD Ryzen Threadripper. The AMD Radeon Graphics Driver Makes Up Roughly 10.5% Of The Linux Kernel Given the impending release of Linux 5.9, I was having some fun with cloc today looking at the current lines of code count for this near-final Linux 5.9 kernel state. AMD vs. Intel Contributions To The Linux Kernel Over The Past Decade Driven by curiosity sake, here is a look at how the total number of AMD and Intel developers contributed to the upstream Linux kernel during the 2010s as well as the total number of commits each year from the respective hardware vendors. Acer Is Launching In Germany What Could Be A Great AMD Ryzen 5 4500U Linux Laptop For those that have been looking out for an AMD Linux laptop powered by a Ryzen 4000 series processor, Acer is set to launch a new laptop at least in Germany that could be quite appealing to Linux users. I've Been Running The AMD Ryzen 7 4700U + Ubuntu 20.04 As My Main System For about one and a half months now I have been using the AMD Ryzen 7 4700U as my main laptop paired with Ubuntu 20.04 LTS. It's been working out very well for not even being the top-of-the-line AMD Renoir SKU. Here is some additional commentary for those thinking about one of the new AMD laptops with Linux use. Benchmarks Of Arch Linux's Zen Kernel Flavor Following the recent Linux kernel tests of Liquorix and other scheduler discussions (and more), some requests from premium supporters rolled in for seeing the performance of Arch Linux's Zen kernel package against the generic kernel. Here are those benchmark results. A Quick Look At The Blender 2.82 Performance On Intel + AMD CPUs With Blender 2.82 having released on Friday, this weekend we've begun our benchmarking of this new Blender release as the leading open-source 3D modeling solution currently available. Here are some preliminary v2.81 vs. v2.82 figures on different higher-end Intel and AMD processors. Ryzen CPUs On Linux Finally See CCD Temperatures, Current + Voltage Reporting One of the few frustrations with the AMD Ryzen CPU support on Linux to date has been besides the often delayed support for CPU temperature reporting has been the mainline kernel not supporting voltage readings and other extra sensors. But that is finally changing with the "k10temp" driver being extended to include current and voltage reporting plus CCD temperature reporting on Zen 2 processors. Rust-Written Redox OS Booting The 128-Thread AMD Ryzen Threadripper 3990X The Rust language focused Redox OS open-source operating system is now able to boot the AMD Ryzen Threadripper 3990X 64-core/128-thread processor and run with full multi-threading capabilities. ASUS TUF Laptops With Ryzen Are Now Patched To Stop Overheating On Linux The AMD Ryzen Linux laptop experience continues improving albeit quite tardy on some elements of the support. In addition to the AMD Sensor Fusion Hub driver finally being released and current/voltage reporting for Zen CPUs on Linux, another step forward in Ryzen mobile support is a fix for ASUS TUF laptops with these processors. LLVM/Clang 10.0 Adds AMD Zen 2 Scheduler Model For Optimized Code Generation It's too bad that it has taken so many months after AMD Zen 2 based Ryzen and EPYC processors began shipping to see this compiler support in place, but the good news now is that for the upcoming release of LLVM 10.0 is now the Zen 2 scheduler model being added to the "znver2" target. AMD Begins Providing PowerPC Builds Of Their "AOMP" GPU Compiler AOMP is the AMD GPU compiler for OpenMP and HIP support on GPUs as part of Radeon Open eCosystem 3.0 (ROCm 3.0). Now they have begun providing PowerPC 64-bit LE builds of AOMP as part of allowing Radeon GPU compute to happen on POWER9 systems. A Slew Of ACO Optimizations For The Radeon Vulkan Driver Landed In Mesa 20.0 The Valve-backed ACO compiler back-end that is optionally used by the RADV Radeon Vulkan driver has continued growing in popularity with Linux gamers and also has continued maturing a lot for Mesa 20.0 that is due out later this quarter. RenderDoc 1.6 Released, NVIDIA + AMD + Intel All Primed For Vulkan 1.2 This morning's release of Vulkan 1.2 is off to a great start. Mesa 20.0 Released With Big Improvements For Intel, AMD Radeon Vulkan/OpenGL Mesa 20.0 is now released as the first quarter 2020 update to the Mesa 3D open-source graphics driver stack. AMD Ryzen 4000 Mobile Series "Renoir" Graphics No Longer Experimental With Linux 5.5 While the Linux 5.5 kernel is expected to be released as soon as this Sunday, a last minute change to the AMDGPU DRM driver makes the Renoir graphics no longer treated as experimental. With that, there is open-source support out-of-the-box rather than being hidden behind a kernel module flag. System76 May Offer AMD Ryzen Laptops When They Begin Their Own Manufacturing System76 is preparing to begin shipping their new Lemur Pro laptop in early April. This will be their most open laptop yet albeit still based on Intel. But it looks like when they move on with their ambitious plans to begin manufacturing their own devices, we may finally see a System76 AMD-powered laptop. FFmpeg 4.3 Released With AMD AMF Encoding, Vulkan Support, AV1 Encode FFmpeg 4.3 is out as the latest version of this key open-source multimedia library. FFmpeg 4.3 is quite a big release. AMD Is Hiring Another Lead Linux Kernel Developer To Work On Their Graphics Driver Should you be experienced in upstream Linux kernel development, AMD is hiring a lead Linux kernel developer. Linux k10temp Driver For AMD CPUs Updated To Better Handle Power/Temp Analysis As we have been eagerly talking about for the past week, the Linux kernel's k10temp driver was updated for better AMD CPU CCD temperatures and voltage/current reporting. Those improvements have been quickly evolving thanks to the work of the open-source community with AMD still sadly holding the datasheets concerning the power/temperature registers close to their vest. A new version of k10temp was sent out on Wednesday.
