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-Sensor-Fusion-Hub-SFH-1.1
|
AMD Sensor Fusion Hub "SFH1.1" Support Coming In Linux 5.20 For Newer Laptops
|
Michael Larabel
|
With the upcoming Linux 5.20 cycle is support for AMD's Sensor Fusion Hub v1.1 revision being found in newer Ryzen laptops.
With Linux 5.11 the AMD Sensor Fusion Hub Linux driver was introduced for supporting this sensor functionality found since original AMD Ryzen laptops but this driver took a while to be published and then mainlined. The AMD Sensor Fusion Hub is akin to the Intel Sensor Hub (ISH) and exposes the accelerometer, gyroscope, magnetometer, ambient light sensor, and human presence detection (HPD) sensors with capable laptops.
Since its mainlining the AMD SFH driver has been updated to support newer laptops, at the time when Cezanne was new. Now for Linux 5.20 it's preparing support for newer AMD Ryzen laptops as well. This time around its for supporting "SFH1.1" as a new revision of the Sensor Fusion Hub. Several hundred lines of new code are needed for bringing up the Sensor Fusion Hub 1.1 revision and ultimately exposes similar functionality to prior SFH hardware.
More AMD Rembrandt laptops continue working their way to market, such as the recent Lenovo ThinkPad X13 Gen 3 AMD model with Ryzen 7 PRO 6850U Zen 3+ SoC.
The patches adding AMD SFH1.1 support only mention that this updated Sensor Fusion Hub is found with "newer AMD SoCs", presumably appearing with new AMD Rembrandt laptops as there have been a few other Rembrandt-related Linux patches volleyed recently on the kernel mailing list.
This merge has added all of the new SFH1.1 code into HID-next ahead of the Linux 5.20 merge window opening in another week.
| 1
| 1,760,719,240.052572
|
https://www.phoronix.com/news/AMD-Raphael-Audio-Driver-Linux
|
AMD Posts New Linux Audio Driver Code For Raphael
|
Michael Larabel
|
AMD today posted new open-source Linux driver code for enabling the audio co-processor (ACP) with the upcoming Raphael platform with the Ryzen 7000 series processors.
The good news is AMD is publishing the ACP support, but the bad news is this is coming rather late... The AMD Ryzen 7000 "Raphael" desktop processors are inching closer to release. The Linux support has appeared to be coming together nicely for these Zen 4 processors over recent kernel releases. Now it turns out this new ACP audio driver code is necessary for those relying on the integrated ACP audio with these upcoming chips.
With the Ryzen 7000 series launch nearing, it was just earlier this week that AMD began teasing more of AM5 with this official shot. Over the span of 5 patches today the new header files and other patches to get the AMD SoC ACP Linux driver into shape for the Raphael "RPL" platform. But due to the timing of this, the Linux 5.20 merge window is opening up in just over one week and not immediately clear if this Raphael code can be rushed through into sound-next in time. Otherwise if missing the Linux 5.20 merge window it won't be merged until Linux 6.0 later in the year, certainly past the Raphael launch window.
The timing of this Raphael ACP audio support is unfortunately late and with the kernel timing may mean no ACP audio support out-of-the-box at launch, especially with some autumn Linux distribution releases possibly sticking with the v5.19 kernel. In any event at least the other areas of the Zen 4 desktop CPU support should be in place for launch including the likes of temperature monitoring and other bits that sometimes have been late to the party.
In any event, and as usual, stay tuned to hear more about the AMD Ryzen 7000 series "Raphael" on Linux when the time comes.
| 1
| 1,760,719,241.758439
|
https://www.phoronix.com/news/Linux-AMD-PSP-Cool-Down
|
Linux Optimization Patch Wants AMD's PSP To Cool Down
|
Michael Larabel
|
The newest AMD Linux optimization patch for the kernel aims to introduce a cool down period for the AMD Platform Security Processor (PSP) after each I2C transaction between the x86 CPU and the PSP.
Jan Dabros of Semihalf, a Polish embedded systems engineering firm that has previously worked on Linux patches around AMD PSP, suggested this kernel patch on Monday as a means of optimizing performance.
This cool down timer for the AMD PSP Linux driver would limit the amount of transactions with the PSP during which the bus isn't released after each I2C transaction. Currently the proposed cool down period is 100ms.
The proposed patch to the i2c-designware-amdpsp Linux driver can be found on the kernel mailing list where it's currently awaiting review. No performance measurements around this AMD PSP driver change were shared as part of the patch.
| 14
| 1,760,719,243.022166
|
https://www.phoronix.com/news/AMD-AOCL-3.2-Released
|
AOCL 3.2 Released As AMD Optimizing CPU Libraries Now With LibMEM & Crypto
|
Michael Larabel
|
Last week AMD quietly released AOCL 3.2 as the newest version of their optimized CPU software libraries for use across Ryzen, Ryzen Threadripper, and EPYC platforms.
The AMD Optimizing CPU Libraries (AOCL) started off as core math libraries optimized for AMD Zen-based processors and has grown to encompass more libraries with time. AOCL includes implementations for libM, FFTW, BLAS, BLIS, libFLAME, ScaLAPACK, and more.
Now with the AOCL 3.2 release AOCL-Cryptography has been added for AES encryption/decryption and SHA-2 hashing functions.
In addition to the AOCL-Cryptography introduction, AOCL 3.2 also brings AOCL-LibMem as AMD optimized memory and string functions with Zen 1/2/3 optimized memcpy/mempcpy/memmove/memset/memcmp functions.
AOCL 3.2 also brings improvements to its BLIS library, AOCL-FFTW is now aligned with upstream FFTW 3.3.10, new complex number variant functions for AOCL-libM, and various other additions/improvements.
Downloads and more details on AMD AOCL 3.2 via developer.amd.com.
| 2
| 1,760,719,243.35671
|
https://www.phoronix.com/news/Retbleed-Ryzen-7-4800U
|
The Current Retbleed Performance Costs With An AMD Ryzen 7 4800U
|
Michael Larabel
|
Following some weekend benchmarks here are more complementary numbers on the Retbleed mitigation performance benchmark costs. These additional numbers are on a Zen 2 based AMD Ryzen 7 4800U APU that has been common both to laptops as well as embedded/low-profile devices for thin client computing, IoT / edge use-cases, and more.
This AMD Ryzen 7 4800U round of benchmarking was carried out after the initial Retbleed patches were merged to the mainline Linux kernel this week following its public disclosure. The benchmarks were carried out in the default state with all relevant mitigations automatically applied and then repeating the run on the same software/hardware when booted with the "retbleed=off" kernel option to disable the Retbleed mitigation for the AMD Zen 2 CPU cores.
This is the current Retbleed costs as kernel developers eye possible mitigation improvements to reduce the overhead costs for both AMD and Intel CPUs. I'm also working on some other Retbleed benchmarks on AMD Zen 1 and on other hardware, including a look now at the compounded mitigation performance costs for the CPU vulnerabilities over the past four years.
The kernel context switching time continues to get much slower, the usual synthetic tests are significantly impacted, and I/O workloads, Java, RocksDB and other database systems and more are impacted from a few percent to double digit losses for Zen 2 with Retbleed.
See all of the Ryzen 7 4800U Retbleed benchmarks from this result page. If you missed it from a few days ago see the rest of my big Retbleed benchmarks while waiting on some follow-up comparisons in the days ahead.
| 16
| 1,760,719,244.642077
|
https://www.phoronix.com/news/AMD-P-State-Precision-Boost
|
AMD P-State Linux Driver Updated With Precision Boost Control, Other Fixes
|
Michael Larabel
|
Back in Linux 5.17 the AMD P-State "amd_pstate" driver was introduced for Ryzen and EPYC systems as an alternative to the ACPU CPUFreq frequency scaling driver with an emphasis on delivering better power efficiency for modern AMD Zen 2 and newer systems. Since the mainlining there hasn't been too much change to this driver but now a new patch series has been sent out with some updates.
Earlier this year AMD made it easier to enable amd_pstate and there has been a few patches here and there, but no major improvements to AMD P-State since its mainline arrival. Meanwhile, ACPI CPUFreq has tended to perform better still in benchmarks when it comes to raw performance / throughput.
Out today though is a new patch series from AMD with updates to this CPU power management driver. First up, the new patches add support for the Precision Boost hardware control with AMD processors. AMD Precision Boost is the fine-grained frequency control mechanism for optimizing power/performance. (There is Precision Boost Overdrive - PBO - as an extension to that for allowing optionally exceeding the rated power specifications of the platform.)
With the patches sent out today, the AMD Precision Boost support can be controlled when using amd-pstate by way of the /sys/devices/system/cpu/cpufreq/boost sysfs interface for monitoring the status or forcing it on/off in real-time.
In addition to the AMD Precision Boost control, the patch series also fixes an issue around the wrong lowest performance value reading, shortening the frequency transition delay time, and other fixes/clean-ups.
The updated patch series can be found on the mailing list for review. I'll be running some tests shortly of these new AMD P-State patches.
| 19
| 1,760,719,245.147597
|
https://www.phoronix.com/news/AMD-2022-Multimedia-Hiring
|
AMD Hiring For Another Open-Source GPU Driver Developer With Multimedia Expertise
|
Michael Larabel
|
It's great seeing AMD continuing to hire for more Linux/open-source driver developers. Beyond their many roles they are still working to fill on the CPU side of the house, they have a new job posting in hiring for their open-source GPU driver stack with a focus on multimedia efforts.
AMD is looking for another multimedia expert to join their "open-source GPU drivers for Linux" team that they note is used by the Steam Deck and Tesla Model S, besides being wildly popular among Linux gaming enthusiasts after open-source drivers.
They are looking for another engineer able to contribute to the Linux kernel and Mesa for AMD GPUs. Besides multimedia experience like around VA-API and more generally with video codecs and other multimedia processing knowledge, OpenGL and Vulkan will come into play too. (Hopefully we'll end up seeing good Vulkan Video support out of AMD once the Vulkan Video specs are ratified and extended to cover AV1 and the like...)
Thanks to Linux-using design wins from Tesla, the Valve Steam Deck, and more, AMD is hiring more Linux GPU driver developers -- in addition to their continued Linux hiring to support Ryzen and EPYC efforts. In any event, see this job posting for more details on AMD's Linux multimedia engineering needs.
Outside of that, AMD continues with many other current Linux jobs on the CPU side of the company to work on virtualization, QA, CXL, and other kernel engineering roles.
| 19
| 1,760,719,246.513476
|
https://www.phoronix.com/news/AMD-Rembrandt-Coreboot
|
AMD Rembrandt SoC Support For Coreboot In Place - Based Off Existing "Sabrina" Code
|
Michael Larabel
|
Today's Coreboot code now has AMD Rembrandt SoC support by splitting it out from the Sabrina SoC support that has been in the works the past several months for this open-source firmware project.
As I wrote about earlier this year, Coreboot had been seeing many patches for an AMD "Sabrina" SoC as well as a Google Chromebook device using this SoC. Sabrina hadn't been referenced on AMD public roadmaps and thus much curiosity over this chip and its purpose. It turns out that Sabrina is derived from Rembrandt, so presumably some special offshot of that dedicated for Google Chromebook purposes.
With Sabrina being derived from Rembrandt, today's Coreboot Git code shifts it around so that AMD Rembrandt support is now exposed as a base SoC from that prior Sabrina code and then Sabrina now based off of that. Namely the change is some Kconfig magic for now exposing AMD Rembrandt.
Granted, this is just concerning the SoC itself and it remains to be seen how many Rembrandt motherboards will end up seeing Coreboot support outside of the Chromebook space. So far in the Coreboot Review are no other AMD Rembrandt related patches pending.
AMD Rembrandt is made up of the Ryzen 6000 mobile series processors as 6nm Zen 3+ parts. Later this month I'll finally have up extensive AMD Rembrandt Linux benchmarks via the Ryzen 7 PRO 6850U with finally getting my hands on some hardware.
| 1
| 1,760,719,247.71356
|
https://www.phoronix.com/news/AMD-Jadeite-Linux-2022
|
AMD Gets Back To Working On Their "Jadeite" Platform For Linux In 2022
|
Michael Larabel
|
Somewhat surprisingly, AMD engineers have been working on some new Linux kernel patches for their aging Jadeite platform.
Jadeite? It didn't ring a bell for me at first either and not many search references in regards to AMD Jadeite... The Linux patches this week summed it up as "Jadeite(JD) platform is Stoney APU variant."
Stoney is several years old now and surprising to see AMD going back to working on this platform with new driver work as usually they are more focused on the much newer hardware... At first I was wondering if AMD was re-spinning Stoney Ridge APUs for some odd but new customer use-case, but Jadeite itself has been out since 2017. The only existing Jadeite reference I was able to find within the Linux kernel source tree was this AMD DRM patch from 2017 for updating YUV plane offsets and the patch noting that it was tested by booting on Jadeite.
AMD working on 600+ new lines of kernel code for improving their Jadeite platform on Linux. This new AMD Jadeite work in 2022 is providing I2S MICSP support for audio under Linux with the ES8336 audio codec. The patches are indeed new and carrying just a 2022 copyright. The patches for enabling I2S MICSP with the ES8336 codec for just the Jadeite platform, not Stoney at large.
With these patches the new audio driver code can be enabled via the SND_SOC_AMD_ST_ES8336_MACH Kconfig build option. The patches under review don't outline AMD's motivation for finally working to improve Jadeite/Stoney Linux support in 2022. Not that we're complaining, but just not used to see AMD going back to tackle major new feature code for 5+ year old platforms given their still limited (but growing) open-source engineering resources / staff compared to the likes of Intel and with Jadeite being just an apparently niche offshoot of Stoney Ridge.
| 6
| 1,760,719,248.713314
|
https://www.phoronix.com/news/AMD-FSR-2.0-Source-Published
|
AMD FidelityFX Super Resolution 2.0 Source Code Published
|
Michael Larabel
|
After AMD announced FidelityFX Super Resolution 2.0 back in March, as of today they have made good on their word to open-source it.
This temporal upscaling solution for game engines is now available under an MIT license. AMD self-describes FidelityFX Super Resolution 2.0 as, " FSR 2 uses cutting-edge temporal algorithms to reconstruct fine geometric and texture detail, producing anti-aliased output from aliased input. FSR 2 technology has been developed from the ground up, and is the result of years of research from AMD. It has been designed to provide higher image quality compared to FSR 1, our original open source spatial upscaling solution launched in June 2021."
The FSR 2.0 source code includes the complete C++ code-base as well as the HLSL source code for Direct3D usage. There is also a code sample implementation and FSR 2.0 API documentation. FSR 2.0 supports Direct3D 12 and Vulkan.
The FSR 2.0 source code and documentation is now available via GitHub. It will be interesting to see what open-source usage comes out of FidelityFX Super Resolution 2.0.
| 47
| 1,760,719,249.322787
|
https://www.phoronix.com/news/AMD-Ryzen-Embedded-R2000
|
AMD Announces Ryzen Embedded R2000 Series With Zen+ Cores, Radeon Graphics
|
Michael Larabel
|
AMD is using the Embedded World conference in Nürnberg to launch the Ryzen Embedded R2000 series for industrial use-cases along with IoT, thin clients, and edge computing.
AMD is advertising the Ryzen Embedded R2000 series as offering up to 81% higher CPU and graphics performance over the R1000 series, but with this new series also comes up to 2x more cores. The Ryzen Embedded R2000 series is making use of the Zen+ architecture and Radeon (Vega-based) graphics. The improved Radeon graphics abilities on the R2000 series allow for driving up to four 4K displays.
The Ryzen Embedded R2000 series currently tops out with the R2544 that in the 35~54 Watt range offers 4 cores / 8 threads, 8 graphics compute units, and a 3.35GHz base frequency. But the Ryzen Embedded R2544 and R2514 models won't be available to system integrators until October while the lower-end R2313 and R2314 models are immediately available. AMD plans to make the Ryzen Embedded R2000 series available for up to 10 years to satisfy industrial use-cases. Windows 11/10 and Ubuntu Linux LTS releases are officially supported for the Ryzen Embedded R2000 series. Granted, most (or all) other modern Linux distributions should play fine with the Zen+ cores and Radeon graphics.
AMD Ryzen Embedded R2000 series line-up. More details on the Ryzen Embedded R2000 series via AMD.com.
| 18
| 1,760,719,250.869563
|
https://www.phoronix.com/news/AMD-PRO-5000-WX-Update
|
AMD PRO 5000 WX Series Coming To More System Integrators, DIY Market Later This Year
|
Michael Larabel
|
After announcing the Threadripper PRO 5000 WX series back in March and with Lenovo being their launch partner for these Zen 3 Ryzen Threadripper CPUs, AMD today shared an update on availability.
AMD announced today that Ryzen Threadripper PRO 5000 WX Series processors will be available from more system integrators beginning in July.
Beyond the initial Lenovo ThinkStation with Ryzen Threadripper PRO 5000 WX series SKUs, the Dell Precision 7865 workstation is also coming as another system with these new Threadripper CPUs. Leading system integrators worldwide are expected to begin offering the PRO 5000 WX Series in July.
Meanwhile, "later this year" is when AMD says these processors will be coming to the DIY market.
AMD's brief update for the Ryzen Threadripper PRO 5000 WX series can be found on the community blog.
| 7
| 1,760,719,251.587827
|
https://www.phoronix.com/news/AMD-Pensando-Elba-Patches
|
Linux SoC Patches Revised For The AMD Pensando Elba
|
Michael Larabel
|
Earlier this year Pensando engineers began posting Linux patches for enabling their Elba DPU SoC. This data processing unit is powered by 16 x Arm Cortex-A72 cores and designed for supporting up to dual 200GE networking with this SoC intended for high-end networking equipment. It didn't take long for the AMD integration less than one month after AMD completed its Pensando acquisition with the new Linux patches now reflected as the AMD Pensando Elba.
Sent out this week was the fifth iteration of the enablment patches for what's now referred to as the AMD Pensando Elba SoC and sent with an AMD.com address, just weeks after the acquisition was completed. Indeed part of these "v5" patches is just about renaming references from Pensando to AMD-Pensando. The patches also re-base against the latest upstream Linux state and other basic changes.
Pensando Systems has been contributing to the Linux kernel even prior to the start-up exiting stealth mode. AMD purchased the young company for $1.9 billion USD to expand its data center offerings.
See the new patches for the latest AMD Pensando Elba SoC enablement work. With things winding down on that enablement, it's quite possible it will be mainlined for the v5.20 kernel cycle.
| 4
| 1,760,719,252.719382
|
https://www.phoronix.com/news/AMD-k10temp-Zen-4-v5
|
AMD Linux CPU Temperature Driver Sees Latest Patches For Zen 4 & Likely Mendocino
|
Michael Larabel
|
One of my personal gripes with AMD's Zen CPU support on Linux has been the lack of timely support for CPU temperature monitoring with their "k10temp" driver. Even though usually just new IDs are often needed and sometimes needing to adjust offsets or other minor changes, it has traditionally been done post-launch and sometimes left up to patches from the open-source community. Thankfully that has been changing and with Zen 4 it looks like that support will be ready for launch-day with the mainline Linux kernel.
AMD with Rembrandt (Yellow Carp) they managed to land the k10temp support pre-launch and for Zen 4 they have also been working for some time now on ensuring CPU temperature monitoring on Linux is working... It's certainly not critical for most users, but a personal frustration since for launch-day CPU reviews it's great including thermal data that hasn't been possible when requiring more than the new IDs when in the past there were at times questions over offsets, etc. While on the AMD side, it's not a huge engineering effort but something that pleases tech-minded Linux enthusiasts.
Going back to last year AMD was preparing for next-gen CPUs with k10temp changes to support up to 12 CCDs and making other driver patches for new PCI IDs. Out today is the latest patch series for k10temp in preparing the Zen 4 support.
The patches going back to last year cited new Family 19h Models while now it turns out some model IDs were missing. The patch series from AMD Linux engineer Mario Limonciello explains, "This series started as what looked like a correction to previous commits, but I missed that the previous commits were for a different family with the same chip models. So while fixing up the series I also noticed that a few upcoming chips have new PCIe IDs and CCD offsets not yet supported, so add them to amd_nb/k10temp."
Lisa Su showing off Zen 4 back during her Computex 2022 keynote. What this fifth iteration of the k10temp enablement patches amounts to is supporting Family 17h A0h-AFh, Family 19h 70h-7Fh, and Family 19h 60h-6Fh. Family 17h is for Zen / Zen 2 while Family 19h so far is for Zen 3. As noted in prior Phoronix articles, based on other Linux kernel patches, it almost definitely appears Family 19h will include Zen 4 CPUs so all indications so far have been the 60h and 70h parts are likely for different Zen 4 (and Zen 4C) chips. The new IDs in Family 17h are presumably for the upcoming Mendocino SoCs for budget laptops... Otherwise I can't think of what other Zen 2 processors are currently not covered by the k10temp Linux driver support.
In any case, the v5 patches are now out there with the AMD Linux temperature driver "k10temp" ready to support these upcoming processors. Hopefully these patches will manage to land for the upcoming Linux 5.20 cycle so that there is the mainlined kernel support ahead of the Zen 4 desktop and server processor launches later this year.
| 20
| 1,760,719,253.382847
|
https://www.phoronix.com/news/Linux-5.20-AMD-NUMA-Imbalance
|
With A Few Lines Of Code, AMD's Nice Performance Optimization For Linux 5.20
|
Michael Larabel
|
A patch from AMD to further tune the Linux kernel's scheduler around NUMA imbalancing has been queued up and slated for introduction in Linux 5.20. For some workloads this scheduler tuning can help out significantly for AMD Zen-based systems and even on Intel Xeon servers has the possibility of helping too.