| 28
| 1,760,719,412.61976
|
https://www.phoronix.com/news/Linux-5.11-Drops-k10temp-V-C
|
Linux 5.11 Drops AMD Zen Voltage/Current Reporting Over Lack Of Documentation
|
Michael Larabel
|
The Linux hardware monitoring "k10temp" driver is dropping support for reporting CPU voltage and current information for AMD Zen-based processors over lack of documentation for being able to properly support the functionality.
Earlier in 2020 this long-standing AMD hwmon temperature driver added support for core/SoC current and voltage reporting with Zen processors based on the work by the community and some best assumptions around the appropriate registers. But now that support is being dropped for lack of accuracy in some configurations and the possibility it might even damage the hardware.
Last week was the main hwmon pull request for the Linux 5.11 cycle while sent in today was a secondary update with the sole change being the removal of this k10temp current/voltage reporting.
The support is being removed as it "turns out that [it] was not worth the trouble."
The actual Git commit removing the support from k10temp goes on to explain:
Voltages and current are reported by Zen CPUs. However, the means to do so is undocumented, changes from CPU to CPU, and the raw data is not calibrated. Calibration information is available, but again not documented. This results in less than perfect user experience, up to concerns that loading the driver might possibly damage the hardware (by reporting out-of range voltages). Effectively support for reporting voltages and current is not maintainable. Drop it.
For those just wanting energy metrics, the "AMD_Energy" driver is available with the mainline kernel as well but currently that just exposes energy details on EPYC CPUs after the initial Ryzen support was eliminated there. Hopefully AMD will publish the necessary documentation in 2021 to allow for this k10temp support to be properly restored. At least and arguably more pressing is the temperature monitoring support continues to work for Zen through Zen 3 (Zen 3 with Linux 5.10+) with this k10temp driver.
| 59
| 1,760,719,414.131277
|
https://www.phoronix.com/news/AMD-S2idle-ACPI-Linux-5.11
|
AMD S2idle Support For Linux Getting Squared Away
|
Michael Larabel
|
Just in time for the upcoming AMD Ryzen 5000 series mobile processors, it's looking like the S2idle support is finally coming together on Linux for increased power savings.
As previously covered, Linux 5.11 picked up an AMD SoC PMC driver that is responsible for handling S2idle transactions driven by the platform firmware on the SMU. Sent in on Tuesday as part of a secondary set of ACPI changes for Linux 5.11 is more AMD S2idle enablement work.
This latest code is about providing initial support for S2idle (suspend to idle S0ix) while largely relying upon the existing kernel code written by Intel. The bit needed wrangling with this code was the BIOS implementation for ACPI methods like _DSM (device specific methods) not being standardized and thus needing vendor-specific handling. So with this code now in the kernel as of yesterday, the Linux S2idle code can correctly use the proper _DSM methods on Linux.
Initial support for S2Idle based on the Intel implementation does not work for AMD as the BIOS implementation for ACPI methods like the _DSM are not standardized.
So, the way in which the UUID's were parsed and the ACPI packages were retrieved out of the ACPI objects are not the same between Intel and AMD.
This patch adds AMD support for S2Idle to parse the UUID, evaluate the _DSM methods, preparing the Idle constraint list etc. This ultimately is about getting the SoC into its lowest power platform idle state. The patch seems to imply that S2idle should now be working for supported AMD platforms -- well, sans the usual quirky BIOS implementations and other power management headaches often seen among consumer laptops. Will be digging further into the AMD S2idle handling over the holidays but in any case this latest code for it hit the Linux mainline tree yesterday following this ACPI pull request and looking like better AMD Linux power savings could be on the horizon.
| 15
| 1,760,719,414.187438
|
https://www.phoronix.com/news/AMD-EPYC-PostgreSQL-Linux-5.11
|
AMD EPYC Seeing Nice Performance Improvements With PostgreSQL On Linux 5.11
|
Michael Larabel
|
For those running PostgreSQL database servers (and potentially similar workloads) on AMD EPYC servers, Linux 5.11 is bringing a very nice Christmas gift in the form of better performance for at least some 2P server configurations.
We are just half-way through the Linux 5.11 merge window with new code continuing to land, but already I've been running Linux 5.11 Git benchmarks on a number of systems in looking for any prominent performance improvements as a result of the new feature code or any signs of performance regressions... One area where I am seeing definite improvement with Linux 5.11 is on the PostgreSQL database server performance for at least AMD EPYC 2P servers.
I am still running more tests and in varying configurations, but for cases like EPYC 7F62 2P and EPYC 7742 2P with speedy NVMe SSD drives, I have been seeing nice uplift from PostgreSQL on Linux 5.11. I haven't yet bisected the cause to see if it's some improvement with the block layer or underlying file-system with Linux 5.11 or quite possibly coming as a result of scheduler improvements and the other changes that cycle so far like the Schedutil work for AMD frequency invariance as one possibility that comes to mind.
In any case, there is some nice PostgreSQL uplift being seen on Linux 5.11 from my preliminary testing. Here are some of the numbers when using an AMD EPYC 7F72 2P setup with a WD_BLACK SN850 NVMe SSD:
Seeing better throughput...
And better latency with AMD EPYC PostgreSQL on Linux 5.11.
I didn't see similar uplift out of MySQL/MariaDB but those numbers were flat. Outside of database workloads, the Linux 5.10 vs. 5.11 numbers were largely flat but I am ramping up testing as the new feature material settles down. I'm continuing to explore the AMD EPYC (and Intel Xeon) performance on Linux 5.11 especially once the merge window has passed. I'll be running many Linux 5.11 benchmarks through the holidays, so stay tuned.
| 5
| 1,760,719,415.586045
|
https://www.phoronix.com/news/AMD-Freq-Invariance-Linux-5.11
|
AMD Frequency Invariance Support Comes With Linux 5.11
|
Michael Larabel
|
The previously reported on work for frequency invariance calculations for AMD CPUs with a focus on the AMD EPYC 7002 series has been merged for Linux 5.11 as part of the "sched/core" material.
Following all of the Intel Linux kernel work in recent months around frequency invariance handling for more accurate load tracking and making more accurate frequency scaling decisions, the initial AMD implementation is here with Linux 5.11 as part of the core scheduler updates. In basic terms, the frequency invariance calculation is for addressing the issue of tasks appearing larger if the CPU is running slower so the frequency invariance takes into account the current frequency relative to the maximum possible frequency.With the AMD implementation the frequency invariance support is relying on a numerical performance state value (0 to 255) as opposed to looking at the current CPU frequency relative to the maximum frequency. ACPI's Collaborative Processor Performance Control interface is queried for that performance state information, but not to be confused with AMD's prior work on their own Zen 2 CPPC CPUFreq driver that seemingly has been dropped for the time being. Due to the use of ACPI CPPC information, this frequency invariance calculation is limited to Zen 2 (and newer) AMD CPUs.
The main user of this support for now is with the CPUFreq "Schedutil" governor that allows making use of the frequency-invariant information for smarter CPU frequency scaling decisions. I'll be running such benchmarks soon once the Linux 5.11 merge window settles down.
This pull was mainlined with the AMD CPU frequency invariance calculation support work by AMD and SUSE. The Linux 5.11 core scheduler updates also include topology and NUMA improvements, better robustness for the global CPU priority tracking, and other mostly small enhancements this cycle.
| 8
| 1,760,719,415.913481
|
https://www.phoronix.com/news/AMD-SB-TSI-Driver-Linux-5.11
|
AMD SB-TSI Sensor Driver Set To Appear With Linux 5.11
|
Michael Larabel
|
There are a lot of changes coming with Linux 5.11 and on the AMD side includes the likes of VanGogh and Dimgrey Cavefish graphics support, AMD EPYC Zen 3 support in the AMD_Energy driver, AMD RAPL Zen1/Zen2/Zen3 PowerCap support, an AMD SoC PMC driver, and the AMD Sensor Fusion Hub driver for Ryzen laptops is finally being mainlined... Another new addition was queued up this weekend by way of hwmon-next and that's the AMD SB-TSI sensor driver.
Going back to early in 2020 we reported on Google engineers working on this AMD SB-TSI code for the Sideband Temperature Sensor Interface. This is about reading the AMD SoC temperature connected to a BMC.
SB-TSI has been part of the public AMD Zen documentation while thanks to Google in cooperation with AMD there is this SB-TSI temperature sensor support coming.
SB Temperature Sensor Interface (SB-TSI) is an SMBus compatible interface that reports AMD SoC's Ttcl (normalized temperature), and resembles a typical 8-pin remote temperature sensor's I2C interface to BMC.
This commit adds basic support using this interface to read CPU temperature, and read/write high/low CPU temp thresholds. The sbtsi_temp driver is added as part of the hardware monitoring "hwmon" code set to go into the kernel once the merge window opens for Linux 5.11, which will be as soon as tomorrow if no additional Linux 5.10 release candidate is warranted later today. This documentation covers more details on this sideband temperature monitoring interface for those interested.
The Linux kernel already supports the AMD Zen CPU temperature monitoring via the k10temp driver. This "sbtsi_temp" driver work will likely be useful for the likes of OpenBMC for temperature monitoring. It should come as little surprise at this stage and given our past articles, but AMD has been working to improve their OpenBMC support and acknowledge the growing customer interest in open-source firmware stacks.
| 2
| 1,760,719,417.06984
|
https://www.phoronix.com/news/AMD-HIP-CPU-Implementation
|
AMD Provides A CPU-Based HIP Implementation For When Lacking A GPU
|
Michael Larabel
|
AMD's HIP C++ Runtime API / Kernel Language for allowing portable, single-source applications on AMD and NVIDIA GPUs can now run on CPUs too.