The change from AMD to the fair scheduler is considering CPU affinity when allowing NUMA imbalance within the find_idlest_group() function.
AMD engineer K Prateek Nayak explained:
In the case of systems containing multiple LLCs per socket, like AMD Zen systems, users want to spread bandwidth hungry applications across multiple LLCs. Stream is one such representative workload where the best performance is obtained by limiting one stream thread per LLC. To ensure this, users are known to pin the tasks to a specify a subset of the CPUs consisting of one CPU per LLC while running such bandwidth hungry tasks.
...
Ideally we would prefer that each stream thread runs on a different CPU from the allowed list of CPUs. However, the current heuristics in find_idlest_group() do not allow this during the initial placement.
[Example behavior]
Once the first four threads are distributed among the allowed CPUs of socket one, the rest of the treads start piling on these same CPUs when clearly there are CPUs on the second socket that can be used.
Following the initial pile up on a small number of CPUs, though the load-balancer eventually kicks in, it takes a while to get to {4}{4} and even {4}{4} isn't stable as we observe a bunch of ping ponging between {4}{4} to {5}{3} and back before a stable state is reached much later (1 Stream thread per allowed CPU) and no more migration is required.
We can detect this piling and avoid it by checking if the number of allowed CPUs in the local group are fewer than the number of tasks running in the local group and use this information to spread the 5th task out into the next socket (after all, the goal in this slowpath is to find the idlest group and the idlest CPU during the initial placement!).
For the Stream memory benchmark test case, this patch was able to increase performance upwards of 40%:
This patch on top of the current Linux kernel code benefited Stream by 36~44% as an example common test case.
While an AMD led optimization for benefiting their Zen-based processors with multiple last level caches per socket, this Linux scheduler change can also benefit Intel CPUs too in cases of multi-socket servers. For Stream on an Intel Xeon Scalable "Ice Lake" server the Stream performance saw a 54~82% improvement over the current Linux performance.
Not bad at all with this kernel patch just being a few lines of code!
As of this morning the patch was queued into sched/core, making it material to be sent in for the Linux 5.20 merge window later this summer unless any issues come about with this code that previously was residing on the kernel mailing list. Once the Linux 5.20 cycle is abound with this and any other optimizations, I'll definitely be around with some fresh Xeon and EPYC benchmarks.
| 18
| 1,760,719,254.982301
|
https://www.phoronix.com/news/LUMI-Supercomputer-Inaugurated
|
LUMI Inaugurated As Europe's Most Powerful Supercomputer - Powered By AMD CPUs/GPUs
|
Michael Larabel
|
While not record-shattering like the 1.1 Exaflops Frontier supercomputer at ORNL that took the Top500 spot this year from Fugaku, LUMI was inaugurated today with the claim of Europe's most powerful supercomputer.
LUMI was inaugurated today as "Europe's most powerful supercomputer" and will be used for matters like fighting climate change. LUMI is owned by the EuroHPC Joint Undertaking as a consortium of 10 European countries while the supercomputer is physically located in Finland.
LUMI consists of HPE Cray EX nodes with 2,560 nodes consisting of one 64-core AMD EPYC "Trento" CPU and four AMD Instinct MI250X GPUs. Meanwhile the "LUMI-C" partition is just CPU-focused and contains 64-core AMD EPYC 7003 series CPUs in 1,536 dual-socket nodes. LUMI also has 64 NVIDIA A40 GPUs used for visualization workloads. While pre-Exascale, globally LUMI currently is ranked the third fastest supercomputer in the world.
LUMI picture courtesy of the EuroHPC JU.
Details on today's inauguration via Lumi-SuperComputer.eu.
We've taken extra interest in LUMI thanks to their HPC researchers having made nice presentations on AMD Linux code porting, the state of ROCm, and related open-source engagements around the growing AMD HPC Linux wins and technical challenges that have been involved from the software side.
| 18
| 1,760,719,254.991457
|
https://www.phoronix.com/news/AMD-Financial-Day-2022
|
AMD Talks Up Zen 4 AVX-512, Genoa, Siena & More At Financial Analyst Day
|
Michael Larabel
|
AMD today hosted their 2022 Financial Analyst Day where they made some new disclosures and firmed up past product road-map plans.
Among the mix of technical information shared at the AMD Financial Analyst Day 2022 included:
- AMD notes 5nm and 4nm products in the Zen 4 space and 4nm and 3nm for Zen 4 processors, without noting the breakdown of SKUs or chiplets between the different processes.
- AMD confirmed AVX-512 and AI acceleration instructions for Zen 4.
- 9~10% instructions per clock (IPC) increase with Zen 4 over Zen 3.
- Up to 125% memory bandwidth per core improvement with Zen 4, thanks to DDR5.
- Zen 5 in 2024 will indeed be a new "grounds-up microarchitecture" redesign. More AI and machine learning optimizations are expected for Zen 5.
- For EPYC 7004 "Genoa", AMD is talking greater than 75% improvement for enterprise Java performance compared to EPYC 7003 series. 12 channel DDR5 confirmed for Genoa as well as PCIe Gen 5 and CXL support. Genoa is on track for Q4 launch.
- EPYC "Bergamo" with Zen 4C cores for cloud computing will be able to offer twice the cloud density as 3rd Gen EPYC. Bergamo CPUs will launch in H1'2023.
- Genoa-X was confirmed as Genoa with 3D V-Cache. Genoa-X will offer more than 1GB of 3D V-Cache per CPU/socket.
- "Siena" was announced as a lower-cost Zen 4 platform intended for edge computing and telecommunications. Will be quite interesting to see how this Siena play is for lower-cost EPYC.
Hopefully with the disclosures now like AVX-512 for Zen 4 publicly confirmed, hopefully AMD will move on to posting the GCC and LLVM/Clang Znver4 compiler target and other remaining Zen 4 enablement patches for Linux.
| 36
| 1,760,719,256.628895
|
https://www.phoronix.com/news/Linux-5.20-AMD-MWAIT
|
Linux 5.20 With AMD Zen Will Prefer MWAIT Over HALT As An HPC Optimization
|
Michael Larabel
|
Earlier this year was an AMD Linux patch to prefer using MWAIT rather than HALT for cases where the CPU idle driver isn't being used. Using MWAIT can lead to significant improvements for the exit latency and now for the Linux 5.20 cycle later this year that change is expected to land.
Intel CPUs going back to the Core 2 days have preferred MWAIT over HALT (HLT) for the C1 code in the Linux kernel. The Monitor Wait "MWAIT" instruction can be used for power management purposes to hint that the processor can enter a specified target C state while waiting for an event or a MONITOR'ed store operation to complete. Usage of MWAIT is intended to be more efficient than the HALT instruction, which AMD CPUs on Linux continue using when not relying on the kernel's idle driver. AMD Zen CPUs though do support MWAIT for more efficient handling (or going back to K10 does support MWAIT but not MWAIT C1 support until Zen 1).
This week now that the Linux 5.19 merge window has passed, the patch has been picked up by TIP's x86/cpu branch. In turn that work should wind up being submitted to the mainline kernel for the Linux 5.20 cycle later this summer. This benefits all generations of AMD Zen processors going back to the original Zen 1 CPU models.
For those like HPC users that disable global C-states / opt for BIOS performance profiles, this change to prefer MWAIT C1 over HALT should yield nice improvements to the exit latency. This change doesn't affect Intel Linux users with the kernel for many years and going back to the Core 2 CPU days already preferring MWAIT. Those wanting to compare or test old behavior can boot the kernel with "idle=nomwait" as a kernel command line parameter to use HALT (HLT) instead.
| 2
| 1,760,719,256.638094
|
https://www.phoronix.com/news/AMD-Multi-GPU-Compute-P2P
|
AMD Kernel Driver Enabling Peer-To-Peer Multi-GPU Compute For Linux
|
Michael Larabel
|
A new patch series posted today by AMD is enabling peer-to-peer support within their AMDKFD kernel compute driver for allowing communication between multiple AMD GPUs over the PCIe bus without needing intermediate copies through system memory. In turn this should help with the multi-GPU compute performance for the Radeon ROCm stack.
The set of patches to the AMDKFD driver and toggled at build-time via the proposed "HSA_AMD_P2P" Kconfig switch enables GPUs to directly access the GPU video memory of other graphics cards without needing to go through system RAM. This AMDKFD feature works for compatible chipsets and where the BAR is large enough to expose the entire video memory capacity on the PCIe bus.
This P2P work also includes extending the KFD (Kernel Fusion Driver) device topology to surface peer-to-peer links and exposing the layout to user-space via sysfs. Previously the Radeon compute libraries in user-space attempted their own peer-to-peer handling while this integration from the kernel driver side should be superior in terms of robustness and reliability.
Given the timing of these new patches, this HSA_AMD_P2P work won't be merged until at least the Linux 5.20 cycle later this summer. This P2P multi-GPU work is just in the context of their AMDKFD compute driver code and not anything targeting their graphics side/APIs that is handled separately and already supported to varying degrees.
| 10
| 1,760,719,258.305263
|
https://www.phoronix.com/news/AMD-Pensando-Complete
|
AMD Completes Pensando Acquisition For Adding DPUs To Their Portfolio
|
Michael Larabel
|
It was just last month that AMD announced plans to acquire Pensando and today that $1.9 billion deal has been completed.
Pensando was the five year old company founded by former Cisco CEO John Chambers and focused on software-defined clouds, compute, and networking with their own DPU. AMD in particular was focused on Pensando's Data Processing Unit (DPU) and software stack for complementing their growing data center portfolio.
AMD (NASDAQ: AMD) today announced that it has completed its acquisition of Pensando Systems in a transaction valued at approximately $1.9 billion. Pensando’s distributed services platform will expand AMD’s data center product portfolio with a high-performance data processing unit (DPU) and software stack that are already deployed at scale across cloud and enterprise customers including Goldman Sachs, IBM Cloud, Microsoft Azure and Oracle Cloud. The Pensando team will join the AMD Data Center Solutions Group, led by AMD Senior Vice President and General Manager Forrest Norrod. AMD announced this morning they have successfully completed the $1.9 billion deal for Pensando Systems.
Pensando engineers have contributed to the Linux kernel over the past two years while moving forward under AMD's leadership it will hopefully be ramping up. Going back to late 2019 is when they began adding upstream Linux support for their hardware for what was still a less well known start-up at the time. Pensando recently has been in the process of upstreaming Linux patches for their Elba DPU for use within enterprise network switches.
| 6
| 1,760,719,259.94555
|
https://www.phoronix.com/news/AMD-Virtual-NMI-VNMI
|
AMD Preparing To Finally Support Virtual NMI With Their CPUs (VNMI)
|
Michael Larabel
|
It appears that with upcoming AMD Zen 4 processors there will finally be Virtual NMI (VNMI) support for virtualization, a feature Intel CPUs have supported for well more than the past decade.
AMD engineers on Thursday sent out the initial Linux kernel patches for enabling Virtual NMI (Non-Maskable Interrupt) support when supported by the processor. The presence of VNMI support is detected by a CPU ID feature bit but given the timing they are either horribly late in supporting it for existing CPUs and not talked about its presence until now, or more than likely: VNMI is being introduced with upcoming AMD Zen 4 processors.
AMD's Santosh Shukla sums up their Virtual NMI implementation as:
Currently, NMI is delivered to the guest using the Event Injection mechanism. The Event Injection mechanism does not block the delivery of subsequent NMIs. So the Hypervisor needs to track the NMI delivery and its completion (by intercepting IRET) before sending a new NMI.
Virtual NMI (VNMI) allows the hypervisor to inject the NMI into the guest w/o using Event Injection mechanism meaning not required to track the guest NMI and intercepting the IRET. To achieve that, VNMI feature provides virtualized NMI and NMI_MASK capability bits in
VMCB intr_control -
V_NMI(11) - Indicates whether a virtual NMI is pending in the guest.
V_NMI_MASK(12) - Indicates whether virtual NMI is masked in the guest.
V_NMI_ENABLE(26) - Enables the NMI virtualization feature for the guest.
When Hypervisor wants to inject NMI, it will set V_NMI bit, Processor will clear the V_NMI bit and Set the V_NMI_MASK which means the Guest is handling NMI, After the guest handled the NMI, The processor will clear the V_NMI_MASK on the successful completion of IRET instruction Or if VMEXIT occurs while delivering the virtual NMI. Basically VNMI amounts to an efficiency optimization with AMD virtualized guests will not need to track the guest's NMI state and worrying about intercepting the IRET for the NMI completion handling. Virtual NMI has long been supported by Intel CPUs - Intel's Linux kernel code for virtualization in fact requires it and has been plumbed into the kernel on Intel's side since 2008.
Given the timing of this Linux patch series, VNMI will presumably premiere with next-generation Zen 4 Ryzen 7000 and AMD EPYC 7004 series hardware.
| 0
| 1,760,719,261.04881
|
https://www.phoronix.com/news/AMD-SEV-SNP-Arrives-Linux-5.19
|
AMD SEV-SNP Finally Being Merged In Linux 5.19 To Enhance Confidential Computing
|
Michael Larabel
|
Last year with the launch of AMD EPYC 7003 "Milan" processors one of the new security features was SEV-SNP, or the "Secure Nested Paging" update to the Secure Encrypted Virtualization functionality that has built up with succeeding EPYC generations. While AMD published out-of-tree kernel patches in a GitHub repository to enable SEV-SNP and has been volleying several revisions to them on the kernel mailing list, one year later it's finally arriving in mainline with the Linux 5.19 kernel.
The Linux 5.19 kernel to be released later this summer will offer AMD SEV-SNP support without having to resort to patching your kernel or using any out-of-tree code. As I wrote at the beginning of April, it looked like SEV-SNP was finally ready for upstreaming with Linux 5.19. Now on the first day of the v5.19 merge window, the SEV-SNP patches have indeed been submitted.
SEV-SNP has been one of many exciting enhancements with EPYC 7003 series.
The hardware-based memory integrity protections provided by AMD SEV-SNP can help prevent malicious hypervisor-based attacks and other functionality beyond what has already available with Secure Encrypted Virtualization of prior EPYC CPUs.
An AMD table showing the feature differences across the SEV tiers. SEV-SNP was introduced in March 2021 with AMD EPYC 7003 series processors.
SEV-SNP adds integrity protections around replay protection, data corruption, memory aliasing, memory re-mapping, TCB rollback, and more. The SEV-SNP for Linux 5.19 has all the initial enablement code in place but there still are some open tasks like handling the lazy validation mode for pages so for the moment it's all pre-validated at boot. The interrupt security enhancements are also still to be worked on for this Linux kernel code. See this AMD whitepaper to learn more about the various tiers of Secure Encrypted Virtualization.
This pull request is what has the initial SEV-SNP enablement for Linux 5.19. It's unfortunate it took over one year after the EPYC Milan launch for this less than 4k lines of code to be upstreamed into the kernel, but at least it's here now and after extensive review/testing. Many hyperscalers and other big EPYC customers likely have already been using SEV-SNP by patching their kernel builds but it's great to have all of this mainlined to make the availability of SEV-SNP more widespread and easier to maintain. Like previously with SEV-ES, it took rather a long time post-launch before being mainlined. This is one of the areas where AMD still has room for improvement with their Linux support is getting such new CPU features out there more punctually -- Intel meanwhile is known for their timely enablement and getting next-generation CPU features generally into the kernel prior to the hardware shipping. For EPYC Zen 4 processors I have already been reporting on various ID additions and basic enablement work for months but so far haven't seen any major feature code work its way to the kernel mailing list to begin the review/upstreaming process. Update: As another example for Intel's timely enablement approach... Today Intel TDX was submitted for Linux 5.19 too with Trust Domain Extensions being Intel's SEV alternative coming with Xeon Scalable Sapphire Rapids.
| 0
| 1,760,719,261.7121
|
https://www.phoronix.com/news/AMD-Kria-KR260-Robotics
|
AMD Launches Xilinx + Linux Powered Robotics Starter Kit
|
Michael Larabel
|
AMD on Tuesday released the Kria KR260 Robotics Starter Kit featuring a Xilinx Kria K26 System-on-Module and tailoring it for robotics, machine vision, and industrial communication/control use-cases while running Ubuntu 22.04 LTS.
The Kria K26 module on the Kria KR260 Robotics Starter Kit features a Xilinx Zynq Ultrascale+ with quad-core Arm Cortex-A53 processor, 4GB of DDR4 memory, and 512Mbit QSPI flash. There is microSD/SDHC support, DisplayPort 1.2a output, four Gigabit Ethernet ports, and one SFP+ connector with 10GbE support.
AMD/Xilinx is officially supporting Ubuntu 22.04 LTS and ROS 2.
Priced at $349 USD, the Kria KR260 Robotics Starter Kit is an interesting alternative to NVIDIA's Jetson developer kits that are very popular for robotics use.
This robotics starter kit is launching immediately though already on Xilinx.com it's listed as having a 20 weeks lead time and DigiKey lists the product as out-of-stock.Learn more about this robotics kit via the AMD press release.
| 11
| 1,760,719,263.337048
|
https://www.phoronix.com/news/AMD-Computex-2022
|
AMD @ Computex 2022 Talks Up Ryzen 7000 Series, Announces Mendocino Budget Laptop APUs
|
Michael Larabel
|
AMD CEO Lisa Su keynoted this morning for Computex 2022 where she talked up some of the company's processor plans for the rest of the year. The focal points were on the much anticipated Ryzen 7000 series desktop processors as well as announcing the "Mendocino" APUs that will be coming to affordable laptops later in the year.
Some of the key takeaways from the AMD Computex 2022 keynote include:
- AMD announced the "Mendocino" laptop processors that will be launching in Q4 for the $399~699 USD mainstream laptops. These Mendocino processors are manufactured on a 6nm TSMC process, sport 4 cores / 8 threads, and have RDNA2 graphics. Mendocino uses the Zen 2 microarchitecture.
- Lisa talked up the Ryzen 7000 series and reiterated its support for PCIe 5.0 and DDR5 memory (no DDR4 support). AMD claims the single-thread performance will be more than 15% greater than current Ryzen 5000 series processors and boast twice the L2 cache per core. The new CPUs can boost above 5GHz and a demo was shown with Ryzen 7000 series boosting to 5.5GHz.
- AMD confirmed the rumors that the Ryzen 7000 series desktop processors will indeed include an RDNA2 integrated GPU.
- The AMD X670 Extreme, X670, and B650 chipsets were introduced for AM5 / Ryzen 7000 series processors. The B650 will be limited to PCIe 4.0 while the X670 non-Extreme makes PCIe 5.0 optional. The B650 will also lack overclocking support.
- The Ryzen 7000 series processors will be shipping in the fall.
The keynote in full is available below.
| 58
| 1,760,719,263.642971
|
https://www.phoronix.com/news/AMD-Memory-Device-Coherent
|
AMD Posts Latest Coherent Device Memory Mapping Linux Code - Designed For Frontier
|
Michael Larabel
|
For over the past year we've seen various patches posted by AMD engineers with a state effort around preparations for the Frontier supercomputer. Most of these patches have involved memory handling under Linux and the special purpose memory handling between the CPU/GPUs. Published on Monday was their latest work on coherent device memory mappings for the Linux kernel.
This "MEMORY_DEVICE_COHERENT" was worked on by AMD engineers for their Frontier supercomputer effort but can be relevant to other, future supercomputers and the code also being of possible interest to other hardware vendors too. This latest effort is summed up as:
This patch series introduces MEMORY_DEVICE_COHERENT, a type of memory owned by a device that can be mapped into CPU page tables like MEMORY_DEVICE_GENERIC and can also be migrated like MEMORY_DEVICE_PRIVATE.
...
System stability and performance are not affected according to our ongoing testing, including xfstests.
How it works: The system BIOS advertises the GPU device memory (aka VRAM) as SPM (special purpose memory) in the UEFI system address map.
The amdgpu driver registers the memory with devmap as MEMORY_DEVICE_COHERENT using devm_memremap_pages. The initial user for this hardware page migration capability is the Frontier supercomputer project. This functionality is not AMD-specific. We expect other GPU vendors to find this functionality useful, and possibly other hardware types in the future.
Our test nodes in the lab are similar to the Frontier configuration, with .5 TB of system memory plus 256 GB of device memory split across 4 GPUs, all in a single coherent address space. Page migration is expected to improve application efficiency significantly. We will report empirical results as they become available. See the latest MEMORY_DEVICE_COHERENT patch series for more technical details if interested.
ORNL photo showing Frontier under construction. Frontier is the exascale supercomputer currently being built for Oak Ridge National Laboatory and expected to reach full capability this calendar year using a combination of AMD EPYC 3rd Gen CPUs and AMD Instinct 250X GPUs. The coherent interconnects between the CPUs and GPUs with xGMI has been what's seeing most of the Frontier-mentioning Linux support patches for getting the software support all in order. Frontier once fully operational should be delivering above 1.5 Exaflops compute performance.
| 1
| 1,760,719,264.817324
|
https://www.phoronix.com/news/AMD-Prefer-MWAIT-v3
|
AMD Updates Linux Patches For Lowering Idle Exit Latency
|
Michael Larabel
|
Last month an AMD engineer began posting Linux kernel patches so the kernel prefers the MWAIT instruction over HALT for lowering the CPU idle exit latency. Preferring MWAIT over HALT has been something Intel CPUs on Linux have preferred going back to the Core 2 days and indeed with modern AMD CPUs there is significant advantages to lowering the exit latency in doing so for the idle code. This morning the latest iteration of the work was posted.