For the past several years AMD has been working on HIP for single-source C++ programming that can work on NVIDIA CUDA-enabled GPUs and AMD Radeon graphics. Their "HIPIFY" tool allows automatically converting CUDA code to HIP. To date HIP has just been about GPU programming but now it's becoming a heterogeneous API at the same time Intel is now promoting their oneAPI alternative.
With HIP-CPU, HIP code can be easily run on CPUs now with this work-in-progress implementation as it strives to become a complete HIP implementation where possible. Granted, cases like inline NVIDIA PTX / Radeon graphics assembly aren't portable.
This AMD HIP-CPU support isn't a run-time implementation but consists of just a header file for building out the support at compile time. HIP-CPU is implemented as a generic C++17 header file that can execute across CPUs/architectures. This header-only C++ project should work across operating systems and with all major compilers supporting C++17 or later -- HIP-CPU makes use of the C++ standard parallel algorithms. Because it is header-only, it also does not require AMD's ROCm stack.
Downloads and more details on the new AMD HIP-CPU project via GitHub.
| 6
| 1,760,719,417.363544
|
https://www.phoronix.com/news/Mesa-21.0-AMD-Smart-Memory
|
Mesa 21.0 Begins Landing Optimizations For AMD Smart Access Memory
|
Michael Larabel
|
While AMD Smart Access Memory has already been supported under Linux for some time with its resizable BAR functionality, only now with all the excitement around the feature being promoted with the Ryzen 5000 series and Radeon RX 6000 series hardware is the Mesa driver code beginning to see some optimizations for it.
Lead RadeonSI developer Marek Olšák of AMD has merged a set of optimizations for AMD Smart Access Memory that will come in next quarter's Mesa 21.0 release. The code that landed today include a helper for determining Smart Access Memory / all vRAM visible, only force the staging uploads for vRAM when Smart Access Memory isn't enabled, and only use staging for linear textures when the feature is disabled. For cases when Smart Access Memory is enabled, vRAM is now used for command buffers, the pipe usage stream is mapped to vRAM, and the uploading code has been unified and going straight to vRAM with this feature enabled. More details within this merge request.
These Smart Access Memory optimizations are focused on the RadeonSI Gallium3D code. The code is queued in Mesa 21.0-devel while the stable release with this functionality should be out in March, hopefully before then we'll see more optimizations.
For those wondering how to check if Smart Access Memory is enabled for your Linux system, see the details in this forum thread.
I was in the process already of running some AMD Smart Access Memory Linux benchmarks while I'll be restarting them now with this latest Mesa Git.
| 10
| 1,760,719,418.641624
|
https://www.phoronix.com/news/Ryzen-5000-Series-One-Month
|
The AMD Ryzen 5000 Series Continue To Impress On Linux
|
Michael Larabel
|
It's been just over one month since AMD launched the Ryzen 5000 series as the first processors part of the Zen 3 family. The Linux performance continues to be terrific with the Ryzen 5600X / 5800X / 5900X / 5950X parts in our continued benchmarking.
Some of the benchmarks that are in the works for publishing in the coming days and weeks include looking at Zen 3 compiler tuning on GCC and LLVM Clang now that patches are beginning to materialize, Radeon RX 6800 series performance with Ryzen 9 on Windows vs. Linux, Windows 10 vs. Ubuntu 20.04/20.10 performance on the Ryzen 9 5900X, Windows Subsystem for Linux 2 performance, an AMD Smart Memory Access comparison on Linux, and also checking out how well the BSDs like FreeBSD and DragonFlyBSD perform with the Ryzen 5000 series. Those are just the tests currently in the works but more will surely be coming as well - throughout all the ongoing tests, the performance continues to impress.
Outside of the performance, the only Linux issues have been the ones I've mentioned since launch day: the unfortunate matter of the compiler support coming after launch and thus in turn taking months before being found in stable distributions, the AMD Zen 3 temperature support only appearing with Linux 5.10 that will debut as stable this weekend, and no AMD_Energy driver support for the Ryzen 5000 series processors at this point. But aside from those issues, which admittedly don't affect many users directly, it's been a very pleasant support experience paired with modern Linux distributions.
From my ongoing testing as well as from other Linux-based reviewers and the community that gets their hands on these new processors, more benchmarks continue to be uploaded via the Phoronix Test Suite to OpenBenchmarking.org.
On the individual search pages for the Ryzen 9 5950X, Ryzen 9 5900X, Ryzen 7 5800X, and Ryzen 5 5600X continue to be the individual percentile rankings for these processors in the different benchmarks out of the hundreds available via the Phoronix Test Suite. For many tests these Zen 3 processors are up around the 99th percentile or better out of all the benchmarks found on OpenBenchmarking.org.
You can also enjoy a Ryzen 5950X vs. 5900X vs. 5800X vs. 5600X performance overview. As of writing there are 236 unique test cases with enough statistically significant data displayed in common for these four Ryzen 5000 series processors. From that page you can also add additional processors to the dynamic comparison for further insight -- some 350+ other processors but obviously the number of test cases in common and with enough statistical significance to display any composite number will vary based on processor. From that page you can also punch in your own CPU prices for custom performance-per-dollar metrics.
Of course via the Phoronix Test Suite you can also fire up your own benchmarks to see how your system(s) compare to these Ryzen 5000 series processors. Enjoy and stay tuned for more Linux benchmarks of these parts.
| 13
| 1,760,719,418.819006
|
https://www.phoronix.com/news/AMD-Zen3-znver3-In-GCC-11
|
Initial AMD Zen 3 Support Successfully Lands In GCC 11
|
Michael Larabel
|
A few days ago AMD finally sent out the initial AMD Zen 3 "znver3" support to the GCC compiler with the LLVM Clang support to follow. That initial "-march=znver3" targeting support has now been merged for GCC 11.
While past the feature development stage of GCC 11, the znver3 patch was permitted to land with new targets / hardware support allowed to land still as it doesn't risk regressing the existing compiler support.
As outlined in the prior article though, this patch just flips on the new CPU instructions supported by Zen 3 compared to Zen 2. But there isn't yet any updated costs table / scheduler model to really optimize the generated for Zen 3 processors. AMD is still working through that portion of the enablement but looks like that might not be ready for either several weeks or months. That "optimized" support might not even be available in time for the GCC 11.1 stable release next March~April while at least the basic "-march=znver3" is ready and out.
The merge happened this morning so the daily builds and development snapshots to follow have this support.
I have already been running some benchmarks on the AMD Zen 3 GCC patch and should have out those initial benchmarks in the coming days.
| 9
| 1,760,719,420.159745
|
https://www.phoronix.com/news/AMD-5900X-Zen-3-Spectre-Windows
|
A Quick Look At The Spectre Mitigation State For AMD Zen 3 On Windows 10
|
Michael Larabel
|
Earlier this week I looked at the Spectre mitigation performance impact on AMD Zen+ / Zen 2 / Zen 3 processors given the recent launch of the Ryzen 5000 series and those newest CPUs still requiring some mitigation handling. Questions were raised about the Spectre mitigation handling on Windows, so I ran some quick tests there as I happened to have a Windows 10 install on the Ryzen 9 5900X test box at the moment for some unrelated Windows vs. Linux gaming.
The prior article looking at the Spectre mitigation performance impact on Linux found the Zen 3 impact to actually be greater than Zen 2. As explained in more detail in that prior article, Zen 3 now relies on "always-on" STIBP (Single Threaded Indirect Branch Predictors) as part of the Spectre V2 handling. Always-on STIBP carries more overhead than the conditional STIBP previously used. Since that former article I was able to confirm with AMD that always-on STIBP is indeed the intended mode of operation for Zen 3 and will remain that way. But in any case, it's not that bad since all of the AMD Zen 3 benchmarks across the various websites you've likely been looking at since launch have been with the default mitigations in place. This is just about the performance if opting to disable the default mitigations. So if disabling all the mitigation handling you may be able to squeeze out slightly better performance than the default/out-of-the-box, but at least it's not a situation where the increased mitigations / performance overhead is coming after the fact as an unexpected and costly surprise to users.
As I've been working on some Windows 10 + WSL vs. Linux benchmarks on the Ryzen 9 5900X, I ran some quick Spectre-focused tests there.
While on Linux disabling all the CPU security mitigations can be as easy as booting the kernel with the "mitigations=off" option, on Windows it's less so. InSpectre is the most straight-forward way of disabling Meltdown/Spectre mitigations under Windows. When disabling the Spectre mitigation on AMD hardware it will report it as such, but there is an important distinction... It doesn't disable the AMD STIBP handling, so that is actually still active.
The MDS tool confirmed that even when disabling the Spectre mitigations for AMD Zen 3, STIBP is left in an enabled state. IBRS (Indirect Branch Restricted Speculation) was also left enabled but from our testing the always-on STIBP is the source of the main performance overhead on Zen 3.
So when quickly firing up some benchmarks on the Ryzen 9 5900X of the out-of-the-box versus InSpectre Spectre disabled state for AMD Zen 3 where IBRS/STIBP is left enabled:
The performance is basically unchanged to little surprise since Zen 3 STIBP is left enabled (all the benchmark details in full via this OpenBenchmarking.org result file). Disabling AMD STIBP under Windows still appears to rely on it being exposed as an option by the motherboard/system vendor, to which most don't expose STIBP is a toggleable option. So long story short, InSpectre doesn't end up being worthwhile on the new AMD Zen 3 CPUs with the tests I've conducted since STIBP doesn't end up getting forced off. But as this has been the default behavior from the get-go for these fantastic new Ryzen 5000 series processors, the benchmark numbers you have been seeing since launch are the default mitigated numbers while delivering the incredible generational uplift from Zen 2 -- this is just about trying to squeezing some extra frosting on the cake.
| 4
| 1,760,719,421.362305
|
https://www.phoronix.com/news/Linux-5.11-AMD-Energy-Zen-3
|
EPYC Zen 3 CPU Support Coming To Linux's AMD_Energy Driver
|
Michael Larabel
|
In addition to AMD Zen 1/2/3 PowerCap RAPL support coming for the Linux 5.11 kernel, the hwmon-next Git branch has also queued initial support for Zen 3 processors within the AMD_Energy driver.
The AMD_Energy driver was introduced earlier this year and merged for Linux 5.8 for easily exposing AMD CPU energy metrics -- albeit the list of supported CPU models was later restricted to EPYC CPUs.