An updated version of the Linux patch series so AMD CPUs prefer the MWAIT usage over HALT (HLT instruction) was sent out today by AMD's Wyes Karny. The MWAIT instruction is more efficient than HLT and this code change will make use of that on systems lacking CPU idle driver support -- such as when global C-states are disabled by the system BIOS or kernel builds without CPU idle enabled. This optimization is done with AMD's high performance computing (HPC) customers in mind where they often disable global C-states intentionally.
AMD has not relied on MWAIT in this configuration since families prior to Family 10h / K10 have not supported the instruction while K10 to Bulldozer CPUs have supported MWAIT but not the MWAIT C1. It's basically with Zen 1 and newer where using MWAIT is clearly the better choice over HALT. The Monitor Wait "MWAIT" instruction hints for the processor to stop instruction execution and can enter an optimized state. The Linux kernel makes use of MWAIT/HLT as part of its idle loop.
With AMD's own numbers on Zen 3, they have found with this patch series going from HLT to MWAIT improves the exit latency by 21.74% for one test or in another test of context switching performance the wake-up latency reduction netted a 45% improvement. AMD has confirmed similarly benefits with older Zen CPUs too.
See this patch series if interested in more details. Hopefully this work will manage to get buttoned up in time for the v5.19 kernel cycle.
| 7
| 1,760,719,266.66936
|
https://www.phoronix.com/news/Xilinx-2022-DRM-Driver
|
AMD's Xilinx Posts New Linux DRM Display Driver
|
Michael Larabel
|
The AMD-owned Xilinx posted a new patch series today implementing a new DRM display driver for supporting their soft MIPI DSI Tx subsystem IP.
This display driver is for use with Xilinx FPGA platforms for soft display purposes from an XI-4 stream interface with programmable DSI subsystem controller.
The Xilinx MIPI DSI Tx Subsystem soft IP is used to display video data from AXI-4 stream interface. It supports upto 4 lanes, optional register interface for the DPHY and multiple RGB color formats. This is a MIPI-DSI host driver and provides DSI bus for panels. This driver also helps to communicate with its panel using panel framework. The Xilinx DSI Tx subsystem makes use of the MIPI protocol and the D-PHY physical layer provides a direct connection to display peripherals.
Those interested in this new open-source DRM driver that is aiming to ultimately be included in the mainline Linux kernel can find the initial patch series now awaiting review.
| 2
| 1,760,719,268.104024
|
https://www.phoronix.com/news/AMD-Zen-4-IBS-Linux
|
AMD Zen 4 IBS Extensions Under Review For Linux
|
Michael Larabel
|
Upcoming AMD Zen 4 processors are bringing improvements to their Instruction-Based Sampling (IBS) capabilities that can be utilized by Linux's wonderful perf utility and subsystem.
At the end of April was the initial patch series with a revised series sent out this morning for new IBS extensions with AMD Zen 4. The patch series is also notable in it being the first Linux kernel patch series explicitly referencing "Zen4" rather than just calling it a future/upcoming architecture. All the other recent Zen 4 patch series have just used generic/vague terminology even though we all know it's for Zen 4 given AMD's Linux upstreaming cadence and history around their Linux support timing.
Zen 4 will improve instruction-based sampling by adding a data source extension as well as a new L3 cache miss filtering capability. These new Zen 4 IBS features are summed up as:
DataSrc extension provides additional data source details for tagged load/store operations. Add support for these new bits in perf report/script raw-dump.
IBS L3 miss filtering works by tagging an instruction on IBS counter overflow and generating an NMI if the tagged instruction causes an L3 miss. Samples without an L3 miss are discarded and counter is reset with random value (between 1-15 for fetch pmu and 1-127 for op pmu). This helps in reducing sampling overhead when user is interested only in such samples. One of the use case of such filtered samples is to feed data to page-migration daemon in tiered memory systems.
Add support for L3 miss filtering in IBS driver via new pmu attribute "l3missonly". See the kernel mailing list for these Zen 4 IBS patches if you are a heavy Linux perf user and wanting to learn more about these new capabilities.
In general, besides perf instruction-based sampling being useful for profiling for possible optimizations and debugging of issues, the sampling is also useful for a growing number of compiler features for being able to feed the perf hardware sampling results back to the compiler for assisting in generating profile-based optimized binaries. With Intel long having been more at the forefront of hardware performance counters and the functionality exposed under Linux, it's good to see some IBS improvements coming with Zen 4.
| 1
| 1,760,719,268.481824
|
https://www.phoronix.com/news/AMD-Hiring-Embedded-Yocto
|
AMD Working to Create A New Yocto Linux Platform For Xilinx SoCs
|
Michael Larabel
|
Following AMD completing its Xilinx acquisition back in February, AMD is now preparing to ramp up their investment into embedded Linux. AMD is hiring for the "creation and maintenance" of a Yocto-based Embedded Linux platform for running on Xilinx SoCs.
While over the past year AMD has been hiring more Linux engineers especially for server and even client CPU computing, following their Xilinx acquisition they also have their sights set on embedded Linux. Yesterday is the first time I've seen AMD hiring for embedded Linux talent with a focus on the currently Xilinx branded hardware.
Most interesting is their recruiting of a "Senior Embedded Linux Software Engineer." This new position will be "the primary technical point of contact for the embedded Linux systems" and "Technical lead for: Creation and maintenance of Yocto based Embedded Linux platform running on Xilinx SoC solutions."
See the job posting if you're an embedded Linux expert and potentially interested in it or their other AMD Linux jobs.
It will be interesting to see what comes of this new embedded Linux push for AMD/Xilinx. There already has been OpenEmbedded and Yocto support for various Xilinx products and developer boards. See this Wiki page for some of the current Yocto support details on Xilinx hardware. Yocto as a refresher is the Linux Foundation backed project that provides a stack for creating IoT/embedded Linux distributions. Yocto has been going on for more than a decade and leading in the embedded Linux space and backed by vendors like Arm and Intel.
Coming back rather full circle, AMD originally joined the Yocto Project back in 2014. Back then they joined and became an advisory board member but not much came of it with their hardware at the time not being that compelling and the company under financial pressure (that was also the time period after AMD joined the MeeGo project that has since been discontinued). It's now a very different situation today with Ryzen Embedded and also the Xilinx portfolio. Yocto members page puts AMD as a silver member while Xilinx is with the likes of Arm, Intel, Cisco, Microsoft, and others as the flagship platinum member tier.
Another decade full circle is this embedded Linux hiring is for Dresden, Germany. AMD hasn't maintained any significant footprint in Dresden since they exited the fab business. However, Xilinx does have an office in Dresden. What's ironic is back during the exciting days of AMD's Fab 36 in Dresden, AMD's since closed Operating System Research Center (OSRC) was based there with at the time their excellent pool of talented open-source/Linux developers. It's been ten years since the AMD OSRC sadly closed and around two dozen Linux engineers let go, but at least they are now back to ramping up their Linux efforts. It's fascinating to think though how better positioned they could have been for today's dominant Linux use on servers and clouds down to IoT and edge computing if the OSRC had never been disbanded during their financial difficulties.In any event, it will be interesting to see how the new AMD/Xilinx Yocto Linux platform plays out. AMD is also working on other interesting software integration that may play out here like their in-progress work bringing Radeon ROCm to Xilinx hardware.
| 5
| 1,760,719,270.184904
|
https://www.phoronix.com/news/AMD-HSMP-New-Zen-4-Features
|
New AMD HSMP Driver Features Prepared Ahead Of Zen 4 EPYC
|
Michael Larabel
|
Merged in Linux 5.18 is the AMD HSMP driver for enabling the "Host System Management Port" usage under Linux as an interface for enabling additional system management functionality on AMD EPYC 7003 servers. For Linux 5.19 this AMD HSMP driver is set to be extended with additional features coming with next-generation AMD EPYC servers.
The AMD HSMP Linux driver in conjunction with current-generation AMD EPYC "Milan" processors allow for reading the current average socket power consumption, setting socket power limits, getting/setting maximum frequency limits, min/max width for xGMI links, reading the memory clocks, reading the average C0 residency, and other similar functionality.
A patch from AMD now under review for the amd_hsmp driver within the x86-platform-drivers area is enabling HSMP protocol version 5 support. AMD HSMP protocol version 5 is coming with AMD Family 19h Model 10h EPYC processors, expected to be Zen 4 / EPYC 7004 series based on all the 10h series patches we've seen in recent months.
Added with this new protocol version is support for DIMM statistics like per-DIMM power consumption and temperatures, bandwidth for I/O and xGMI links, setting of power efficiency modes, and other new features.
The AMD HSMP protocol v5 patch can be found here as it undergoes review and given the timing should ultimately find its way into the Linux 5.19 kernel.
| 0
| 1,760,719,273.034021
|
https://www.phoronix.com/news/AMD-GFX11-Patches
|
AMD Sends Out New Linux Patches For RDNA3 "GFX11"
|
Michael Larabel
|
AMD continues working on their open-source Linux driver support for next-gen GPUs... The latest patches posted on Friday are for "GFX11", pointing to the major new graphics IP version with RDNA3 graphics processors due out later this year.
AMD continues bringing up their new unreleased GPU support in a new "block by block" approach rather than the large, lofty patch series as with prior generations. As noted this has a number of benefits for improving their development workflow and making it easier to get the patches out there in a timely manner ahead of launch.
The latest excitement to note with their latest patch series is the introducing of the "GFX11" block. GFX9 was Vega/CDNA and GFX10 is for Navi 1x/2x (RDNA / RDNA2) while GFX11 is now almost definitively for RDNA3. The latest patches for ending out the week begin introducing the open-source driver support for GFX11.
AMD began landing patches inside the LLVM compiler for the "GFX11" target for their AMDGPU shader compiler back-end. AMD then moved on with beginning to post AMDGPU kernel driver patches around GFX11.
The GFX11 headers were posted and add 24.6k lines to the driver right there. There were then 73 patches sent out on Friday for enabling the MES - Micro Engine Scheduler. The new AMD Micro Engine Scheduler (MES) is a hardware scheduler micro-controller for scheduling engine queues to hardware slots. Following that were another 29 patches for enabling the graphics and compute support for the new GC 11.0 block that goes along with the MES engine management.
The open-source Linux driver enablement work for RDNA3 GPUs is certainly heating up and getting to the more interesting code drops. No super exciting details though were revealed by this latest series beyond confirming RDNA3 is ushering in GFX11 as opposed to another GFX10 rev. If all goes well there will be upstream support in the Linux kernel and Mesa for RDNA3 GPUs ahead of the graphics cards launching later this year - stay tuned as the enablement work continues and my monitoring of the Linux support state.
Sent out on Friday was also another pull of readied material for DRM-Next ahead of the Linux 5.19 kernel merge window at the end of May. Notable with that pull is enabling the SoC21 block but besides that is more IP-discovery work and other fixes and low-level code improvements.
| 1
| 1,760,719,274.69696
|
https://www.phoronix.com/news/Linux-s2idle-More-AMD-Lenovo
|
Linux Workaround Coming For Better s2idle Resume On More AMD Lenovo Laptops
|
Michael Larabel
|
Going along with many recent s2idle (suspend to idle) fixes as well as other fixes/workarounds/improvements like around S0ix, a patch is pending as a fix/workaround to get s2idle behaving correctly -- or rather, more timely -- on more AMD Ryzen powered Lenovo laptops.
Mario Limonciello of AMD's Linux client team who in recent months has been working on many suspend/resume and power management fixes for AMD Ryzen laptops has continued this quest with the latest patch now pending as a "fix" within the x86 platform drivers subsystem.
The patch was queued this week into the "fixes" branch by x86 platform drivers subsystem maintainer Hans de Goede. The code change is a s2idle resume quirk for various ThinkPad laptops. This should fix s2idle behavior on various AMD Ryzen Lenovo ThinkPads -- or making the resume happen much more timely. The workaround is coming for what equates to problematic system firmware.
Limonciello explained in the patch, "Lenovo laptops that contain NVME SSDs across a variety of generations have
trouble resuming from suspend to idle when the IOMMU translation layer is active for the NVME storage device. This generally manifests as a large resume delay or page faults. These delays and page faults occur as a result of a Lenovo BIOS specific SMI that runs during the D3->D0 transition on NVME devices...Create a quirk that will run early in the resume process to prevent this SMI from running. As any of these machines are fixed, they can be peeled back from this quirk or narrowed down to individual firmware versions."
This should fix s2idle resume handling for Lenovo laptops such as the L14 Gen2 AMD, T14s Gen2 AMD, X13 Gen2 AMD, T14 Gen2 AMD, T14 Gen1 AMD, T14s Gen1 AMD, P14s Gen1 AMD, P14s Gen2 AMD, and P14s Gen2 AMD models. Given it's in the "fixes" branch, it should be sent in as part of the next batch of platform-drivers-x86 fixes for the in-development Linux 5.18 kernel.
| 15
| 1,760,719,276.211488
|
https://www.phoronix.com/news/Kripke-AMD-HIP
|
LLNL's Kripke Ported To AMD HIP With More HPC Software Seeing Radeon/Instinct Support
|
Michael Larabel
|
Lawrence Livermore National Laboratory now has their Kripke software ported to running on AMD's HIP for GPU acceleration.
Kripke as a simple, scalable, 3D Sn deterministic particle transport code solver is up to version 1.2.5 today and introduces AMD HIP support as an alternative to its existing NVIDIA CUDA support. The Heterogeneous-Computing Interface for Portability (HIP) is AMD's means of easing conversion of traditionally NVIDIA CUDA applications over to portable C++ code for execution on AMD graphics processors.
This LLNL open-source project has added AMD HIP support by way of their RAJA software library that serves as an abstraction layer for HPC software. Kripke has supported NVIDIA CUDA for a while and also OpenMP/MPI for CPU execution while finally there is AMD accelerated support with this native HIP support.
AMD HIP adoption was off to a slow start but in more recent times with AMD's increasingly competitive graphics hardware and more supercomputer wins, AMD HIP support in the upstream code-bases of more open-source HPC projects has been increasing. The AMD HIP software stack as part of ROCm continues maturing too and new projects come about like Orochi for continuing to improve the AMD GPU compute story.
Should Kripke be of interest to you, check out the new release with HIP support at GitHub.
LLNL, El Capitan Lawrence Livermore National Laboratory continues their AMD support bring-up as part of preparing for the El Capitan supercomputer expected to go online next year with performance expected to be about 2 ExaFLOPS using AMD EPYC "Zen 4" CPUs with next-generation Radeon Instinct GPUs.
| 5
| 1,760,719,276.970594
|
https://www.phoronix.com/news/AMD-PerfMonV2-V2
|
AMD Sends Out Updated Linux Patches For PerfMonV2 That's Expected With Zen 4
|
Michael Larabel
|
AMD's open-source Linux engineers on the CPU side of the house continue being quite busy with all sorts of new feature enablement work, which given their timing and other factors is almost all definitively for upcoming Zen 4. AMD this week sent out updated patches in getting "PerfMonV2" support in order that is updated performance monitoring abilities with upcoming processors.
Back in March I wrote about PerfMonV2 back when AMD originally published the Linux support patches. This Performance Monitoring Version Two is for new/updated hardware performance counter abilities with a "new generation" of AMD CPUs and ensuring this new behavior is supported under Linux. Hardware performance counters under Linux via the perf subsystem is becoming increasingly used by a variety of parties for cases from debugging to using the performance counters for feedback/profile-driven compiler optimizations in the name of being able to achieve greater performance.
New with AMD Performance Monitoring V2 is the addition of "global" registers that allow enabling/disabling multiple performance counters at the same time. With the AMD Performance Monitoring up to this point, the different performance counter controls all had to be set individually while now can be set easily in one-go using the global registers (i.e. a single MSR write).
AMD PerfMonV2 also adds new core PMU features for the dynamic detection of the number of available PMCs rather than expecting a static set number in the code.
These latest AMD PerfMonV2 patches can be found on the kernel mailing list. If all goes well these patches could be mainlined in the v5.19 kernel and thus still supported by mainline ahead of Zen 4 processors launching later in the calendar year.
| 0
| 1,760,719,278.304055
|
https://www.phoronix.com/news/Parallel-CPU-Bringup-AMD-Snag
|
Faster Booting Via Parallel CPU Bringup Hits A Snag With Older AMD CPUs
|
Michael Larabel
|
At the end of last year you may recall the talked about Linux kernel patches for booting systems faster by allowing the parallel bring-up of secondary CPU cores. It's been a while since hearing much about that effort but seems to have hit a snag in that the code is running into problems on early Zen CPUs and older.
Going back to last December when that latest code was posted for supporting parallel CPU bring-up on x86_64, there were reports of problems for some AMD hardware. However, debugging the issue is a challenge with the system getting stuck early on in the boot process with minimal information available and completely crashes/hangs the system.
Finally today is an update on that matter - confirming from additional testing that Zen/Zen+ CPUs appear affected by this issue with parallel CPU bring-up while newer Zen 2 and Zen 3 processors appear to behave fine. Debugging the Zen/Zen+ issues remains a challenge due to failing quickly and not getting any output from the system at boot. Besides Zen/Zen+ being botched with this parallel CPU boot code, previously it was also reported pre-Zen APUs/CPUs having trouble too.
So while this CPU parallel boot support for Intel/AMD x86_64 is promising especially for significant boot time savings with higher core count systems, given the lack of notable progress in recent months it doesn't appear to be much of a priority and also remains complicated by this AMD issue and going through the mailing list there may be some other quirks/issues too. Hopefully they'll be able to sort through these problems otherwise if/when these patches move forward could end up having to blacklist/whitelist CPUs for enjoying this functionality.
| 7
| 1,760,719,278.952298
|
https://www.phoronix.com/news/AMD-SMCA-Syndrome-FRU-Text
|
AMD Continues With MCE/SMCA Linux Driver Changes Ahead Of Zen 4 CPUs
|
Michael Larabel
|
This year AMD engineers working on hardware enablement for Linux have been busy with EDAC driver improvements like RDDR5 and LRDDR5 handling, AMD Scalable Machine Check Architecture (SMCA) additions for "future" CPUs, and the various other areas outside of the error detection and correction field. Today though is a new patch series back in that hardware error handling space with new SMCA code.
A new patch series posted on Monday for the AMD MCE (Machine Check Exception) driver adds support for two new "syndrome" registers used in "future AMD Scalable MCA systems" and as part of that implementing a new FRU Text feature. Given the timing of this work and AMD's cadence around Linux hardware enablement timing, this is almost certainly for EPYC 7004 "Genoa" and "Bergamo" server processors.
AMD engineers remain very busy working on Linux support ahead of Zen 4 processors launching later this year. The intention with the new syndrome registers to be found as part of the SMCA IP with future AMD CPUs is for providing supplemental error information. The FRU text feature is for a Field Replaceable Unit (FRU) string that is represented in the new syndrome registers. The FRU text string can vary based on MCA bank and is populated dynamically for each error state. This FRU string will be included as part of all AMD MCE reports for hardware errors.
The new AMD MCE driver patches are now out for review on the kernel mailing list and given the timing could be merged for the v5.19 cycle if no issues turn up. Long story short, this is another patch series pointing at the seemingly more than usual hardware error detection/reporting changes coming for next-generation EPYC server processors and all should be welcomed improvements by server administrators for helping to deal with any hardware/system issues.
| 2
| 1,760,719,280.205648
|
https://www.phoronix.com/news/AMD-HIP-RT
|
AMD Launches HIP-RT Ray-Tracing Library
|
Michael Larabel
|
The newest software addition under AMD's GPUOpen software umbrella is HIP-RT as a ray-tracing library for HIP.
HIP-RT is a ray-tracing library that makes it easy to write ray-tracing software against HIP and easy integration for existing HIP code-bases.
This isn't just trying to replicate existing NVIDIA RTX/OptiX interfaces or the like but is said to take a slightly different design approach, "Although there are a few other ray tracing APIs which introduce many new aspects, we designed HIP RT in a slightly different way by eliminating the need to learn many new kernel types. HIP RT introduces new object types like hiprtGeometry and hiprtScene. Once the geometric information is passed to HIP RT, the process builds the data structure, which in turn is passed to the HIP kernel. At this stage, the device-side library API can be used to perform an intersection test. Current generation graphics cards, such as AMD RDNA 2 architecture-based GPUs, support hardware ray tracing acceleration to further optimize render times. However, up until now, applications that support HIP have not been able to utilize this hardware acceleration. HIP RT is designed to allow developers to take full advantage of the Ray Accelerators used for hardware ray tracing in AMD RDNA 2 architecture-based GPUs."
HIP-RT is available on GPUOpen.com. This library has been tested on Navi 1x/2x and Vega 1x/2x class GPUs. HIP RT can also run on NVIDIA GPUs via CUDA APIs but the hardware accelerated ray-tracing only works for RDNA2 GPUs.
More details on the HIP-RT Radeon ray-tracing library via today's announcement. HIP-RT is available under an MIT license.
| 16
| 1,760,719,280.60899
|
https://www.phoronix.com/news/AMD-SEV-SNP-TIP-Linux-5.19
|
AMD SEV-SNP Appears Ready For Upstreaming In Linux 5.19
|
Michael Larabel
|
Introduced last year with the AMD EPYC 7003 "Milan" processors was SEV-SNP as the latest iteration of their Secure Encrypted Virtualization technology. SEV-SNP adds additional integrity protections and safeguards as part of this "Secure Nested Paging" extension of SEV. Finally with Linux 5.19 the SEV-SNP support should premiere in the mainline kernel.
Right after the EPYC 7003 series was introduced in March 2021, AMD began sending out Linux patches for SEV-SNP. They continued over the past year sending updated versions of the patches to implement more of the supported functionality and addressing developer feedback. Going back to the hardware launch they've also hosted this code in a GitHub repository for AMD EPYC customers wanting to use their supported kernel build / patches for building a kernel with this functionality in place beyond the SEV/SEV-ES code already in upstream. The hardware-based memory integrity protections can help prevent malicious hypervisor-based attacks and other functionality beyond what is available with Secure Encrypted Virtualization of prior EPYC CPUs.
After going through a number of rounds of review (12+), it looks like the initial SEV-SNP code with working guest OS support is ready for what will land in the Linux 5.19 kernel this summer.
An AMD table comparing SEV, SEV-ES "Encrypted State", and SEV-SNP "Secure Nested Paging" that has been building up on succeeding generations of EPYC processors. Queued this morning into TIP's x86/sev branch is the AMD SEV-SNP guest driver and other SEV-SNP feature code. With this code now part of TIP.git, it's the last stepping stone prior to being sent in next merge window as part of the various TIP feature pull requests... This puts it hitting mainline around the end of May once the Linux 5.19 cycle gets started.
This is the whole SEV-SNP guest patch series that had been carried via AMDESE's sev-snp-v12 GitHub branch and now part of TIP x86/sev until the Linux 5.19 cycle rolls around -- barring any last minute issues coming up that would lead to these patches being dropped from TIP / x86/sev. There does still appear to be areas for improvement with this SEV-SNP code such as supporting the "lazy validation" mode for pages where as now it's all done under pre-validation. Interrupt security is another SEV-SNP feature still to be addressed too with future patches.
More details on AMD SEV-SNP protections with EPYC 7003 series processors can be found via the AMD whitepaper. It's unfortunate it has taken more than one year after launch for this SEV-SNP functionality to make it into a mainline kernel, but at least it's looking like that initial support will be all in good shape this summer and mainlined in time for the autumn Linux distribution updates and ahead of next-gen EPYC processors releasing.
| 0
| 1,760,719,282.212216
|
https://www.phoronix.com/news/Zen-1-To-Zen-3-2022-AMD-ucode
|
Updated AMD Zen 1 Through Zen 3 CPU Microcode Published
|
Michael Larabel
|
On Friday AMD published new CPU microcode files for both Family 17h and Family 19h for Zen 1/2/3 processors. At the moment there isn't any public insight into the changes with this updated microcode but it may be significant.
AMD on Friday published new CPU microcode files for both Family 17h (Zen / Zen+ / Zen 2) and Family 19h (Zen 3). Sadly as is normal, the microcode binary update comes without any public change-log.
But it does seem a bit different this time around with Family 17h being included too... We see Family 19h CPU microcode updates every few months but it's been a while since the last Family 17h update. The AMD Family 17h CPU microcode included in linux-firmware.git was last updated in December 2019. Not only has it been a long time since the last in-tree update, but this new microcode is much larger. The Family 17h microcode goes from 6476 to 9700 bytes in size.
The Family 19h microcode meanwhile is the same size as its prior update from back in February.
Raising my curiosity/concerns about this AMD microcode update is that it's the first Zen/Zen2 CPU microcode update for linux-firmware.git since the end of 2019 and this update is much larger than prior versions of the microcode. The Zen 3 microcode is updated too but that has been happening every few months for the current-gen hardware. With no official announcement yet on the new CPU microcode we are left to only speculate what may be changed in this version... But given the Family 17h conditions it's likely one of two: just a lot of routine fixes/updates included over the past 2+ years and only now deciding to bundle it up into a new version or the new CPU microcodes are being published now with some security mitigation(s). It very well could be the latter, but we'll see soon if that's the case.
For now the new AMD Zen CPU microcode is on the mailing list but should soon make it into linux-firmware.git and in turn pulled by the various Linux distributions for packaging up nicely for end-users.
| 24
| 1,760,719,282.876301
|
https://www.phoronix.com/news/AMD-MWAIT-Over-HALT-2022
|
AMD Patch To Use MWAIT Instead Of HALT For Certain Cases Yield A ~21% Improvement
|
Michael Larabel
|
As a Linux kernel change for benefiting AMD CPUs going back to Zen 1 and for matching behavior Intel has had in place since the Core 2 times, AMD submitted a patch for having the Linux kernel use the MWAIT instruction instead of HALT for when the system isn't using the CPU idle driver either for C-states being disabled by the BIOS or the driver not part of the kernel build. In turn this can lead to around a 21% improvement in exit latency on affected systems.
The Monitor Wait "MWAIT" instruction can be used for power management purposes to hint that the processor can enter a specified target C state while waiting for an event or a MONITOR'ed store operation to complete. Usage of MWAIT is intended to be more efficient than the HALT instruction.
Intel Core 2 and later have already preferred MWAIT over HALT instruction for entering the C1 state and that has been the behavior in the Linux kernel. AMD CPUs though continue using HALT in this code path for cases like not using the CPU idle driver (either due to being disabled or not part of the kernel build) or when C-states are disabled by the system BIOS, so admittedly somewhat niche.
The patch proposed this week by AMD now switches over to using MWAIT for all AMD Zen processors for Zen 1 and newer plus Hygon Dhyana too. On AMD Zen 3 (and similar results for older Zen too) the exit latency was found to be around 21% lower using MWAIT rather than HALT. Most users will find themselves using the CPU idle driver and thus unaffected by this patch, but for those that aren't this change can be beneficial. The high latency with HALT had apparently come up in real-world testing/usage when exploring packet loss in an unspecified network program as part of "customer performance observations".
This AMD change is currently under review on the kernel mailing list.
| 8
| 1,760,719,284.783417
|
https://www.phoronix.com/news/AMD-Linux-x2AVIC-Patches
|
AMD Readies Linux Patches For x2AVIC Support
|
Michael Larabel
|
Back in February AMD engineers sent out a "request for comments" on x2APIC virtualization (x2AVIC) support. Those patches have now been refined beyond the RFC state and sent out today as a patch series for further review on the Linux kernel mailing list.
With AMD's Advanced Virtual Interrupt Controller (AVIC) code for the past number of years, x2APIC has not been supported concurrently. Now with x2AVIC it overcomes that limitation for interrupt virtualization with x2APIC enabled. X2APIC is particularly important for supporting larger vCPU counts ahead of AMD's next-generation server processors with EPYC Genoa expected to have 96 cores / 192 threads per socket and EPYC Bergamo up to 128 cores / 256 threads per socket... Or even further out, AMD EPYC Turin rumored to have 256 cores / 512 threads per socket. The x2AVIC code has been successfully tested now by AMD engineers with more than 255 vCPUs active, verifying its functionality up to 512 vCPUs for now with their internal hardware.
Current generation EPYC 7003 "Milan" processors offer up to 64 cores / 128 threads per socket.
Previously, with AVIC, guest needs to disable x2APIC capability and can only run in APIC mode to activate hardware-accelerated interrupt virtualization. With x2AVIC, guest can run in x2APIC mode. This feature is indicated by the CPUID Fn8000_000A EDX[14], and it can be activated by setting bit 31 (enable AVIC) and bit 30 (x2APIC mode) of VMCB offset 60h.
The mode of interrupt virtualization can dynamically change during runtime. For example, when AVIC is enabled, the hypervisor currently keeps track of the AVIC activation and set the VMCB bit 31 accordingly. With x2AVIC, the guest OS can also switch between APIC and x2APIC modes during runtime. The kvm_amd driver needs to also keep track and set the VMCB bit 30 accordingly.
More details on AMD's x2APIC virtualization support for Linux KVM usage via this patch series.
| 0
| 1,760,719,284.866694
|
https://www.phoronix.com/news/AMD-Branch-Sampling-v5.19
|
AMD Branch Sampling "BRS" Feature To Land With Linux 5.19
|
Michael Larabel
|
While there are many new features with Linux 5.18 with its merge window having just ended days ago, feature code is already beginning to accumulate within the various "-next" branches for what will be Linux 5.19 this summer. Patches merged today get AMD Branch Sampling (BRS) functionality in place for Zen 3 processors with that next kernel cycle.
Last year I wrote about Google engineers working on AMD Branch Sampling support for Linux with Zen 3 processors. This is about making use of the processor-exposed information about branches taken during code execution. In turn this AMD Branch Sampling information will be exposed via Linux's perf subsystem for collection. This AMD Branch Sampling support can be leveraged for better compiler optimizations on AMD processors by analyzing the collected hardware data for making more informed/accurate optimization decisions like with AutoFDO. See the aforelinked article for more commentary.
While it's been some months since last talking about the AMD Branch Sampling support, based on the information within the public AMD programmer reference manual they were able to get the support ironed out and is now ready for mainlining. After re-basing against Linux 5.18-rc1, pulled into TIP's perf/core branch is the AMD Zen 3 Branch Sampling code. In turn these perf/core patches along with whatever patches accumulate over the next month and a half will in turn be submitted for the Linux 5.19 merge window when that kicks off around the end of May.
With Zen 3 processors there is 16-deep branch sampling accessible via CPU MSR registers. With the perf tool the RETIRED_BRANCH_INSTRUCTIONS can be used for matching the BRS behavior or the new branch-brs event. The AMD Zen 3 BRS support is going to be an opt-in feature with the Linux CONFIG_PERF_EVENTS_AMD_BRS switch at build-time. It's great that AMD BRS is now being exposed under Linux but unfortunate it's taken so long after Zen 3 CPUs premiered with this feature, at least it should still prove useful with Zen 4 and beyond for profiling / AutoFDO-like optimizing.
| 0
| 1,760,719,286.196515
|
https://www.phoronix.com/news/AMD-Acquiring-Pensando
|
AMD Acquiring Pensando For $1.9B USD
|
Michael Larabel
|
It was just two months ago AMD completed its acquisition of Xilinx and now its newest data center play is entering into a definitive agreement to acquire Pensando.
Pensando is the five year old company focused on distributed computing solutions for software-defined cloud, compute, networking, and other areas. Pensando was founded by former Cisco CEO John Chambers and the company has been gaining ground over the past half-decade in the cloud/edge computing space. Now to bolster AMD's data center offerings, AMD is acquiring Pensando for about $1.9 billion (USD).
AMD expects the Pensando acquisition to close in Q2.
More details on this acquisition announced this morning at AMD.com.
| 17
| 1,760,719,286.757266
|
https://www.phoronix.com/news/AMD-Orochi
|
AMD Announces "Orochi" For HIP/CUDA Run-Time Handling
|
Michael Larabel
|
AMD's GPUOpen team today announced "Orochi" as their latest open-source software offering in the HIP GPU compute space.
While HIP has eased the process for developers to port their traditional NVIDIA CUDA code over to AMD's ROCm stack for GPU execution on Radeon / Instinct hardware, that has resulted in a separate HIP backend from CUDA.
AMD Orochi. This should be great for helping more robust CUDA/HIP support within packaged applications, assuming Orochi's adoption takes off. AMD Orochi aims to allow NVIDIA CUDA and AMD HIP support to exist within a single code-base and binaries. Orochi provides a library that loads the HIP and CUDA driver APIs dynamically at run-time. Therefore a single application binary can work on both AMD GPUs and NVIDIA GPUs.
Orochi is a library that loads HIP and CUDA® driver APIs dynamically at runtime. By switching the CUDA/HIP calls in your app to Orochi calls, you can compile a single executable that will run on both AMD and NVIDIA GPUs. This streamlines the deployment of your GPU accelerated applications and makes it easy to provide cross-platform support with a single API, thus eliminating the overhead of maintaining separate backends for each platform. We chose the name Orochi as it’s named after the legendary eight-headed Japanese dragon, which goes well with the purpose of our library – to allow a single library with multiple backends at runtime. AMD announced Orochi today on GPUOpen.com. The code is available via GitHub under an MIT license.
| 7
| 1,760,719,287.850392
|
https://www.phoronix.com/news/AMD-PSP-Sysfs-Expose
|
AMD To Expose More PSP Security Information Under Linux, Including State Of CPU Fuses
|
Michael Larabel
|
Right now under Linux it isn't quick and easy to figure out if the likes of (Transparent) Secure Memory Encryption are enabled and working but a new patch series will more easily expose the security attributes of the AMD Platform Security Processor (PSP) to users on Linux. Among the information to be exposed will also include whether the CPU is fused in the name of tampering prevention.
AMD Linux engineer Mario Limonciello has been working on a patch series for exporting various AMD PSP security attributes under Linux and exposing that information to user-space via sysfs.
Among the information to be exposed via sysfs includes whether the CPU/APU is a fused part to prevent tampering but limits the CPU to working in certain system vendor motherboards (Platform Secure Boot) with effectively vendor-locking that given part. The sysfs information also indicates whether the CPU/APU is unlocked for debugging purposes, the TSME state, whether the PSP is enforcing rollback protection, the status of the Replay Protected Monotonic Counter (RPMC), whether the HSP TPM is acxtivated, and whether RomArmor SPI protection is enforced. This work is only about reporting the state of these various PSP features and doesn't allow altering their value/behavior.
This current patch series allows the ability to detect Secure Memory Encryption (SME) and Transparent Secure Memory Encryption (TSME) too and possibly expanding that in the future so both wouldn't be redundantly enabled at the same time, but now at least the user will know.
The AMD PSP information will be exported under /sys/bus/pci/devices/. This information reporting is being handled by AMD's CCP (Crypto Co-Processor) driver. This patch series thus also now allows the AMD CCP Linux driver to load even for CPUs without SEV/TEE. AMD's Platform Security Processor is the Arm core inserted onto the CPU die with on-chip firmware that is responsible for various security responsibilities on Ryzen and EPYC systems.
For now the patches are on the kernel mailing list while hopefully they will be readied in time for the v5.19 cycle this summer for this useful AMD PSP information reporting.
| 3
| 1,760,719,289.399537
|
https://www.phoronix.com/news/AMD-More-Linux-CXL-Debug
|
AMD Recruiting More Linux Engineers For Debug, CXL Enablement & More
|
Michael Larabel
|
I was informed that AMD has a few more Linux positions open at the company. While they have in past years been rather nimble with their Linux staffing, things continue to change thanks to their ongoing successes in the marketplace from the consumer side with Steam Deck through the likes of Tesla's infotainment system up through high-end server platforms.
AMD over particularly the past 1~2 years has begun significantly ramping up their Linux hires thanks to EPYC servers being well received in the marketplace along with their supercomputer design wins and most of the servers these days running Linux. Then over the past year they have been focusing more on the Linux client side too for Ryzen and embedded SoC designs with increasing Linux usage there too.
In addition to some of their Linux job postings from earlier this month still being unfilled, they have also opened up some new positions.
Notable with their latest batch of Linux openings is hiring a CXL engineer. This role will focus on AMD's support for Compute Express Link (CXL) hardware enablement under Linux. Intel so far has been leading the development of the Linux kernel's Compute Express Link subsystem. Over the past number of kernels with the still maturing CXL subsystem Intel engineers have been responsible for nearly all of that enablement work for this new open standard. AMD has been part of the CXL organization so it's good to see them now (albeit late) hiring to work on CXL for Linux too.
AMD's newly released EPYC Milan-X processors are a fantastic performer under Linux already but certainly eager to see more timely contributions to the Linux kernel for next-gen platforms and relentless optimizing for new and existing hardware, especially considering Intel's vast open-source/Linux talent pool and their massive scale of open-source work across the board.
Over on the client side, another new job opening is for a client Linux platform debug lead. This will focus on desktop/notebook platform debugging for Linux. Great to see they'll have an additional person helping to work through Linux client bugs, including mentioned in the job posting are around power management / system standby -- an area where some modern AMD Ryzen notebooks have had issues on Linux with s2idle, etc, but more recent kernels have begun working through those problems.
There remains a number of other open Linux engineering positions too. Those on the hunt can see the AMD Linux job board.
| 1
| 1,760,719,289.715311
|
https://www.phoronix.com/news/AMD-Easier-P-State-Usage
|
AMD Making It Easier To Switch To Their New P-State CPU Frequency Scaling Driver
|
Michael Larabel
|
Following our how-to guide for enabling the new AMD P-State driver that premiered in Linux 5.17 after finding many users were unsure to go about using this new CPU frequency scaling driver, AMD is now making it easier to switch from ACPI CPUFreq to AMD P-State.
As noted in the how-to article, while the likes of the Ubuntu Mainline Kernel PPA are building amd_pstate already as a kernel module, it's not trivial to switch to with ACPI CPUFreq taking precedence and getting in the way of loading the amd_pstate module. Using the "initcall_blacklist=acpi_cpufreq_init" kernel option detailed in the guide will block the ACPI CPUFreq driver from initializing and thus allowing the AMD P-State driver to go on and claim the supported AMD CPU. But with pending patches to the AMD P-State driver, that blacklisting won't be needed anymore for kernels building in ACPI CPUFreq but leaving AMD P-State as a module.
AMD engineer Mario Limonciello, who joined AMD's Linux client team last year, sent out a patch series improving the usability for this new driver. The patch series adds a new "replace" module option for AMD P-State to allow it to replace existing CPUFreq drivers when loaded. Additionally, another patch adds a module device table so the amd_pstate driver will auto-load if encountering an AMD CPU with ACPI CPPC supported (introduced with Zen 2 and a requirement for this driver).
So with these pending patches it would allow simply having amd_pstate.replace=1 as a kernel parameter to boot the system and use AMD P-State without blacklisting ACPI CPUFreq or rebuilding your kernel with this new driver built-in. This would also allow users to toss "options amd-pstate replace=1" into the /etc/modprobe.d directory on distributions like Ubuntu as another easy means of switching. Thus this convenient patch series is making things much simpler for the kernel builds out there having AMD P-State as a module and ultimately making it even easier for users to opt into using this new driver focused on better power efficiency.
This patch series isn't forcing AMD P-State by default or so, but we'll see if that comes up in a future kernel version once AMD P-State has proven itself. Right now these patches are residing just on the kernel mailing list while we'll see if they get picked up in time to try to make it into v5.18 or is held off until the v5.19 cycle.
| 3
| 1,760,719,291.062926
|
https://www.phoronix.com/news/Linux-5.18-EDAC
|
Linux 5.18 EDAC Continues Making Preparations For AMD Zen 4
|
Michael Larabel
|
AMD continues improving their Error Detection And Correction (EDAC) driver code within the Linux kernel ahead of next-generation processors debuting later this year.
The Linux 5.17 AMD EDAC changes included adding support for Registered DDR5 (RDDR5) and Load-Reduced DDR5 (LRDDR5) system memory and adding in support for new Family 19h models.
With Linux 5.18 there is more AMD EDAC work still focused around these new AMD Family 19h models. There has been some disagreeing with my assessment that AMD Zen 4 will still be Family 19h, but the patch work continues to be pointing to that being the case. Zen 4 looks like it will be in Family 19h -- the same as Zen 3 -- and similar to how Zen / Zen+ / Zen 2 spanned Family 17h.
While these patches do not spell out Zen 4, these model IDs don't match existing AMD CPUs including the newly released and exciting Milan-X parts. The AMD EDAC changes for Linux 5.18 add the ability to set the memory type per DIMM -- not for say DDR5 and DDR4, but the likes of LRDDR5 / DDR5 / RDDR5 memory types across different memory controllers. The other work is setting a new "family flags" bitmask and using new offsets for AMD Family 19h Model 10h. A new bitfield is added for indicating DDR5 memory use on a memory controller.
The Linux kernel continues preparing AMD code paths around DDR5 support for next-generation platforms.
Besides the EDAC work, there has been thermal monitoring and other additions in recent and ongoing kernels ahead of AMD's next-gen hardware launches. See the EDAC pull request for more details on the changes that were merged on the Error Detection and Correction front for Linux 5.18.
| 5
| 1,760,719,291.349824
|
https://www.phoronix.com/news/AMD-P-State-How-To
|
How To Use The New AMD P-State Driver With Linux 5.17
|
Michael Larabel
|
Since the release of the Linux 5.17 kernel the leading question in my inbox has been from readers asking how to actually make use of the AMD P-State driver. Right now this driver isn't the default over ACPI CPUFreq and I haven't seen any Linux distribution vendors announce their plans to immediately default to this new driver, but over the months ahead I expect that to change. In any case, if wanting to use amd_pstate on Linux 5.17 today here is a brief how-to guide for making the transition.