The AMD_Energy driver to now has not supported any Zen 3 processors but it's fairly straight-forward to do so. With the code set to go into Linux 5.11, this one line addition adds support for Family 19h "Zen 3", but currently limited to just model 01h. Thus like the current support, at least for now it looks like the AMD_Energy support is still being limited to EPYC processors (the Ryzen 5000 series parts launched so far have been Family 19h, Model 33).
Hopefully the AMD_Energy driver will properly support AMD Ryzen desktop/mobile processors in due course as aside from that limitation this driver has been working out quite well over the past number of months with my EPYC testing.
This hwmon-next work along with the aforementioned PowerCap RAPL support should be merged to mainline as soon as the Linux 5.11 merge window opens in a week or two. Linux 5.11 stable is expected in February.
| 3
| 1,760,719,421.7173
|
https://www.phoronix.com/news/AMD-Zen-3-GCC-Plus-AOCC-2.3
|
AMD Sends Out Zen 3 Compiler Support For GCC + AOCC 2.3 Compiler Released
|
Michael Larabel
|
Following last month's release of the Ryzen 5000 "Zen 3" processors, AMD has now begun publishing their official compiler support for this extremely compelling processor family.
For as extremely great as Zen 3 is, it's the belated compiler support as one of the few critiques we've had -- normally on the Intel side they are often plumbing their compiler targets and new instruction set extension support a year or more ahead of CPU launches (e.g. the most recent example back in July Intel added Alder Lake and Sapphire Rapids to GCC), and that's for when those processors are shipping on schedule. Having the compiler support out well ahead of the launches ensure the support is worked into stable compiler releases by the time the CPUs ship and ideally already used as the default compiler version in major Linux distribution releases. Intel generally remains spot-on in that regard while AMD has been much tighter -- or in the case of Zen 3, basically one month after launch.
Out this morning is the initial patch adding support for Zen 3 CPUs to the GNU Compiler Collection with the new "-march=znver3" option. The patch though is admittedly the "very basic enablement" and doesn't yet have any Zen 3 changes around the instruction cost tables, tuning, and scheduler differences with the architecture changes of Zen 3. For now all that specialized tuning is kept the same as Zen 2 -- which in the case of the scheduling at least, is still the same as Zen 1. At a later date that more precise tuning will come to the znver3 code for GCC.
This basic Znver3 enablement is looking to land in trunk now for GCC 11, which should debut in March~April as the annual update to the GNU Compiler Collection. That basic enablement patch will likely be allowed to land even though GCC 11 recently passed the end of feature development but given the timing likely means that any scheduler updates and other fine tuning won't land in time. That polished Znver3 support then won't come until GCC 12 or sought as a back-port later on for a GCC 11 point release. But due to the timing of these projects and the various launches, basically means a released/stable GCC compiler won't take full advantage of the Zen 3 microarchitecture until well into 2021. Hopefully AMD will be able to work on better timing alignment moving forward.
For now the GCC znver3 support basically takes the Zen 2 state while adding in the instructions for PKU, VPCLMULQDQ, and VAES. It's basically quite similar to the basic Znver3 support for LLVM that was done while waiting on AMD. I've heard AMD will be putting out their official Znver3 LLVM/Clang patch out soon.
Today they also released AOCC 2.3 as their LLVM Clang based compiler with various AMD Zen patches added. AOCC 2.3 remains tuned for Zen 2 and based on LLVM 11.0 upstream but does contain their initial Zen 3 support.
I'll be running some benchmarks shortly to see any impact from these Znver3 compiler updates on common workloads.
| 11
| 1,760,719,422.772401
|
https://www.phoronix.com/news/AMD-OpenBMC-2020-Progress
|
AMD Is Making Progress On Open-Source Firmware - Initially With OpenBMC
|
Michael Larabel
|
While we are still waiting to see what AMD might do for returning to open-source AGESA or better supporting Coreboot and the like, they are making some inroads with open-source firmware support -- beyond the context of Chromebooks where they continue to engage due to Google's engineering requirements. AMD is working to "align with the industry direction of open-source firmware stacks" with their initial focus being on open-source OpenBMC firmware support for their server platforms.
AMD's Supreeth Venkatesh spoke at this week's virtual Open-Source Firmware Conference (OSFC 2020) on the work they are pursuing around OpenBMC. It was acknowledged that this work is being done due to the industry direction these days of preferring open-source firmware stacks (and being "a good open-source citizen") but stopped short of outlining any other open-source firmware plans at this time outside of OpenBMC. Given the customer interest and industry trends they have been working to support open-source OpenBMC support on the AMD server reference platforms. From the presentation, it looks like Twitter's engineering team has been involved with the bring-up and among the interested users but surely other key industry players are also taking note.
OpenBMC for those unaware is the effort around providing a Linux-based open-source firmware stack for baseboard management controllers. OpenBMC is backed by companies like IBM, Intel, Google, and Facebook. OpenBMC is used by the likes of Raptor Computing System's Talos II server and Blackbird desktop as part of their fully-open hardware approach.
Currently they have OpenBMC working on their AMD EPYC "Ethanol" reference platform while support for the newer EPYC "Daytona" platform is in the works. They are hoping soon to upstream the DeviceTree support in the Linux kernel around the Daytona platform that was their reference platform initially for the EPYC 7002 "Rome" series.