First up, the easiest way to check to verify your CPU frequency scaling driver is by reading /sys/devices/system/cpu/cpu0/cpufreq/scaling_driver to see if you are on acpi-cpufreq or amd-pstate. As a reminder, AMD P-State is only supported with AMD Zen 2 processors and newer. AMD P-State depends upon ACPI Collaborative Processor Performance Contols (CPPC) and that is only with Zen 2 and newer -- Zen 1 / Zen 1+ will not be seeing any support. You also need to ensure CPPC is enabled in your BIOS in case it happened to be disabled. Many motherboard vendors do have a "ACPI CPPC" option in the BIOS that can be set to enabled or auto on Zen 2 and newer processors.
This guide was written making use of Ubuntu with their Ubuntu Mainline Kernel PPA kernel builds. ACPI CPUFreq is the default still with their Linux 5.17+ kernel builds but the AMD P-State module is included as part of the kernel builds. Most other distro kernel builds will at least likely have the CONFIG_X86_AMD_PSTATE=m option enabled so it's at least built. If you don't see CONFIG_X86_AMD_PSTATE and it's commented out or "n", you would need to first rebuild your kernel so that driver is included.
On most (all?) distribution kernels the ACPI CPUFreq driver is built into the kernel and not a module, so simply blacklisting it will not work. To avoid loading ACPI CPUFreq even if it's built-in, the "initcall_blacklist=acpi_cpufreq_init" option can be set in your /boot/grub/grub.cfg or respective place for editing your kernel command line options at boot (e.g. "/etc/default/grub" for a more permanent solution). This initcall_blacklist is needed to prevent the ACPI CPUFreq initialization function from being called and thus then allowing AMD P-State to successfully initialize. For Ubuntu/initramfs users wanting to have AMD P-State loaded by default, you can also add "amd_pstate" to the "/etc/initramfs-tools/modules " file followed by running update-initramfs -u.
After a reboot with those changes made to blacklist the ACPI CPUFreq driver and to load the AMD P-State driver, you should then be on that new AMD CPU frequency scaling driver code path. Again, you can check /sys/devices/system/cpu/cpu0/cpufreq/scaling_driver to verify. If you are on the new driver, it should print amd-pstate. If that file isn't present you may have successfully black-listed ACPI CPUFreq but AMD P-State didn't load yet in which case you can try modprobe amd-pstate.
If AMD P-State still isn't working, you may need to add amd_pstate.shared_mem=1 as part of your kernel command line configuration. With Linux 5.17 some systems require using the "shared_mem" module option for AMD P-State to correctly function with some hardware. Hopefully soon that limitation will be addressed to not require setting that extra option.
That's the quick overview if left wondering how to enable AMD P-State on Linux 5.17. If any questions or issues stop by our forums / commenting on this article. I'll have up some new AMD P-State benchmarks off Linux 5.17 final shortly. Keep in mind though this driver is initially focused on Ryzen mobile APUs like the Steam Deck while moving forward expect to see more desktop (and perhaps AMD EPYC server) tuning.
| 35
| 1,760,719,292.901724
|
https://www.phoronix.com/news/AMD-MI200-ISA-Docs
|
AMD Publishes New Instinct MI200 Instruction Set Documentation
|
Michael Larabel
|
AMD quietly posted a new version of its instruction set architecture documentation concerning its Instinct MI200 accelerator. AMD originally published the ISA documentation for the MI200 back in November but it seems to have gone unnoticed (including by me) while in February they went ahead and released a new version of that technical documentation.
In the early days of AMD's open-source Linux graphics driver effort they accompanied their driver source code with GPU documentation releases too for helping the community and other parties bring-up their original graphics driver stack. There were thousand and thousands of pages of GPU technical documentation published by AMD for helping this open-source driver effort. However, with time they have focused less on the documentation and more on quality open-source code. The documentation clearing for public release is quite time consuming especially for going through the legal and technical review processes to get those official docs out there in the public light. AMD has found that time better spent working on the open-source Linux driver stack itself for ensuring its good quality and trying to ensure its code is fairly documented and also having complete register header files in the driver as opposed to assembling documentation on the registers.
One area though where AMD has continued with publishing proper GPU documentation is on the actual graphics processor instruction set. As properly understanding the ISA is of benefit to many external parties, that public documentation effort continues.
Noted this week in LLVM Git was reference to the AMD MI200 documentation being published.
The docs referenced are marked as published in February 2022 for the AMD Instinct MI200. It seems that documentation push went unnoticed/unannounced as I haven't seen it mentioned anywhere yet besides when spotting the LLVM Git change. Looking around further, it appears there was also the original AMD Instinct MI200 ISA documentation published back in November that too went without fanfare. There isn't any change-log for the alterations between the November and February documentation revisions.
But in any case for those interested in learning more technical details about the CDNA-based AMD Instinct MI200 accelerator, see the AMD reference guide (PDF). This ISA guide comes in at 275 pages.
| 2
| 1,760,719,293.237635
|
https://www.phoronix.com/news/AMD-Joins-AlmaLinux
|
AMD Now Backing AlmaLinux As This Increasingly Popular RHEL/CentOS Alternative
|
Michael Larabel
|
AMD is now among the latest companies backing the AlmaLinux OS Foundation for that increasingly popular free build derived from the Red Hat Enterprise Linux sources now that CentOS 8 is end-of-life.
AlmaLinux as the Linux distribution started from CloudLinux for providing binary-compatibility with upstream Red Hat Enterprise Linux (RHEL) is approaching the one year anniversary since its first stable release. Ahead of that 30 March anniversary, the foundation announced today a new set of corporate members. The new backers for AlmaLinux include Sine Nomine Associates, BlackHOST, Knownhost, and arguably most notable is AMD.
Other existing backers of AlmaLinux include the likes of Amazon AWS, Microsoft Azure, Arm, Equinix, Hivelocity, and various other hosting organizations and companies. Notably absent from the current backers list is Intel.
AlmaLinux is finding increasing use by public cloud service providers, dedicated web hosting servers, and other environments where formerly CentOS had a stronghold until the Red Hat decision to premature EOL CentOS 8 at the end of 2021. Given AMD's market successes with EPYC in the data center, it doesn't come as much of a surprise to see them now backing this no-cost alternative to RHEL.
Backers to the AlmaLinux OS Foundation provide funding for its continued development of the operating system and help drive its direction by voting and participating within the foundation's board of directors.
More details on the new and existing AlmaLinux OS Foundation members and other information on this CentOS/RHEL alternative via AlmaLinux.org.
AMD has also backed other Linux operating systems over the years ranging from being a longtime sponsor of openSUSE (and close partner of SUSE) to the failed efforts around MeeGo Linux, among others.
| 19
| 1,760,719,294.558593
|
https://www.phoronix.com/news/AMD-PerfMonV2-Linux-Patches
|
AMD Preparing Linux "PerfMonV2" Support In Preparation For Zen 4
|
Michael Larabel
|
The latest Linux kernel patch activity out of AMD in preparation for next-generation "Zen 4" processors is enabling AMD Performance Monitoring Version Two "PerfMonV2" support.
AMD's Linux kernel activity around "upcoming" and "new generation" processors continues ramping up over the past number of weeks. Given the timing and AMD not at a stage of pushing out patches publicly for processor features multiple generations in advance, it basically means Zen 4.
This morning's set of Linux kernel patches is for changes enabling AMD Performance Monitoring V2 capabilities. This is about enabling support for AMD's performance counters under Linux. For years they have allowed their performance counters to be exposed through the perf subsystem and now are preparing PerfMonV2 in advance of future CPUs.
The PerfMonV2 support is checked based on a new CPU feature bit, but given the timing is surely for the Zen 4 processors coming down the pipe in a few months. AMD Performance Monitoring V2 has new "global" registers to allow enabling/disabling multiple performance counters at the same time. With the AMD Performance Monitoring up to this point, the different performance counter controls all had to be set individually while now can be set easily in one go using the global registers where present.
AMD Performance Monitoring V2 also allows for systematically detecting the number of core PMCs rather than being statically set on a per-family basis. There is also a fix to the x86 KVM code as part of this series to avoid attempting to access an invalid MSR that results in KVM guest start-up breaking when detecting PerfMonV2. This patch should get back-ported to the Linux kernel stable series and properly fix things up in time for Zen 4 CPUs appearing.
That's it as far as these initial AMD PerfMonV2 enablement patches are concerned with no other features being exposed yet by this code work. See this patch series for this initial PerfMonV2 support which given the timing will see its first opportunity for mainlining come to the v5.19 cycle this summer.
| 1
| 1,760,719,295.186668
|
https://www.phoronix.com/news/AMD-5800X3D-New-CPUs
|
AMD Announces Ryzen 7 5800X3D Shipping On 4/20, New Mainstream CPUs
|
Michael Larabel
|
AMD today announced the ship date and suggested pricing for their much anticipated Ryzen 7 5800X3D processor as well as new Ryzen 7/5/3 series processors.
First up the AMD Ryzen 7 5800X3D processor is expected to be available globally on April 20 with an SEP of $449 USD. The Ryzen 7 5800X3D is their first desktop processor with AMD 3D V-Cache technology.
AMD also is announcing today new Ryzen 7 / 5 / 3 series processors and extended Ryzen 5000 series support to older AMD 300 Series chipsets.
The new processors announced today include the Zen 3 Ryzen 5700X / Ryzen 5 5600 / Ryzen 5 5500 processors and then the Zen 2 based Ryzen 5 4600G / Ryzen 5 4500 / Ryzen 3 4100 processors. The Ryzen 3 7100 Zen 2 processor delivers four cores / eight threads with a 65 Watt TDP at $99 USD.
The newly announced processors should see availability starting 4 April.
For those with older AMD 300 series motherboards, updated BIOSes are expected to support the Ryzen 5000 series officially now across the X370 / B350 / A320 motherboards.
More details on today's CPU announcements via AMD.com. We'll hopefully have our hands on these new CPUs shortly for Linux benchmarking.
| 55
| 1,760,719,296.194259
|
https://www.phoronix.com/news/AMD-P-State-Tracer
|
AMD P-State Tracer Tool To Be Included With Linux 5.18
|
Michael Larabel
|
One of the most prominent additions to the Linux 5.17 kernel is the introduction of the AMD P-State driver akin to Intel's P-State driver and aims to deliver better energy efficiency than AMD Zen 2 and newer processors currently on the ACPI CPUFreq driver. With Linux 5.18 an AMD P-State tracer tool is to be included with the kernel source tree for helping to analyze and tune this new driver.
The AMD P-State driver makes use of ACPI Collaborative Processor Performance Controls (CPPC) found with Zen 2 and newer systems when exposed by the platform/BIOS. The new AMD P-State driver should really help with desktops/mobile such as with the Steam Deck, as Valve did work on this driver collaboratively with AMD over the course of the past year.
The Linux 5.18 source tree is picking up the "amd_pstate_tracer.py" tool that is used for recording and parsing amd-pstate driver trace logs. The intent is to use this new tool for debugging and tuning the driver.
While not part of the kernel per se, amd_pstate_tracer.py is like a number of other tools that live within the kernel source tree. Those interested will eb able to find this tool within linux/tools/power/x86/amd_pstate_tracer. As root this Python script can record traces looking at the performance behavior of P-State and then report details how much time was spent in each performance state and other details on the behavior.
In the process of introducing the AMD P-State tracer, this also imports/updates the Intel P-State Tracer code already within the kernel for which AMD is then building atop.
This tool along with adding new trace events to the amd_pstate driver itself were picked up a few days ago via the Linux power management's linux-next branch ahead of the Linux 5.18 merge window opening up. The Linux 5.18 merge window should kick off tomorrow if Linux 5.17 goes ahead and releases today and isn't delayed to next week.
| 7
| 1,760,719,296.753934
|
https://www.phoronix.com/news/AMD-Linux-UAI-Zen-4-Tagging
|
AMD Posts New Linux Code For Zen 4's UAI Feature
|
Michael Larabel
|
AMD posted this morning a new Linux kernel patch series for enabling a new feature for "upcoming processors" that is almost definitively for Zen 4, continuing their work in recent weeks around more open-source patches in preparing for their next-generation processors.
The patch series sent out today, which was marked as a "request for comments" (RFC), is making use of the Upper Address Ignore (UAI) functionality found with upcoming AMD processors in order to implement user-space address tagging functionality.
The AMD64 Architecture Programmer's Manual updated back in November did lay out User Address Ignore but obviously didn't specify when it will appear in future AMD processors. With UAI patches now out for the Linux kernel, it pretty much guarantees that it's coming with Zen 4... AMD isn't yet at the stage where they are working publicly on Linux kernel features for two generations ahead.
AMD's programmer manual previously outlined the Upper Address Ignore feature. Now they are working on the first planned usage of UAI for Linux.
The AMD Upper Address Ignore functionality can be used in a similar manner to Arm's Memory Tagging Extension (MTE) that was added with Armv8.5-A and more broadly as part of AArch64 is the Top Byte Ignore (TBI) support. In Arm's MTE case is primarily intended as a memory safety feature while with AMD UAI is mainly open-ended tagging with otherwise unused bits of a memory address like AArch64 TBI.
Intel's ISA programming reference manual previously outlined their Linear Address Masking (LAM) functionality similar in nature to AMD UAI.
Intel for their part has also been working on something similar. Since 2020 I've noted how Intel has been working on Linear Address Masking (LAM). Linear Address Masking is about making use of the untranslated address bits of 64-bit linear addresses for metadata. Intel LAM is coming with future processors and the Linux kernel usage around that is still being finalized with some overlap to today's AMD UAI patches.
With the patches proposed today, AMD is looking to make use of UAI for user address tagging. The AMD patch-set builds upon the Intel LAM Linux patches for extending AArch64's SET_TAGGED_ADDR_CTRL/GET_TAGGED_ADDR_CTRL prctl() options for opening up the x86_64 address tagging to user-space. Memory tagging can be used for detecting memory bugs and other issues by tracking illegal memory operations, among other possible use-cases.
This AMD UAI support for Linux memory tagging has been "lightly tested" so far and out today under the RFC flag in seeking feedback from other kernel developers. See the patch series for this initial AMD work on user-space address tagging for the Linux kernel with their upcoming (Zen 4) processors.
| 7
| 1,760,719,298.334691
|
https://www.phoronix.com/news/AMD-Linux-Jobs-March-2022
|
AMD Posts Some New Linux Job Openings From Client CPU To Server
|
Michael Larabel
|
AMD continues recruiting more Linux engineers to join the company not only for their EPYC server processors given the dominance of Linux on the server/HPC front but also as part of their growing Linux client ambitions covering custom SoCs using Linux from Valve's Steam Deck to the Tesla in-vehicle infotainment system over to just running AMD Ryzen processors on Linux. This is good to see given AMD's traditionally much smaller Linux pool of talent compared to Intel's massive Linux/open-source engineering headcount.
It was just last month that I pointed out AMD hiring more again for Linux on the client side. That builds upon last year AMD establishing a "new organization" within AMD focused on Linux client computing.
Last month they were hiring for a client Linux systems engineer while this month among their open positions is looking for a Linux engineer focused on semiconductors. This again is on the client side with the job posting mentioning their usual phrase for that team, "Step up into a relatively new organization built to engage more strategically and deeply with our commercial customers..."
This is on the client CPU side to support new processor features under Linux and the role is based out of Austin, Texas. "Gain expertise on development efforts in the Linux kernel and the open-source ecosystem. Support new processor capabilities by implementing improving new features in Linux. Be a part of our new team as we shape platform configurations and components that represent an efficient integration of hardware and software/firmware. Want to rebuild a kernel with optional components and build flags that can fine tune a kernel for a specific CPU and platform configuration? Want to contribute to the Linux ecosystem by adding & optimizing kernel features for OEM Laptop/Desktop/Workstation designs? This is the ideal opportunity!"
See this job posting if interested in this newest Linux client engineer position.
They are also hiring as well a Linux kernel software engineer on the server side. "A Linux kernel developer with strong analysis and problem-solving skills who will deliver responsive and efficient solutions that bring new AMD processors to market with virtualization solutions. Engage proven technical collaboration to guide the process of AMD hardware enablement for new generation CPUs. Utilize polished written and verbal communication skills to work with on-site and offshore teams, analyze system requirements, coordinate feature design and development with other teams."
See their other new/open Linux positions here. It's great to see them ramping up their Linux staff given their hardware successes of recent years and with their chief competitor Intel already being the most significant contributor to the Linux kernel and numerous other open-source projects... AMD would need a lot more Linux engineers to achieve the same level of timely Linux support and low-level kernel enhancements that Intel has been focused on for years, especially when it comes to Intel's open-source work beyond just the actual hardware device enablement.
| 8
| 1,760,719,298.412417
|
https://www.phoronix.com/news/AMD-Lenovo-Platform-Profile
|
AMD-Powered Lenovo ThinkPads To Soon Have Working Platform Profile Support On Linux
|
Michael Larabel
|
Last month I covered the issue of Lenovo's ACPI Platform Profile support for AMD-powered laptops was busted on Linux. The platform profile controls were exposed but in reality did not work. Fortunately, fixed up support for this feature is now on the way to the Linux kernel for letting users choose between better performance or extended battery life and cooler operating device.
Confirmed in my own tests and discovered independently by Lenovo as well is that the ACPI platform profile support for their ThinkPad laptops on Linux didn't actually work. Choosing between the performance / balanced / power saver modes did not impact the performance/power unlike this functionality with their Intel-powered ThinkPads.
Last month this broken support was disabled in the mainline Linux kernel while now proper support is on the way, likely for the v5.18 kernel. Under review is adding PSC mode support to the ThinkPad ACPI Linux driver as needed for platform profile support on AMD laptops. The PSC mode is properly handled where as previously only the MMC mode used by Intel was supported and taking that code path broke for AMD Ryzen systems.
Thus once that patch makes it to the mainline kernel, the platform profile support should be working on AMD Ryzen powered Lenovo laptops for switching between performance and power-savings preferences.
The ThinkPad T14s Gen2a with Ryzen 7 PRO 5850U running Linux well. For improving ACPI Platform Profile support in general, Lenovo is also proposing support for switching between Platform Profiles in an easier manner whether on AC (wall charging) or DC (battery) modes. Lenovo laptops at least can easily switch platform profiles based on power mode and so under a "request for comments" was providing an implementation that allows for easily setting the platform profile preference depending upon power source.
| 5
| 1,760,719,300.020018
|
https://www.phoronix.com/news/AMD-HSMP-Linux-Driver
|
AMD "HSMP" System Management Driver On The Way To The Linux Kernel
|
Michael Larabel
|
The AMD HSMP kernel driver is currently under review for possible inclusion into the Linux 5.18 cycle. HSMP in this context is the Host System Management Port.
AMD EPYC 7003 series processors introduce the Host System Management Port (HSMP) as a new interface for providing access to additional system management functions on AMD servers. The amd_hsmp driver is on its way for supporting this interface under Linux. The driver in turn provides /dev/hsmp as an interface for sending HSMP mailbox register commands.
The AMD HSMP currently supports messages for reading the current average socket power consumption, setting the socket power limit, getting/setting a maximum frequency limit, setting the min/max width of the xGMI link, enabling the DF P-State Performance Boost support, getting the FCLK and MEMCLK for the current socket, reading the average C0 residency for that socket, getting per-DIMM temperatures and refresh rates, and getting the maximum/current DDR bandwidth usage, among other possible options (messages).
The AMD HSMP is only found with AMD EPYC processors for Family 19h and newer. With EPYC 7003 series having been introduced nearly one year ago, it's a bit unfortunate that this 600+ lines of code kernel driver is only working its way to the mainline kernel now, but at least better late than never.
The driver code as of yesterday made it into platform-drivers-x86's review branch, so if it's all reviewed well and in time this driver should be ready for inclusion into next month's Linux 5.18 merge window.
This documentation commit has more information on the AMD HSMP interface. The AMD esmi_ib_library in user-space for EPYC system management is able to make use of this new /dev/hsmp interface besides any other user-space software now being able to also read/set the various features exposed by the Host System Management Port.
| 1
| 1,760,719,300.242994
|
https://www.phoronix.com/news/AMD-x2AVIC-Linux-KVM-SVM
|
AMD Sends Out Linux Patches For x2APIC Virtualization, Increases KVM Limit To 511 vCPUs
|
Michael Larabel
|
AMD engineers on Sunday night sent out a patch series getting x2APIC virtualization "x2AVIC" support for the AMD SVM driver with the Kernel-based Virtual Machine (KVM).
Some areas of AMD's virtualization support for the mainline kernel haven't been as punctual as desired. For example, SEV-SNP with last year's EPYC 7003 series processors still doesn't have its new feature code mainlined but still going through the review process every so often with new revisions. (At least AMD has been offering a tree on GitHub with the SEV-SNP in-progress kernel code for interested customers.)
Another AMD virtualization item that I thought was squared away already but turns out is only now seeing patches is for AMD x2APIC virtualization support. AMD x2APIC support outside of the virtualization scope has been supported since 2018 with Linux for EPYC 7002 series and newer; the work this week is about x2APIC virtualization, which Intel has had squared away including their Linux support for years (Intel VT-x with Sandy Bridge originally added x2APIC virtualization there). Sent out Sunday night were a set of "request for comments" patches implementing AMD x2APIC virtualization support. The x2APIC virtualization support allows for VMs to have more than 255 CPU threads and allows for better performance than legacy APIC mode for virtualization.
Yep, more vCPUs! This code gets AMD x2APIC virtualization (x2AVIC) squared away for their KVM kernel code. The patches do go ahead and lift the thread count now from 255 vCPUs up to now a possible 511 vCPU limit. With up through AMD EPYC 7003 series processors having up to 256 threads (128 cores) between two sockets, the existing 255 vCPU limit hasn't been much of an issue with normally some cores/threads being reserved for the host OS for management. However, with Zen 4's Genoa will be up to 96 CPU cores (192 threads) now per socket or with Zen 4C "Bergamo" will be 128 cores per socket.
Besides allowing for up to 511 vCPUs now, the AMD x2APIC virtualization support should lead to better efficiency when used. See the RFC patch series for this tentative AMD x2AVIC support for the Linux kernel.
| 1
| 1,760,719,301.684736
|
https://www.phoronix.com/news/AMD-Feb-2022-19h-ucode
|
AMD Releases Updated CPU Microcode For Zen 3 CPUs
|
Michael Larabel
|
AMD today pushed updated Family 19h / Zen 3 CPU microcode to the linux-firmware.git tree.
There is now updated firmware available for Zen 3 CPU owners on Linux.
Sadly, however, these CPU microcode drops still happen without any public change-log... This is the first Family 19h CPU microcode update since November but no indication if it's due to security fixes, general bug fixes, or what the cause of this latest update is for.
Short of major disclosures, AMD normally doesn't list the microcode changes they are pushing to linux-firmware.git. This is an area where Intel does a better job with their CPU microcode releases as they do actually publish detailed change-logs / release manifests.
I'll be poking at the microcode this weekend and running some tests, but for now just hope it's a routine update with any nice improvements as a plus and not a security issue or the like.
Also landing today in linux-firmware.git were updated AMDGPU firmware/microcode files too for matching what's shipping in the Radeon Software 21.50 packaged driver set. There again though is sadly no change-log / notes.
| 16
| 1,760,719,302.855792
|
https://www.phoronix.com/news/AMD-Linux-5.18-Sched-Zen-LLC
|
Linux 5.18 Scheduler Change To Further Boost AMD EPYC Performance For Some Workloads
|
Michael Larabel
|
While AMD EPYC processors already deliver great performance under Linux, with the Linux 5.18 kernel this spring is a scheduler improvement that can provide measurable speed-ups for various workloads on processors where there are multiple last level caches (LLCs) per node, such as with the case of EPYC.
A patch entitled "sched/fair: Adjust the allowed NUMA imbalance when SD_NUMA spans multiple LLCs" may not sound exciting and in fact I almost overlooked it through my usual patch monitoring of mailing lists and Git repositories. But this kernel scheduler change is actually rather significant in the case of AMD EPYC performance on Linux. It's the then-tentative code last year I wrote about (and then forgot about with the wide range of patches and coverage I carry out on Phoronix) but now revised and ready for the mainline Linux kernel.
Linux continues squeezing even more performance out of AMD EPYC/Zen platforms.
Longtime kernel developer Mel Gorman who authored the change explained, "[A kernel scheduler change from 2020] allowed an imbalance between NUMA nodes such that communicating tasks would not be pulled apart by the load balancer. This works fine when there is a 1:1 relationship between LLC and node but can be suboptimal for multiple LLCs if independent tasks prematurely use CPUs sharing cache. Zen* has multiple LLCs per node with local memory channels and due to the allowed imbalance, it's far harder to tune some workloads to run optimally than it is on hardware that has 1 LLC per node. This patch allows an imbalance to exist up to the point where LLCs should be balanced between nodes."
This Linux scheduler change for balancing between NUMA nodes is being improved for cases of the CPU having multiple LLCs per node.
What's exciting though is the end result and that is with an AMD Zen 3 platform he's been testing, the OpenMP-parallelized Stream memory benchmark was 173~272% faster depending upon the memory operation tested. It's a huge win for the upstream Stream memory benchmark but also other workloads depending upon behavior.
There can be huge improvements to performance and lower variation between runs depending upon the particular workload...
For the common Coremark CPU benchmark, the harmonic mean performance was up by 10% with this patch or the maximum result was 17% faster. For the SPECjbb Java benchmark, the performance was up by as much as 18%. The NPB EP benchmark saw a ~17% improvement in performance too and less deviation between runs. Even for workloads where the overall benchmark result didn't see much change, the deviation between runs was lower with this scheduler patch.
The sched/fair patch was pulled into sched/core, which means that barring any issues turning up in the next few weeks, this should be sent in for the Linux 5.18 merge window next month. Linux 5.18 is looking more and more exciting for its spring kernel release with a ton of great improvements. I'll, of course, have out my own benchmarks with this patch and other changes in due course.
| 6
| 1,760,719,303.236736
|
https://www.phoronix.com/news/CPUPower-AMD-P-State
|
AMD P-State Support Coming For The CPUPower Tool
|
Michael Larabel
|
With the forthcoming Linux 5.17 kernel there is the new AMD P-State driver aiming to provide better power efficiency than the ACPI CPUFreq driver that has long been used on AMD platforms. For complementing that AMD P-State driver, AMD has also been working on adding their CPU P-State support to Linux's cpupower tool.
The Linux cpupower tool is used for showing and setting various processor-related values. CPUPower can be used for examining the power-related behavior of the processor and also controlling various CPUFreq tunables. Currently the CPUPower tool doesn't have knowledge of the AMD performance states exposed via ACPI CPPC with Zen 2 and newer, but AMD has been working to change that.
Sent out today were the latest patches wiring up AMD P-State support within the CPUPower tool.
This support allows making use of the finer-grained AMD P-State features rather than legacy ACPI P-States, support for reading the AMD P-State kernel driver's sysfs data, displaying the P-State information, and more.
CPUPower lives within the Linux kernel source tree and hopefully this AMD P-State support will be buttoned up in time for the v5.18 cycle. This cpupower support is for convenience and not necessary for the AMD P-State kernel driver to properly behave, etc. I'll have up some fresh AMD P-State benchmarks from Linux 5.17 shortly on Phoronix.
| 7
| 1,760,719,304.417975
|
https://www.phoronix.com/news/AMD-Sabrina-SoC-Skyrim
|
More Open-Source Patches Continue Surfacing For AMD's Sabrina SoC
|
Michael Larabel
|
The Linux and Coreboot support for the AMD "Sabrina" SoC continues to be worked on while recently Google has merged a new motherboard target for a Sabrina-powered Chromebook.
Over the past month I've been seeing numerous references to the AMD "Sabrina" SoC in open-source patch series, mostly when it comes to the open-source Coreboot code. We haven't seen Sabrina mentioned on AMD roadmaps, unless it's another Linux-specific codename being used by AMD for early platform enablement. But with Sabrina's Coreboot support being focused on, AMD's contributions there have been largely driven by Google and their requirements for Chromebooks / Chrome OS powered devices.
Reaffirming the Chromebook avenue, Coreboot has merged support for the "Skyrim" mainboard as a unreleased device making use of the AMD Sabrina SoC. Much of the Sabrina code within Coreboot is based on the existing AMD Cezanne code.
Sabrina has a CPU ID of Family 19h Model A0h. Some of the other open-source patches for Sabrina point to this SoC having a next-gen Audio Co-Processor "Gen2", Sabrina exclusively uses LPDDR5 memory, and other new additions. But at the same time it's rather odd that Sabrina does away with ACPI CPPC (Collaborative Processor Performance Controls) that is used now on Linux by the new AMD P-State driver. It's unknown why ACPI CPPC support is removed for Sabrina. So at this point we have many open questions around the AMD Sabrina SoC support but at least we continue seeing more open-source/Linux patches around it thanks to the Chromebook aim.
| 15
| 1,760,719,304.932123
|
https://www.phoronix.com/news/AMD-Completes-Xilinx-Deal
|
AMD Completes Its Acquisition Of Xilinx
|
Michael Larabel
|
As was expected with last week AMD receiving all necessary regulatory approvals for its acquisition of Xilinx, today the deal successfully closed.
AMD has completed its acquisition of Xilinx on Valentine's Day. AMD hopes this semiconductor mega deal will further cement it as "the industry's high performance and adaptive computing leader."
Xilinx is becoming the AMD Adaptive and Embedded Computing Group (AECG), led by former Xilinx CEO Victor Peng. This deal increases AMD's total addressable market from around $80B to $135B, expands its customer base and leverages AMD into new markets, will provide R&D benefits moving forward, and AMD hopes it will strengthen their financial model.
AMD. AMD Acquires Xilinx. More information on the completed acquisition of Xilinx at AMD.com.
| 8
| 1,760,719,306.117263
|
https://www.phoronix.com/news/PowerTOP-ThinkPad-Ryzen-5000
|
Benchmarks - Is PowerTOP Tuning Worthwhile For Modern AMD Linux Laptops?
|
Michael Larabel
|
While PowerTOP was immensely helpful when the Intel open-source project started out in 2007 for reporting untuned kernel parameters and noting what's keeping the CPU from reaching its deeper sleep states, over the past decade Linux has greatly improved when it comes to power management and better behavior out-of-the-box. PowerTOP continues to see occasional commits and new releases, but there's less talk about it these days than going back a number of years when it was a must-have for x86_64 laptops. In any case I was curious to see if following its tips still provided any meaningful difference on a modern AMD Ryzen powered laptop.
This quick round of testing was with a Lenovo ThinkPad T14s Gen2 featuring a AMD Ryzen 7 PRO 5850U.
Testing happened out-of-the-box and then after running PowerTOP as root and setting all the "bad" defaults to "good".
Before/after various benchmarks were run on this Ryzen-powered ThinkPad laptop while also monitoring the power consumption on battery and various thermals too.
Long story short, the overall battery power consumption changed on the Lenovo ThinkPad T14s Gen2 AMD Ryzen 7 PRO powered laptop when PowerTOP-tuned.
There were some minor improvements to the CPU temperature operating temperature.
Given no measurable CPU power savings difference for this Zen 3 powered laptop, there also wasn't any measurable difference in the performance of tested workloads from browser benchmarks to other common laptop real-world tests.
This largely jives with what I've seen on other laptops in recent times of outside of some niche cases, PowerTOP on modern Linux distributions generally doesn't yield a dramatic difference with modern hardware, thanks to ongoing Linux improvements and more sane defaults by Linux distributions and power management behavior with open-source software generally being less buggy in recent years.
| 17
| 1,760,719,306.522105
|
https://www.phoronix.com/news/AMD-Linux-Is-Thunderbolt
|
AMD Preparing More Linux Improvements Around USB4/Thunderbolt
|
Michael Larabel
|
As part of AMD Rembrandt APUs having USB4 support with that specification based on the Thunderbolt 3 protocol, AMD in recent months has been making a number of Linux driver improvements to enhance the USB4/Thunderbolt support for their platforms.
AMD patch series in recent times have included USB4 DisplayPort Tunneling and other USB4/Thunderbolt work. Their latest is refactoring various Linux kernel around an "is_thunderbolt" check used by drivers within their kernel for altering their behavior if the device is connected via Thunderbolt as opposed directly via PCIe and as a means of determining if the device is potentially removable / externally connected. That is_thunderbolt check started out for early Intel Thunderbolt controllers that lacked command completed events.
AMD Linux engineer Mario Limonciello has been posting a few revisions of the "is_thunderbolt" patch series this past week so that ultimately the intended driver behavior covers "non-Intel USB4 designs" (a.k.a. AMD). The is_thunderbolt check is now treated as a kernel quirk and various other Thunderbolt-related kernel code changes are also part of the proposed patch series. As part of the 12 patches is also cleaning up AMD and Nouveau driver code for its paths involving eGPU / removable GPU support and other quirks.
See this patch series for more details but long story short AMD is working on more USB4/Thunderbolt handling improvements for Linux to improve non-Intel platforms.
USB4 on the AMD side is being introduced with the Ryzen 6000 mobile series APUs.
| 10
| 1,760,719,307.668597
|
https://www.phoronix.com/news/AMD-Xilinx-Regulators-Approve
|
AMD Expected To Complete Its Acquisition Of Xilinx Next Week
|
Michael Larabel
|
AMD just announced that it has received approval from all necessary regulators to proceed with its acquisition of Xilinx.
Back in October 2020 AMD's acquisition of Xilinx was announced. While they hoped to close in Q4'2021, that draw into this quarter but now everything is all aligned that they should be closing on the deal next week.
AMD just issued a press release that it has received all necessary approvals for its proposed acquisition of Xilinx. Now having all the necessary governmental approvals, AMD expects the transaction to close on or about 14 February -- quite a Valentine's Day gift for AMD.
It will be very interesting to see how this plays out and hopefully will yield more Linux/open-source contributions. AMD and Xilinx have already been working on bringing ROCm to Xilinx FPGAs, as just one example of the Linux/open-source software impact.
| 40
| 1,760,719,308.339946
|
https://www.phoronix.com/news/AMD-Linux-Client-Hiring-2022
|
AMD Is Hiring Again For Their Linux - Client - Effort
|
Michael Larabel
|
After establishing a new organization within AMD last year focused on improving AMD client platforms on Linux, they are now hiring again for this endeavor.
While AMD has been ramping up their Linux staff on the server front given their EPYC successes, last February they also began hiring for Linux on the client side within "a new organization" at the company. That new organization is off the ground and they were able to add some notable hires last year and hit the ground running with both client fixes and new features / hardware enablement.
Those AMD Linux client efforts appear to be paying off already with seeing a number of laptop/SoC power management fixes in recent months/kernels, Yellow Carp (Rembrandt) support already in the k10temp driver where as traditionally AMD has been late to the game and only providing post-launch thermal monitoring support, and other efforts that will hopefully play out for Linux users/customers over the quarters to come on the client side... If you are a frequent Phoronix reader, you've likely seen many of their fruits covered in numerous articles over recent months. It's been nice seeing more AMD fixes and improvements directly from their engineers on the client side.
There have also been design wins for AMD on the client side that involve Linux support such as Valve's Steam Deck relying on a VanGogh APU while running their Arch Linux based SteamOS and also Tesla's latest Ryzen-powered in-car infotainment system. AMD Linux client efforts are moving in the right direction although there are still areas to improve upon such as better supporting the upstream open-source compilers, ensuring their new hardware enablment aligns more optimally for just not being in the mainline tree at launch but found within distribution kernels, and ideally to see more open-source firmware.
I was informed this morning that AMD is hiring again on the Linux client front. The job posting from the end of last week notes, "Step up into a relatively new organization built to engage more strategically and deeply with our commercial customers - specifically their architecture and engineering teams...As a key engineer on our team, you configure and tune Linux distribution packages to take advantage of AMD technologies on customer platforms. Analyze, configure, and propose changes to platform system firmware to allow the kernel and other services extract platform configuration and runtime information. Join a group of engineers collaborating with customers to help understand how the configuration and enablement of platform and software features fulfill end-to-end user scenarios in commercial platforms. Participate in planning the next generation CPU/APUs. Figure out how changes in the kernel upstream pipeline help meet the goals of new platforms."
The latest job post for an AMD Client Linux Systems Engineer is looking for a software engineer already experienced in this area and also having good knowledge of hardware and low-level debugging too. The position is for on-site in Austin, Texas.
Much of the verbage is the same as last year's AMD Linux client job postings, "Want to rebuild a kernel with optional components and build flags that can fine tune a kernel for a specific CPU and platform configuration? Want to contribute to the Linux ecosystem by adding & optimizing kernel features for OEM Laptop/Desktop/Workstation designs? This is the ideal opportunity!"
In addition to that job posting, outside of their Linux client organization they are also hiring on the Linux server front still for Linux networking I/O lead, Linux software system engineer, virtualization performance developer, Linux system software design engineer, and other jobs with an EPYC focus. (If applying, be sure to mention Phoronix!) It's good to see since Intel has traditionally hired far more Linux/open-source engineers than AMD.
| 7
| 1,760,719,309.995806
|
https://www.phoronix.com/news/LUMI-AMD-OSS-Tuning-2022
|
Europe's AMD-Powered LUMI Supercomputer Continues With Code Porting, Open-Source Tuning
|
Michael Larabel
|
The LUMI supercomputer in Finland is still being assembled with its 2,560 nodes consisting of a 64-core AMD Trento CPU and four AMD Instinct MI250X GPU accelerators per node. This 375+ PFLOPs was supposed to come online by the end of 2021 but was challenged by the supply chain crisis and is now aiming for general availability by the middle of the year. While the hardware is still coming together, their HPC engineers have been hard at work optimizing the open-source Linux software stack.
Last February there was a great presentation by Georgios Markomanolis who is the lead HPC scientist at CSC. He talked about the state of the Radeon Open eCosystem last year and all their porting efforts involved in getting traditionally NVIDIA centered HPC workloads working in their AMD test hardware at the time -- originally MI100 while waiting on the MI250X to be used in LUMI.
Georgios Markomanolis presented last weekend at FOSDEM 2022 to share more about their latest open-source software efforts over the past year.
The HPC engineers continue becoming more experienced with AMD's ROCm open-source software stack for GPU computing, continue hammering on HIP for porting more CUDA codebases into AMD GPU supportive software, and then working to fine-tune the codes for maximum performance once running on the AMD hardware.
One of the new software components they have added to their toolbag over the past year is AMD's open-sourced GPUFort for helping to move OpenACC and CUDA Fortran code into AMD's architecture.
CSC LUMI It still is a rather complex jungle moving complex codebases into the AMD ROCm space for optimal GPU performance.
CSC LUMI Those wanting to learn more about the open-source porting and tuning work going on at CSC to prepare for LUMI coming online can see Georgios Markomanolis' great virtual presentation at FOSDEM.org with the slides and video recordings. Those wanting to learn more about LUMI can do so at lumi-supercomputer.eu.
| 13
| 1,760,719,311.158603
|
https://www.phoronix.com/news/AMD-OSF-2022
|
The Less Than Ideal State Of AMD Open-Source Firmware Support In 2022
|
Michael Larabel
|
There's been some activity on AMD open-source firmware support for newer hardware platforms but for those wanting a fully open-source firmware stack, there remains work on older generations of AMD server platforms. Michał Żygowski of firmware consulting firm 3mdeb presented today at FOSDEM 2022 as to the current state AMD open-source firmware efforts around Coreboot.
Since Michał Żygowski's talk last year at FOSDEM on open-source AMD firmware support, Coreboot has seen support for AMD Ryzen 5000 series "Cezanne" APUs. That Cezanne APU support for Coreboot seems to be primarily driven for AMD's Chromebook efforts with Google continuing to mandate it. However, AMD has not yet made available any public Cezanne firmware support package for use with Coreboot.
Something new and not previously covered on Phoronix is AMD working on "Sabrina" SoC support in Coreboot. Sabrina is for Family 17h Models A0h to AFh. Those model IDs have turned up already in Linux 5.17 kernel patches and are next-generation Zen cores. So it's good to see AMD preparing this Sabrina support now, presumably though is just driven with Chromebook intentions.
One addition to Coreboot's amdfetool is working on AMD BIOS recovery support for when the system fails to POST. This amdfwtool for supported systems is allowing it to boot from the backup BIOS image.
While last year there was hope expressed of seeing open-source firmware support for modern AMD EPYC server platforms, at the moment there are no updates. There is also "concerning information" about the AMD server open-source firmware prospects. I haven't heard anything new myself in private channels, but still holding out hope for change...
Also concerning open-source AMD firmware developers is AMD's Platform Secure Boot functionality where CPUs can be locked to particular vendors. Besides this feature blocking use of select AMD CPUs from use in other vendor motherboards, it would prevent complications around being able to load open-source firmware on said platforms/processors.
Meanwhile some older AMD Optera era motherboards with Coreboot ports have seen their code fall behind and not being well maintained in all cases. Those ports not being updated for modern Coreboot interfaces run the risk of ultimately being removed from the source tree.
Some older platforms though like the ASUS KGPE-D16 are seeing a revival in their open-source firmware support thanks to sponsorship with some organizations interested in that platform for blockchain developers and other purposes. Dasharo is being worked on as an open-source firmware distribution downstream focused on simple code, long-term maintenance, and privacy-respecting.
Those wanting to learn more about the open-source AMD firmware efforts by 3mdeb's Michał Żygowski can see his other presentation assets from today up on FOSDEM.org.
Here's to hoping more AMD open-source firmware efforts materialize in 2022...
| 21
| 1,760,719,314.940829
|
https://www.phoronix.com/news/Shadow-Stacks-Linux-AMD
|
Intel's Linux Shadow Stack Patches Should Work Fine With AMD CPUs
|
Michael Larabel
|
Intel has for a while been posting Linux kernel patches for implementing Control Flow Enforcement (CET) technology, both for the Indirect Branch Tracking and Shadow Stack features. However, as written about earlier this week, Intel is focusing on the shadow stack support for user-space. The patches posted this past week by Intel for Linux Shadow Stack for User-Space support was limited to their own processors but fortunately it's appearing to be work out fine for AMD CPUs too.
The shadow stack functionality is about defending against return-oriented programming (ROP) attacks. The Shadow Stack keeps a copy of each CALL and upon a return (RET) will check the return address stored in the normal stack to verify it matches the contents of the Shadow Stack otherwise will generate a fault.
An Intel graphic on Shadow Stack as part of CET. With the 35 patches posted this past week, the code was limited to being enabled with Intel CPUs given that is what Intel engineers have been obviously testing. But AMD Zen 3 processors also support the Shadow Stack functionality and as acknowledged in the Intel patches there was just a lack of being able to test these patches there.