Features they are working on moving forward include IPMI 2.0 compliance, Redfish compliance, handling remote debugging with OpenBMC, an APML SB-RMI kernel driver, CPU power management, and RAS functionality.
AMD's open-source OpenBMC changes are being staged via this GitHub branch. More details within the OSFC slide deck. Great to see more AMD engagement around open-source firmware support and will be very interesting to see where this journey leads in 2021.
Also worth noting as well is the independent work 3mdeb is pursuing on Trenchboot DRTM (Dynamic Root of Trust for Measurement) on AMD platforms. If AMD Secure Startup type functionality is of interest to you, see this slide deck for those details on working out open-source AMD Secure Startup by the firm 3mdeb.
| 22
| 1,760,719,423.609065
|
https://www.phoronix.com/news/AMD-EPYC-Rome-Freq-Invariance
|
AMD+SUSE Tackling Frequency Invariance For AMD EPYC 7002 CPUs
|
Michael Larabel
|
Thanks to work by AMD and SUSE engineers, the Linux kernel could soon be seeing frequency invariance support for EPYC 7002 "Rome" processors for yielding greater performance and power efficiency.
Over the past year we have seen a lot of Linux kernel work for dealing with frequency invariance but to now that on the x86 side has been focused on Intel Xeon processors. Now through the cooperation of AMD with patches led by SUSE, frequency invariance is being worked on for the EPYC 7002 "Rome" processors.
Frequency invariance ultimately is about more accurate load tracking and being able to make better frequency scaling decisions. All the fine details around frequency invariance can be found outlined via the Kernel.org documentation. Fundamentally it's regarding the issue of tasks appearing larger if the CPU is running slower so the frequency invariance takes into account the current frequency relative to the maximum possible frequency (or with the way these AMD patches are implemented, relative to the maximum possible performance state).
With the pending AMD Rome implementation, ACPI is queried for figuring out the maximum boost performance state of the system via the CPPC (Collaborative Processor Performance Control) support that is new to Zen 2 and thus why Zen 1 CPUs are not supported by this code. The CPPC performance value is used as opposed to the boost clock frequency itself for lack of being able to properly determine that on the current AMD processors but in relative terms still works out all the same.
The AMD EPYC frequency invariance patches for the Linux kernel remain under review and discussion on the kernel mailing list.
No performance figures were provided by the developers on these AMD patches, so it will be fun to benchmark it when the time comes. At least on the Intel side the frequency invariance support with Xeon processors has allowed for better performance and efficiency. It will be interesting to see how this AMD frequency invariance support pans out with the Linux Schedutil governor that makes use of scheduler utilization data for making its frequency scaling decisions. The patches also note it will serve as a baseline for comparison should AMD ever get back to working on their previously proposed AMD CPPC CPUFreq driver that hasn't seen any new public activity since the patches were first volleyed back in July 2019 after Zen 2 was first announced with ACPI CPPC support.
| 2
| 1,760,719,424.255396
|
https://www.phoronix.com/news/AMD-ROCm-Xilinx
|
AMD ROCm Open-Source Stack Coming To Xilinx FPGAs
|
Michael Larabel
|
Now here is some darn interesting software news from SC20... AMD, which is in the process of acquiring Xilinx, is bringing the Radeon Open eCosystem "ROCm" stack to Xilinx hardware.
Xilinx and AMD are offering a technology demonstration of the AMD ROCm open-source stack atop the Xilinx Alveo accelerator cards. AMD and Xilinx are working to fully support the FPGAs within the ROCm platform and integration between AMD Instinct GPUs and Alveo accelerators for compute, networking, and storage solutions. Ultimately they aim to provide a fully integrated ROCm runtime to span from AMD CPUs and GPUs through FPGAs.
With AMD pulling Xilinx into the organization and offering this unified ROCm stack, they will now be squarely facing off with Intel's oneAPI software stack.
The ROCm integration is still in the works while their technology demonstration was announced via Xilinx.com.
| 22
| 1,760,719,425.039528
|
https://www.phoronix.com/news/AMD-Zen-PowerCap-RAPL-5.11
|
AMD Zen1/Zen2/Zen3 PowerCap RAPL Support Queued For Linux 5.11
|
Michael Larabel
|
The work reported on back in October for RAPL PowerCap patches for AMD Zen CPUs from Zen 1 through Zen 3 are set to arrive with Linux 5.11 in early 2021.
This work led by Google engineers allow the AMD Family 17h/19h RAPL (Runtime Average Power Limiting) support within the Linux kernel's power-capping framework in the PowerCap driver.
Those unfamiliar with Linux's power-capping framework can see the kernel.org documentation. To user-space this provides a consistent interface for the power information now for both Intel/AMD processors.
Up to now the PowerCap driver has been Intel-only but extended by Google to support AMD CPUs in the form of Zen 1 and newer. This patch has the Zen 1 / Zen 2 support while a follow-up patch adds the support for Family 19h (Zen 3) given the MSRs are identical to the earlier Zen CPUs.
These patches have been queued into the Linux power management "-next" branch making them Linux 5.11 material for when that merge window opens in December.