This patch can hopefully be dropped now that there is AMD testing. Fortunately, an AMD Linux engineer has been testing the CET Shadow Stack patches and commented that the patches appear to be running fine on AMD processors - including when testing a patched CET version of the GNU C Library and passing various reference tests.
So assuming no issues turn up moving forward, the CET Shadow Stack support once finally mainlined into the Linux kernel should work for both Intel and AMD CPUs as a security improvement.
| 4
| 1,760,719,315.93382
|
https://www.phoronix.com/news/LLVM-15-Faster-Zen-Sqrt
|
LLVM Clang 15 Enables Faster Square Root Instructions For AMD Zen
|
Michael Larabel
|
As part of an effort to update LLVM Clang's "-mtune" handling to cater to newer processors, AMD Zen processors with LLVM/Clang 15 later this year will be able to enjoy faster and more accurate square root calculations with tuning to use SQRTSS/SQRTPS instructions.
Merged today to mainline for LLVM/Clang 15 (not to be confused with the branched LLVM/Clang 14 releasing next month) is fast SQRTSS (Compute Square Root of Scalar Single-Precision Value) / SQRTPS (Square Root of Single-Precision Floating-Point Values) tuning for AMD Zen processor cores. With Zen 1 and newer those instructions are found to be fast enough and worthwhile than the existing code path while also being more accurate.
This tuning for AMD Zen comes while on the Intel side they already enabled TuningFastScalarFSQRT going back to Sandy Bridge and the TuningFastVectorFSQRT has been in place since Skylake. While this LLVM tuning change affects all Zen CPUs going back to Zen 1, the LLVM change is only happening now in 2022.
This square root instruction tuning for AMD Zen came up as part of a broader discussion for improving the -mtune generic behavior for more modern CPUs, similar to GCC's -mtune default catering to Haswell. As noted in that discussion, "znver1/znver2 schedule models are, well, leave a lot to be desired." Sadly, there isn't as much aggressive AMD compiler tuning by LLVM (and GCC) as there is on the Intel side.
Zen 1 is already a half-decade old while this change for LLVM/Clang 15 will be out as stable around September 2022. Sadly this change is just another example of AMD software optimizations coming in late (and often times left up to independent parties / the open-source community), especially on the compiler side while Intel generally is very early in their new CPU family targeting and ensuring they are well optimized with accurate cost tables, able to make use of new instructions, etc.
This SQRTSS/SQRTPS tuning for Zen is the first AMD Zen specific activity for LLVM going back to last September. Hopefully we'll see more AMD open-source compiler tuning happen this year -- we still haven't seen znver4 introduced yet while Intel started their Alder Lake and Sapphire Rapids compiler patchwork back in mid-2020.
| 27
| 1,760,719,316.849664
|
https://www.phoronix.com/news/AMD-MI200-Not-Experimental
|
AMD MI200 "Aldebaran" Linux Driver Support No Longer "Experimental"
|
Michael Larabel
|
Back in November AMD announced the MI200 accelerator that has seen its Linux open-source driver support developed under the "Aldebaran" codename going back to February of last year. The AMD developers are now removing the "experimental" flag from that Aldebaran class GPU support.
It was twelve months ago that AMD published the initial Aldebaran Linux support and continued working on the upstream Linux open-source support for the CDNA-based hardware accelerator well ahead of the official Instinct MI200 debut. This Arcturus successor has matured with new Linux kernel versions and settled down in time for launch.
But to now the Aldebaran support has been hidden behind the "AMD_EXP_HW_SUPPORT" flag. That AMD experimental hardware support flag prevents the support from being activated by default with the AMDGPU/AMDKFD kernel driver unless an override module parameter is set. The amdgpu.exp_hw_support=1 module parameter is what's needed to make use of that "experimental" hardware support, similar to the Intel graphics driver on Linux also requiring a "force probe" option for its experimental/early hardware support.
With a patch sent out today, Aldebaran PCI IDs will no longer be behind that experimental hardware support flag and thus supported out of the box. AMDGPU maintainer Alex Deucher of AMD commented on the patch, "These [Aldebaran devices] been at production level for a while. Drop the flag."
So it's good to know that the Aldebaran support is already considered production-ready and thus moving forward with the Linux 5.18 kernel (or Linux 5.17 if submitted as a "fix" in the coming weeks) will have this Aldebaran support out-of-the-box for the AMD Instinct MI200 series. This is in regards to the upstream open-source Linux kernel support while AMD also maintains their packaged Radeon Software for Linux enterprise driver stack for use on the various RHEL/CentOS/SUSE/Ubuntu enterprise/LTS Linux distributions on older kernels.
| 14
| 1,760,719,317.58075
|
https://www.phoronix.com/news/Lenovo-AMD-PP-Not-Working
|
Lenovo's Platform Profile Support For AMD Systems On Linux Has Been Busted
|
Michael Larabel
|
ACPI Platform Profile support on Linux has been useful for catering to balancing your power or performance preferences with modern laptops on Linux. It has worked well in general across various devices tested but it turns out to be a dud currently when it comes to AMD Ryzen powered Lenovo systems.
ACPI Platform Profile support allows setting the power/balance/power-savings preference (among other possible profiles) on Linux with recent kernel activity that so far has seen major support available for major laptop brands like Lenovo, Dell, and ASUS. The user's profile preference can be set via a sysfs interface or desktops like KDE Plasma and GNOME have already added convenient user interfaces for setting it nicely from the system settings area.
On supported systems, GNOME offers the ACPI Platform Profile to be set via the "Power Mode" area of the system settings. While Lenovo ThinkPad laptops with AMD Ryzen processors may be advertising ACPI Platform Profile support on Linux, it's not actually behaving correctly. I had stumbled upon this myself in December with the ThinkPad T14s Gen2 / AMD Ryzen 7 PRO 5850U. Back then I ran some power profile benchmarks on it and was surprised to see no change between the different profiles, unlike Intel-powered Lenovo systems and other laptop brands I have tested over the past year with this ACPI Platform Profile support being available in the kernel.
I hadn't the time yet to do any further testing or dig into it any deeper with my ever-long TODO list, while this week an explanation arose... While the code has been in the kernel, Lenovo's support is actually busted for AMD laptops.
This patch working its way to the mainline kernel at least for Linux 5.17 ends up disabling the support for its non-working state. The Lenovo contributed patch explains, "Lenovo AMD based platforms have been offering platform_profiles but they are not working correctly. This is because the mode we are using on the Intel platforms (MMC) is not available on the AMD platforms. This commit adds checking of the functional capabilities returned by the BIOS to confirm if MMC is supported or not. Profiles will not be available if the platform is not MMC capable."
In other words, the Platform Profile support will just be disabled for now on Lenovo systems with AMD processors given this pending patch. Lenovo is looking at other means of properly supporting it for AMD hardware but no solution yet.
| 17
| 1,760,719,319.916111
|
https://www.phoronix.com/news/AMD-PAN-Linux-RFC
|
AMD Cooking Up A "PAN" Feature That Can Help Boost Linux Performance
|
Michael Larabel
|
AMD open-source engineers sent out a request for comments on a new kernel feature called "PAN", or Process Adaptive autoNUMA. Early numbers shown by AMD indicate that PAN can help with performance in some workloads on their latest server hardware by a measurable amount.
The proposed PAN is Process Adaptive autoNUMA and is an adaptive algorithm calculating the AutoNUMA scan period. AMD's Bharata B Rao further explained in the request for comments (RFC) Linux kernel patch series, "In this new approach (Process Adaptive autoNUMA or PAN), we gather NUMA fault stats at per-process level which allows for capturing the application behaviour better. In addition, the algorithm learns and adjusts the scan rate based on remote fault rate. By not sticking to a static threshold, the algorithm can respond better to different workload behaviours. Since the threads of a processes are already considered as a group, we add a bunch of metrics to the task's [memory management] to track the various types of faults and derive the scan rate from them. The new per-process fault stats contribute only to the per-process scan period calculation, while the existing per-thread stats continue to contribute towards the numa_group stats which eventually determine the thresholds for migrating memory and threads across nodes."
The important part for end-users / AMD EPYC customers is how PAN can benefit Linux performance. With a PAN'ed Linux kernel build, they found the Graph500 interconnect HPC benchmark to benefit by as much as 14.93% compared to a default Linux kernel build, NAS benchmarks were up to 8% faster, PageRank only about 0.37% faster, and other results from less than 1% to the more significant numbers noted. That's just with the limited selection of tests evaluated so far by AMD - it will certainly be fun to benchmark this patch series if it moves past the RFC stage and is something other kernel maintainers get behind and ultimately be upstreamed into the kernel.
So far no other kernel developers have commented on the Process Adaptive autoNUMA proposal but those interested can see the RFC series for learning more about this feature or testing it out. In its current form is less than 400 lines of new code to improve the Linux NUMA behavior.
| 7
| 1,760,719,321.300333
|
https://www.phoronix.com/news/AMD-SEV-SNP-Guest-v9
|
AMD Publishes Latest Linux Patches For Enabling SEV-SNP Guest Support
|
Michael Larabel
|
One of the additions with EPYC 7003 "Milan" processors introduced last year was SEV-SNP as the "Secure Nested Paging" addition to AMD's Secure Encrypted Virtualization found with EPYC processors. While they have maintained an out-of-tree Linux source repository with the SEV-SNP patches, the mainline kernel is still lacking support for these latest security features but the code continues to undergo revisions and review for its eventual upstreaming.
Over SEV and SEV-ES "Encrypted State" introduced with prior EPYC processors, SEV-SNP is able to offer greater integrity with replay protection, data corruption safeguards, and fend off a variety of other possible attacks to VMs.
AMD didn't begin posting their SEV-SNP kernel patches publicly until after the EPYC 7003 series processors were first announced and thus the ongoing process still for getting the guest and KVM hypervisor support squared away and upstreamed so EPYC server users will be able to run off a mainline kernel without relying on any distribution-patched kernel or other third-party kernel builds to take advantage of the latest security features on EPYC.
Sent out on Friday were the latest guest patches. The SEV-SNP guest support is now up to its ninth revision for the code needed to be running in the VM guest kernels to make use of these strong memory integrity protections and other features. As with prior series, this SEV-SNP support isn't yet complete but features like interrupt protection are expected to be added after the initial SEV-SNP code is accepted to mainline.
The SEV-SNP guest v9 support is now out for review on the kernel mailing list. The KVM hypervisor support for SEV-SNP is maintained as a separate patch series and wasn't updated last week. It's unfortunate that it's going to be a year after the EPYC Milan launch at the earliest before this code is all mainlined, but then again at times we have seen Intel also running behind in their mainline kernel support for Intel CPU features like SGX and now the ongoing work with TDX. In any case at least the code once mainlined should be in good shape.
EPYC users can grab the SEV-SNP kernel patches via this GitHub repository as another easy source if wanting to spin your own kernel.Meanwhile on the Intel side this week brought the v2 patches of Intel's TDX guest core support. This is the series for Intel's Trust Domain Extensions for confidential guest VMs from the host and physical attacks. Intel's TDX provides similar functionality to AMD SEV for future Intel CPUs.
| 1
| 1,760,719,321.875897
|
https://www.phoronix.com/news/Linux-5.17-AMD
|
Linux 5.17 Is Bringing Big Improvements For AMD Hardware
|
Michael Larabel
|
Thanks to hiring more Linux developers and preparing to ramp up for next-generation hardware support, the in-development Linux 5.17 kernel is going to be another exciting step forward for AMD Linux customers.
Yesterday I wrote up the Linux 5.17 feature overview now that Linux 5.17-rc1 has been tagged, which marks the end of new features being accepted for this kernel. Linux 5.17 will now see weekly release candidates until the stable kernel is ready around the end of March. While it won't be the default kernel of Ubuntu 22.04 LTS, it should be for the likes of Fedora 36 and other spring Linux distribution releases and obviously the rolling kernels. Below is a look at the many AMD improvements to find with Linux 5.17.
First and most impactful, the new AMD P-State driver is merged. This is the driver developed by AMD in cooperation with Valve for improving the power efficiency on Linux compared to the generic ACPI CPUFreq driver. The AMD P-State driver relies upon ACPI Collaborative Processor Performance Controls (CPPC) exposed by the platform for making more fine-grained CPU frequency / performance state information available. Using AMD P-State with the Schedutil governor should lead to better performance than CPUFreq with Schedutil as commonly used today. I'll have fresh benchmarks of AMD P-State vs. ACPI CPUFreq with different governors and different hardware in the coming days on Phoronix. Due to ACPI CPPC dependence, AMD P-State only will work with Zen 2 and newer processors and CPPC must be enabled on the system.
There is also a lot of next-gen (Zen 4) processor work including SMCA updates and EDAC updates, including identifying Registered DDR5 (RDDR5) and Load Reduced DDR5 (LDDR5) memory types for reporting. Also useful is seeing k10 temperature reporting support merged ahead of time rather than after the fact as sadly has been common during Zen 1/2/3 for not seeing timely CPU temperature monitoring support... Hopefully the recent trend continues holding up for timely temperature sensor support.
AMD Smart Trace Buffer support is ready with Linux 5.17 both on the CPU side and with the Radeon dGPU side with newer hardware. AMD Smart Trace Buffer is basically a cyclic data buffer having log information about system execution for helping to analyze failures. The STB is always running and can be dumped when an error occurs without having to reproduce it or run any extra instrumentation.
AMD Renoir hardware now has Sound Open Firmware Support for its audio co-processor and is the first AMD platform supporting Sound Open Firmware for what originally started as an Intel project. Also on the AMD Linux laptop front there is an AMD s2idle failure fix, the latest in an array of S2idle / S0ix related AMD Linux work. Meanwhile many newer ASUS motherboards with X570/B550/B450/X470 chipsets now have working sensor support finally under Linux. Linux 5.17 also introduces Rembrandt SoC network support>
Over on the Radeon side of the house, there is GPU recovery support for Rembrandt APUs, Seamless Boot for Van Gogh, and bug fixes. The AMDGPU DRM kernel driver side wasn't as exciting this time around as some of the other recent kernel releases.
While not impactful for modern hardware, Linux 5.17 does retire AMD 3DNow! instruction usage from within the kernel code. Also on the older AMD hardware front, AMD Fusion APU systems with Hudson D4 chipsets may see a boot time improvement.
Stay tuned for Linux 5.17 kernel benchmarks ramping up on Phoronix over the days/weeks ahead.
| 13
| 1,760,719,323.227322
|
https://www.phoronix.com/news/AMD-SMCA-Linux-5.17
|
AMD SMCA Updates Land In Linux 5.17 For Future CPUs
|
Michael Larabel
|
Last week I noted about EDAC changes in Linux 5.17 for future AMD CPUs. The "Error Detection and Correction" work included AMD adding RDDR5 / LRDDR5 support to their driver and new CPU model IDs that appear to be for Zen 4. Also working on next-gen AMD processor support in Linux 5.17 are recent SMCA changes.
Back in early December I wrote about AMD volleying new SMCA driver changes for a new generation of AMD processors. Those changes are interesting for their Scalable Machine Check Architecture as it introduces the notion of having possibly different bank type layouts depending upon the logical CPU core. Thus the preparations are being made -- and now merged in Linux 5.17 -- for where the machine check architecture bank types/layout are different depending upon the particular CPU. The patches don't provide AMD's reasoning for this change but it's possible for hybrid core designs or other reasons.
The patch series also introduces a number of new SMCA bank types. Among those new bank types are for an updated "v2" unified memory controller, an updated PCIe unit, and various xGMI units.
Those AMD SMCA changes for future AMD CPUs were merged last week as part of ras/core updates for Linux 5.17.
Other next-gen AMD CPU work in Linux 5.17 beyond the SMCA and EDAC additions is temperature monitoring support for those new CPU model IDs touched by the EDAC work. It's good seeing this work happening now ahead of Zen 4 processors launching later in 2022.
| 0
| 1,760,719,323.708128
|
https://www.phoronix.com/news/Linux-5.17-AMD-Hudson-AHCI
|
Some Older AMD Systems Can Boot Faster On Linux 5.17+
|
Michael Larabel
|
A change merged overnight with the libata subsystem updates for Linux 5.17 means that some older AMD hardware will be able to boot quicker by avoiding an otherwise mandated sleep period.
Merged this morning were the ATA subsystem updates for Linux 5.17. Usually the ATA changes don't amount to many noteworthy changes but "Add support for AMD A85 FCH (Hudson D4) AHCI adapters" got my attention... Yeah, the chipset from the early AMD "Fusion" APU days.
What that support change is about is adding the AMD A85 Hudson D4 chipset to the AHCI driver code so it can apply a quirk/workaround.
Within the libata SATA driver code there is a forced 200ms delay with an accompanying code comment, "Some PHYs react badly if SStatus is pounded immediately after resuming. Delay 200ms before debouncing." But the 200 ms delay can be noticeable if the system is otherwise fast at booting. For those still using an AMD A85 FCH (Hudson D4) it's been deemed safe to skip this 200 ms debounce delay in the SATA link resume code.
For now the AMD A85 Hudson D4 is the only chipset listed while hopefully other hardware can be added when deemed safe and now that this "board_ahci_no_debounce_delay" has been added to the AHCI code.
The commit mentions the difference for the developer who tracked down the boot time slowdown, "On the ASUS F2A85-M PRO it reduces the Linux kernel boot time by the expected 200 ms from 787 ms to 585 ms." This is far from the first time we have seen Linux kernel boot time optimizations come by avoiding arbitrary delays catering to select hardware.
| 7
| 1,760,719,324.856951
|
https://www.phoronix.com/news/AMD-s2idle-Check-FW
|
AMD Prepares Linux Fix For Some Laptops Not Resuming From s2idle Suspend
|
Michael Larabel
|
Recently there have been reports of some AMD Ryzen powered notebooks being unable to correctly suspend from resume in s2idle mode. It appears the issue ultimately stems from a firmware setting issue and a set of Linux patches were sent out today to address the condition.
For some modern Ryzen-powered laptops there have been reports recently of some failures in resuming from suspend-to-idle, even after recent AMD s2idle Linux fixes.
The issue appears to stem from the fact that currently Linux just assumes to offer s2idle even if the Fixed ACPI Hardware Table (FADT) doesn't directly indicate it or if there is not an low power S0 (LPS0) device activated. For at least some Intel hardware, s2idle can handle the suspend/resume cycle out even without proper firmware support... AMD hardware, however, cannot without the necessary firmware bits aligning.
This new patch series thus checks for the proper prerequisites before enabling s2idle support on AMD hardware. So now if the firmware support is not present/enabled, s2idle won't be offered to avoid potentially failing to resume when S3 should be used instead.
One of the patches from AMD's Mario Limonciello goes on to explain, "On some OEM platforms a BIOS option is offered that will set the sleep mode between S3 and S2idle. This option will change certain [hardware] behaviors. When in S2idle mode, Linux works properly. However when S3 mode is picked but the user chooses S2idle in Linux the platform may not be properly resumed. To avoid users getting into this situation, don't offer s2idle on AMD systems missing the LPS0 device either by a BIOS option or by using the acpi "no_sleep_lps0" module parameter."
The Lenovo P14s Gen2 with Ryzen 5000 series "Cezanne" SoCs are known to be affected by this issue but likely numerous other laptops too. It's good to see these patches out there now but surprising such firmware checks were not added in the first place for the x86 ACPI s2idle code or spotted previously with AMD working on the s2idle Linux support for more than one year.
| 11
| 1,760,719,325.319412
|
https://www.phoronix.com/news/AMD-P-State-Linux-5.17
|
AMD P-State Driver To Premiere In Linux 5.17 With Aim To Deliver Better Power Efficiency
|
Michael Larabel
|
The AMD P-State driver that has been available in patch form since September and stems from AMD's collaborations with Valve around the Steam Deck will be introduced to mainline with the upcoming Linux 5.17 kernel.
After going through several rounds of patch review the past number of months, the AMD P-State driver as an alternative to the common ACPI CPUFreq driver is going mainline. This AMD P-State driver relies on ACPI CPPC (Collaborative Processor Performance Controls) for making finer CPU frequency scaling / performance state decisions than what is afforded by ACPI CPUFreq. But because of the dependence on ACPI CPPC information, it only supports Zen 2 processors and newer. Additionally, CPPC functionality must be enabled by the system firmware/BIOS as well otherwise you will continue needing to use CPUFreq.
AMD-pstate is very promising for helping to boost the power efficiency of particularly AMD Ryzen desktop and mobile processors. Valve has been working with AMD on this driver in part to help the forthcoming AMD-powered Steam Deck handheld gaming console.
From late November you can see some of my recent AMD P-State Linux power/performance testing while when Linux 5.17 settles down in the weeks ahead I'll be running some fresh benchmarks on multiple systems.
As of yesterday the driver was pulled into the Linux power management subsystem's "linux-next" branch. With PM maintainer Rafael Wysocki now having pulled it into subsystem's "-next" branch, it will be submitted to mainline next week once the Linux 5.17 merge window is opened followed this weekend's release of Linux 5.16 stable. Linux 5.17 stable in turn should be out around the end of March but sadly will likely be too late for finding it in Ubuntu 22.04 LTS.
| 10
| 1,760,719,326.511433
|
https://www.phoronix.com/news/AMD-Ryzen-6000-Pluton
|
AMD Ryzen 6000 Series Mobile CPUs Feature Microsoft's Pluton Security
|
Michael Larabel
|
Back in 2020 Microsoft announced their "Pluton" security chip that woulld be coming to future AMD, Intel, and Qualcomm processors. The Pluton security processor is designed to improve the system security under Windows and now we find out that AMD's forthcoming Ryzen 6000 "Rembrandt" mobile processors will be the first featuring this security feature that may prove controversial to Linux/open-source fans.