Along related lines and for those not aware, Linux 5.8 added the "AMD_Energy" driver for exposing CPU energy information to user-space via the hardware monitoring (HWMON) interfaces. Though at the moment that driver does not yet support Zen 3 processors.
| 9
| 1,760,719,425.651331
|
https://www.phoronix.com/news/AMD-Ryzen-5000-Zen-3-Linux-Perf
|
Many More AMD Ryzen 5000 Series "Zen 3" Linux Benchmarks
|
Michael Larabel
|
Over the past week we have published our Linux performance reviews of the Ryzen 5 5600X, Ryzen 7 5800X, and Ryzen 9 5900X + 5950X "Zen 3" processors. For as much data that has been available in those reviews, here is even more data accumulating thanks to the open-source Phoronix Test Suite and OpenBenchmarking.org. Thousands of data points are building up for these very exciting AMD Zen 3 desktop processors.
Since the initial reviews on Phoronix I continue running more benchmarks on the AMD Zen 3 CPUs on Ubuntu 20.10 and in other hardware/software configurations -- both for future Phoronix articles and also just running more benchmarks to explore more areas of these new AMD CPUs. All of that public benchmarking funnels into OpenBenchmarking.org. Other sites using the Phoronix Test Suite and opting to upload their results like L1 Techs to OpenBenchmarking.org and other early Ryzen 5000 owners lucky enough to buy the CPUs already have also begun uploading their results as well.
So thanks to sufficient data accumulating in areas where deemed statistically significant, more insights can be learned from the OpenBenchmarking.org pages for those eyeing the Ryzen 5000 series.
There are the search pages for the Ryzen 5 5600X, Ryzen 7 5800X, Ryzen 9 5900X, and Ryzen 9 5950X processors. From there you can view the lscpu/cpuinfo outputs on those parts as well as seeing percentile rankings for how these processors rank relevant to all of the other matching data on OpenBenchmarking.org where statistically significant. So like shown above you can easily see which workloads Zen 3 is offering smashing results on and other areas where there is room for improvement -- granted the percentile rankings also take into account server CPUs and other multi-socket solutions, etc. In any case, these Zen 3 CPUs are at the top of the stack for many of the benchmarks.
By navigating to individual test profile pages you can also have a look at how various Zen 3 processors rank for particular workloads... Take for instance the Linux kernel build performance, x265, eSpeak-NG, neural network benchmarks, and the hundreds of other test profiles. For most test profiles there is already enough statistically significant data on the Ryzen 5000 series that you should begin seeing the results listed.
With all the aggregated results, you can also slice and dice the data by looking at particular CPUs directly. Such as the Ryzen 5600X vs. 5800X vs. 5900X vs. 5950X. From there you can look at how the results compare to other processors where there is enough statistically, significant and matching data. All the results listed are backed by multiple data points and the other existing. From those component pages you can also put in your own local pricing information for the selected processors and see the performance-per-dollar graphs.
There is plenty of fun and interesting data to explore like Ryen 9 5900X vs. Core i9 10900K in 262 different tests, 5900X vs. 3900X vs. 2700X vs. 1800X in 66+ tests, and much more so go forth and explore the mass amount of open-source Linux benchmarking data available on OpenBenchmarking.org.
For those running the Phoronix Test Suite on your own systems, for test configurations where there is enough data accumulated on these new processors their data should begin appearing too if you are connected to the Internet and have the OB auto comparison functionality enabled.
Some other interesting Ryzen 5000 series Linux comparisons are also coming up on Phoronix over the weeks ahead from compiler benchmarking to Linux distribution comparisons.
As always, let me know of any OpenBenchmarking.org feature requests as continually adding new functionality for PTS/OB. Given the minimal ads on OpenBenchmarking.org, if you appreciate the resource you can also show your support by joining Phoronix Premium or PayPal tip as ultimately premium members also allow me to spend more time working on that code too when not busy with Phoronix.com content.
| 24
| 1,760,719,426.919093
|
https://www.phoronix.com/news/AMD-IBM-Confidential-Computing
|
AMD + IBM Team Up To Tackle Confidential Computing
|
Michael Larabel
|
AMD and IBM are this morning announcing a multi-year, joint development agreement focused on "building upon open-source software, open standards, and open system architectures to drive Confidential Computing in the cloud and support a broad range of accelerators across high-performance computing (HPC), and enterprise critical capabilities such as virtualization and encryption."
Researchers from both AMD and IBM will collaborate on hardware-based Confidential Computing to strengthen their presence in the cloud and enhancing artificial intelligence workloads. This makes more inroads for AMD EPYC in the cloud and from IBM's side strengthens their cloud offerings.
The brief press release being sent out notes that the joint development activities under the agreement are now underway. Unfortunately, however, when briefed under embargo they were not able to shed any light yet on what the initial focus is or what are the open-source projects they are looking to improve upon for this Confidential Computing initiative.
So for now any interesting technical details of this agreement are not announced but I'll be monitoring all of the usual mailing lists and Git repositories as always looking for any signs of joint AMD/IBM patches in this area over the months ahead. In any case it's good to see AMD engaging on what should be more open-source software infrastructure improvements with IBM.
| 22
| 1,760,719,427.106956
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.