We hadn't heard much of Microsoft's Pluton since 2020 but integration into the silicon obviously takes a while and now we find out AMD is ready to introduce it with their forthcoming Ryzen 6000 series processors. AMD's CES 2022 keynote is later today but the CES website has exposed early that the Ryzen 6000 series mobile processors will indeed have Microsoft Pluton:
This ces.tech page makes note of the new processors featuring Microsoft Pluton.
Back to the Microsoft 2020 announcement of Pluton they talked up cloud security and all the benefits albeit only for Windows PCs with no mention of Linux.
Of Pluton's functionality, that Microsoft post sums it up as, "The Pluton design removes the potential for that communication channel to be attacked by building security directly into the CPU. Windows PCs using the Pluton architecture will first emulate a TPM that works with the existing TPM specifications and APIs, which will allow customers to immediately benefit from enhanced security for Windows features that rely on TPMs like BitLocker and System Guard. Windows devices with Pluton will use the Pluton security processor to protect credentials, user identities, encryption keys, and personal data. None of this information can be removed from Pluton even if an attacker has installed malware or has complete physical possession of the PC."
We also haven't seen any patches (yet?) from Microsoft or AMD looking to enable Pluton security processor functionality under Linux... We'll see what comes of this new security feature and whether it means anything for Linux users -- either for improving system security or causing any obstacles in Linux/BSD open-source usage. Given UEFI SecureBoot issues in the early days and other nightmares for Linux users when Microsoft has pushed new standards and hardware, Linux users may be apprehensive of Pluton but it hopefully won't cause issue with AMD's growing stellar open-source Linux support.
AMD's CES 2022 press conference is coming up in just a few hours so stay tuned for more new hardware details then and what the company has to say about Pluton.
| 64
| 1,760,719,327.042746
|
https://www.phoronix.com/news/Top-AMD-News-2021
|
The Most Exciting AMD Linux / Open-Source News Of 2021
|
Michael Larabel
|
As part of our various year-end articles, here is a look back at the most popular AMD Linux/open-source news of the year with the many milestones they achieved in ramping up their support both for desktop/mobile and server hardware and continued successes when it comes to their open-source Radeon graphics driver stack.
Below is a look at the twenty most popular AMD open-source/Linux news stories on Phoronix for the year. This is just looking at our original news items and not the hardware reviews or other comparison benchmark articles. AMD this year published their new CPU frequency scaling driver that has been working its way to mainline, they have been ramping up their Linux talent at the company, they squared away their Zen 3 support and have been making early preparations for Zen 4, and their GPU software efforts remain very active from the AMDGPU kernel driver and Mesa to their ROCm software efforts.
What do you hope to see out of AMD on Linux in 2022? Let us know in the forums but for now below is a look at the most popular AMD Linux news items for this year.
AMD + Valve Working On New Linux CPU Performance Scaling Design 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. 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. Linux 5.15 Is A Very Exciting Kernel For AMD 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 Achieves 5.1M IOPS Per-Core With AMD Zen 3 + Intel Optane Linux kernel developers have been working tirelessly to squeeze more performance out of IO_uring and the block / I/O code in general. IO_uring lead developer Jens Axboe who also serves as the Linux block subsystem's maintainer (among other roles and major contributions over the years) has used his system as a baseline for evaluating such kernel improvements. He's now moved to using AMD Zen 3 while sticking to Intel Optane storage and is seeing a mighty speed boost out of AMD's latest processors. Minecraft Now 30% Faster With Open-Source AMD Radeon Driver On Linux Those using the open-source AMD Radeon OpenGL driver "RadeonSI" on Linux the performance within the popular Minecraft game is about to be a lot better. Linux To No Longer Enable AMD SME Usage By Default Due To Problems With Some Hardware 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. 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. Linux Prepares For Next-Gen AMD CPUs With Up To 12 CCDs The latest Linux kernel patches confirm that next-gen AMD Zen processors are capable of featuring up to twelve CCDs. Radeon Linux Driver Has A Huge Optimization Two Decades Later For ATI R300~R500 GPUs While earlier this year AMD dropped pre-Polaris support from their mainline Radeon Software Windows driver, under Linux with open-source software older GPUs can live on much longer with superior driver support... Pending for Mesa 22.0 and as a surprise Christmas gift for those with nearly two decade old GPUs, a big optimization is pending for those with ATI Radeon R300/R400/R500 series graphics cards still in operation. AMD Publishes Open-Source "GPUFORT" As Newest Effort To Help Transition Away From CUDA I've just been informed by AMD that they have now made their code public to a new project called GPUFORT. This new GPUFORT project will live under the Radeon Open eCosystem (ROCm) umbrella and is their latest endeavor in helping developers with large CUDA code-bases transition away from NVIDIA's closed ecosystem. AMD Hiring For Open-Source GPU Driver Work With Mentions Of Tesla Model S, Steam Deck With AMD's increasing marketshare on the CPU and GPU front, scoring more data center wins, and also scoring custom design wins for Linux-based environments such as with the Tesla Model S and most recently with the Steam Deck, AMD continues hiring more Linux engineers. AMD Posts New "AMD-PSTATE" CPUFreq Driver Leveraging CPPC For Better Perf-Per-Watt 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. An Early Look At The GCC 12 Compiler Performance On AMD Zen 3 GCC 12 isn't seeing its stable release until around March~April as usual, but with feature development slowly wrapping up as approaching the next stage of development next month to focus on fixes, recently I wrapped up some preliminary benchmarks for how GCC 12.0 is currently performing against GCC 11.2 on an AMD Ryzen 9 5950X (Zen 3) system. 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. Valve Announces Steam Deck As Portable SteamOS + AMD Powered Portable PC Following months of rumors about new gaming hardware from Valve, today they announced Steam Deck as a new handheld PC gaming device starting at $399. 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. 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. A Prominent, Longtime Dell Linux Engineer Recently Joined AMD's Linux Team 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. 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 To Optimize C3 Entry On Linux By Finally Skipping The Cache Flush A minor optimization was posted by an AMD engineer on Wednesday for the Linux kernel. AMD still has room to improve with their Linux support particularly around the robustness of their upstream support at launch and cases like SEV-SNP support still not being yet fully upstream after the March EPYC 7003 launch or ACPI CPPC having been around since Zen 2 but only now getting to supporting it on Linux with amd-pstate, etc, but nevertheless it's been a great year of progress for AMD on the open-source side. We're also still holding out hope for more open-source system firmware efforts ahead.
| 1
| 1,760,719,328.535538
|
https://www.phoronix.com/news/amd-pstate-v7
|
AMD P-State v7 Driver Posted For Delivering Better Ryzen Efficiency In 2022
|
Michael Larabel
|
AMD on Christmas Eve posted their seventh iteration of the AMD P-State Linux driver as their new CPU frequency scaling solution for Zen 2+ to make use of ACPI CPPC for ultimately striving toward optimal power efficiency with a focus on mobile and desktop systems.
This driver was developed in cooperation with Valve and has been out for review since September. Now as we end out the calendar year they are up to the seventh revision of this driver and is appearing that it may be ready for mainline soon.
The amd-pstate driver has been evolving nicely and now working across more Zen 2 and Zen 3 systems (older processors won't be supported due to needing ACPI Collaborative Processor Performance Controls, the CPPC support must also be enabled from the system BIOS). This driver can make more informed CPU frequency scaling decisions than the much more basic ACPI CPUFreq driver currently used by Ryzen and EPYC systems.
With the v7 patches not much has changed besides altering the some macro names in the code and capitalization tweaks. Assuming no major issues come up, it's looking like this amd-pstate driver will be ready for mainlining in the near future. At that point I'll also be around with more benchmarks of this driver for raw performance and power efficiency across a range of AMD Linux systems.
| 1
| 1,760,719,330.499861
|
https://www.phoronix.com/news/AMD-STB-Linux-5.17
|
AMD Smart Trace Buffer Support Is Ready For Linux 5.17
|
Michael Larabel
|
AMD Smart Trace Buffer "STB" support is ready for the upcoming Linux 5.17 kernel cycle.
AMD Smart Trace Buffer is an APU/SoC feature for helping to isolate failures on the SoC by analyzing the last feature the system was utilizing when hitting a failure. AMD STB runs transparently in the background and a trace is then stored into the SoC for newer AMD hardware supporting this functionality. The Smart Trace Buffer trace after a hardware failure can then be read by the user via a DebugFS interface.
The code enablement patches don't make clear all what hardware is covered by AMD STB currently, but from other code hits point to it being supported in at least current generation Cezanne SoCs.
On the Radeon graphics side, initial STB support is ready for Linux 5.17 while also for this next kernel release the AMD PMC driver has integrated STB support.
This patch now in the platform-drivers-x86 for-next branch adds AMD Smart Trace Buffer support to its power management controller (PMC) driver.
The Linux 5.17 merge window is formally opening up in January while that stable kernel release should happen around the end of March.
| 0
| 1,760,719,331.298098
|
https://www.phoronix.com/news/Linux-AMD-i2c-PSP-Sharing
|
New Linux Patches For AMD i2c Bus Sharing With The PSP
|
Michael Larabel
|
The newest Linux hardware support patches for the kernel revolve around i2c bus sharing support for newer SoCs where the i2c bus is being shared by AMD's Platform Security Processor (PSP). This i2c controller is based on common DesignWare IP but new kernel code is being crafted for handling that bus sharing between the kernel and the PSP co-processor.
A new patch series is extending the existing designware i2c Linux driver for supporting the i2c controller on some newer SoCs at least of Cezanne era. The x86 cores are sharing the i2c with the PSP with the latter acting as the arbitrator for access.
The in-progress kernel code is establishing a PSP semaphore arbitration mechanism for handling the sharing of i2c support between Linux and the PSP. The notion of i2c sharing isn't entirely new but existing hardware like Intel's Bay Trail employs a similar i2c bus sharing from the platform firmware when using the X-Powers AXP288 PMIC.
This in-progress patch series working on this Designware i2c PSP support isn't from AMD directly but from Jan Dabros of Semihalf, a Polish embedded systems engineering firm.
| 3
| 1,760,719,332.455217
|
https://www.phoronix.com/news/Linux-5.17-AMD-YC-Ethernet
|
Linux 5.17 Will Add Ethernet Support For AMD Yellow Carp (Rembrandt)
|
Michael Larabel
|
AMD's Yellow Carp enablement has been going back to early summer for this next-generation APU that is better known as Rembrandt for the Ryzen 6000 mobile series. While there has already been the graphics support to land, sensor support, and various other functionality, only coming now with the next kernel cycle will be Ethernet support.
Coming seemingly late compared to the other Rembrandt / Yellow Carp feature code introduced prior cycles is now having Ethernet support, especially with wired network connectivity still rather important to many users. This Yellow Carp Ethernet support doesn't require some shiny new driver either but is being added onto the existing amd-xgbe driver. As well, Yellow Carp uses an existing PCI ID (0x14b5) but requires a few changes for properly supporting.
Compared to Raven Ridge when amd-xgbe was last altered, there are different window settings for indirect PCS access and needed to be modified for different register values. There are also slightly different port speed bits with Yellow Carp and disabling a workaround for auto-negotiation CDR.
Those few basic changes to the amd-xgbe network driver for allowing Yellow Carp (Rembrandt) Ethernet support have been merged into net-next as of yesterday. Those networking subsystem updates in turn will land in the upcoming Linux 5.17 merge window. The Linux 5.17 merge window will open up in January while that kernel will debut as stable around the end of March while Ryzen 6000 "Rembrandt" mobile APUs will be out in 2022 with Zen 3+ CPU cores and RDNA2 graphics.
| 17
| 1,760,719,332.930487
|
https://www.phoronix.com/news/amd-pstate-v6
|
AMD P-State CPU Frequency Control Driver Revised A 6th Time
|
Michael Larabel
|
Making a Sunday debut are the amd-pstate v6 patches as the latest iteration of this work for improving the AMD CPU frequency control behavior on Linux for more optimized power efficiency with modern Zen 2 / Zen 3 series (and future) processors.
The AMD P-State Linux driver is what their engineers have been iterating through the past number of months with cooperation from Valve and other stakeholders.
Compared to ACPI CPUFreq as used currently across AMD processors, the AMD-Pstate driver for use on Zen 2 and newer processors makes use of ACPI CPPC (Collaborative Processor Performance Controls) for improving AMD CPU efficiency on Linux by being able to make more more informed and accurate performance state decisions.
The amd-pstate v6 patches today have various bug fixes, restructuring of some of the code across the patch series, documentation updates, and other mostly small changes at this stage.
Those interested in trying out the newest AMD P-State Linux patches can find the v6 series on the kernel mailing list.
We'll see how much more work is ahead before ready for mainlining in the Linux kernel. Linux ACPI / power management subsystem maintainer Rafael Wysocki did comment a few days ago of some of the patches still needing additional sign-offs before he is ready for pulling in the patches for working their way upstream.
| 40
| 1,760,719,334.093849
|
https://www.phoronix.com/news/AMD-Accelerator-Cloud
|
AMD Launches The Accelerator Cloud To Try Out EPYC CPUs, Instinct GPUs + ROCm
|
Michael Larabel
|
AMD has made public the AMD Accelerator Cloud. No, they aren't getting into the cloud game per se, but rather allowing a place for customers to try out new EPYC processors and AMD Instinct accelerators running with the latest ROCm software components.
With the AMD Accelerator Cloud are AMD's latest wares both in the form of their newest processors and GPUs/accelerators as well as their latest software stack deployed for ensuring a turn-key trial of AMD's offerings.
AMD wrote in their announcement, "The AAC is designed to alleviate many of those potential lingering questions by offering customers an environment to try their code on AMD Instinct accelerators and AMD ROCm software to see how it performs. Whether it’s porting legacy code, benchmarking an application or testing multi-GPU or multi-node scaling, the AMD Accelerator Cloud gives you the tools to make a confident decision."
AMD intends to complement the AAC with ROCm training sessions and hackathons, including the possibility of private events for prospective customers. Accessing the AMD Accelerator Cloud requires a registration process and AMD evaluating all requests. The focus of the AMD Accelerator Cloud at this point is all about HPC and AI workloads. Coming to the AAC soon will be Instinct MI200 accelerators.
This is great especially with many HPC/AI developers not being too experienced with AMD's current software offerings around the Radeon Open eCosystem. The AMD Accelerator Cloud will allow customers to better evaluate the software and hardware combination together prior to making any big investments. Granted, hopefully AMD will soon offer long-awaited ROCm support for RDNA/RDNA2 GPUs and the like to better allow developers to try out ROCm code on their own consumer hardware, just as NVIDIA's CUDA has offered ubiquitous support over the years unlike AMD's current very narrow scope of GPUs supported by ROCm.
More information on the AMD Accelerator Cloud is available at community.amd.com.
| 30
| 1,760,719,334.451769
|
https://www.phoronix.com/news/Linux-Kernel-Drop-AMD-3DNow
|
Linux Kernel Set To Finally Retire AMD 3DNow!
|
Michael Larabel
|
Queued up as part of the x86/core changes intended for the Linux 5.17 cycle is dropping of the AMD 3DNow! code within the kernel. While 3DNow! brings back fond memories from the days of AMD's K6 and early Athlon processors, AMD deprecated the instructions a decade ago and no longer found in newer processors. Removing of the 3DNow! kernel code is being done as part of some code improvements.
Hitting tip's x86/core branch is this commit dropping the 3DNow! code from the kernel. 3DNow! built upon MMX instruction set to offer faster performance for vector processing of floating point data. But 3DNow! ultimately didn't see too much adoption amid Intel's successful SSE introduction.
The X86_USE_3DNOW option is removed that defaulted to on for kernel builds targeting now-old processors like the Geode LX, K7, and Cyrix III. This X86_USE_3DNOW option enabled the 3DNow! helper library code in the kernel and allowed for MMX 3DNow! accelerated memcpy usage and accelerated clearing and copying of pages.
Removing this 3DNow! support from the kernel allows freeing around 500 lines of code that had been in the kernel for over two decades. 3DNow! was found in AMD processors from the K6-2 up until Bulldozer where it was retired. Now it's time to let the kernel code go come 2022 with the Linux 5.17 cycle.
| 66
| 1,760,719,336.031535
|
https://www.phoronix.com/news/AMD-AOCC-3.2
|
AMD AOCC 3.2 Compiler Released Along With AOCL 3.1 CPU Libraries
|
Michael Larabel
|
AMD has issued a nice end-of-year update to the AMD Optimizing C/C++ Compiler (AOCC) that also includes Fortran support as well as a new release of their AMD Optimizing CPU Libraries (AOCL).
AMD AOCC 3.2 is the new version of their LLVM Clang downstream focused on providing the latest optimized compiler support for Zen-based processors whether it be EPYC, Ryzen, or Ryzen Threadripper processors. With AOCC 3.2 they have re-based to LLVM/Clang 13.0 as their compiler release. LLVM 13.0 was released this autumn and is the latest stable LLVM version. AMD now has all of their yet-to-be-upstreamed or unsuitable for upstream patches re-based on this new version.
AOCC 3.2 features a number of improvements around its Flang-based Fortran compiler front-end for LLVM, including various sanitizers working. The (NO)FREEFORM pragma is also now supported and there is improved debug support. OpenMP 4.5 is also now working for Fortran and better support for the Fortran 2008 language version.
AOCC 3.2 remains binary-only and that AMD official Zen code compiler can be downloaded from developer.amd.com. I'll be working on some fresh AOCC 3.2 comparison compiler benchmarks in the days ahead.
AMD also released AOCL 3.1 as their Optimizing CPU Libraries. AOCL 3.1 consists of their optimized versions of BLIS, libFLAME, FFTW, LibM, Sparse, ScaLAPCK, MUMPS, and other optimized libraries. These various libraries popular for different math purposes and used in numerical computing and HPC feature various optimizations aimed for enhancing the performance on AMD Zen-based processors. AOCL 3.1 is available for download as well from developer.amd.com.
| 3
| 1,760,719,336.399467
|
https://www.phoronix.com/news/AMD-EDAC-RDDR5-LRDDR5
|
AMD Linux EDAC Driver Prepares For Zen 4, RDDR5 / LRDDR5 Memory
|
Michael Larabel
|
AMD's Linux engineers continue preparing for next-gen EPYC server processors based on Zen 4 and supporting DDR5 memory.
In addition to recent work like preparing for up to 12 CCDs per socket, temperature monitoring, and other bits, out today is a set of patches for AMD's EDAC (Error Detection and Correction) driver code for the next-generation Zen 4 server processors.
The work sent out today includes adding support for RDDR5 and LRDDR5 memory support to the driver (conventional DDR5 support was already mainlined). This is for Registered DDR5 memory support as well as Load-Reduced DDR5 memory support. LRDDR5 support is for the higher memory density servers, similar to LRDIMMs with prior DDR generations.
The patches do also confirm up to twelve memory controllers per socket with the next-gen processors, compared to the current limit of eight.
With these additions being squarely for next-generation parts and RDDR5/LRDDR5, it does further solidify that Zen 4's Family/Model IDs as AMD Family 19h Models 10h-1Fh and A0h-AFh. These IDs were previously reported on Phoronix in relation to other Linux patches, which some doubted were accurate for Zen 4 given that it's Family 19h still, but is similar to what was seen with 17h for Zen/Zen+/Zen2, etc.
The patches are now out on the kernel mailing list for adding the new IDs and RDDR5/LRDDR5 support to the AMD EDAC driver and in turn will likely be mainlined for the upcoming 5.17 cycle given no surprises.
| 8
| 1,760,719,337.904338
|
https://www.phoronix.com/news/AMD-SMCA-Different-Layouts
|
AMD Makes Some Interesting SMCA Driver Changes For Future CPUs
|
Michael Larabel
|
AMD is preparing updates to their SMCA (Scalable Machine Check Architecture) driver code for future CPUs and points to processors having different bank layouts between CPU cores on the package.
A set of three patches sent out by AMD last week add new SMCA bank types in preparation for new CPUs and also bank layout changes for future AMD systems. What makes these patches interesting is, "Future AMD systems will have different bank type layouts between logical CPUs. So having a single system-wide cache of the layout won't be correct....Future AMD systems will lay out MCA bank types such that the type of bank number may be different across CPUs."
The patches though don't elaborate though on why future AMD CPUs may begin seeing different bank layouts between the logical cores of the system. It is a real possibility though that it's in relation to a hybrid processor design with a combination of different cores, similar to Alder Lake / Arm big.LITTLE. AMD patents over the past two years along with rumors have alleged AMD is developing such a hybrid processor and that could be one explanation for why future AMD CPUs may be seeing the different bank layouts between cores.
When it comes to the new SMCA bank types added for future processors as part of this same patch series. the additions include: MPDMA unit, NBIF unit, system hub unit, SATA unit, USB unit, GMI PCS unit, and GMI PHY unit.
These latest AMD machine check error driver patches can be found on the kernel mailing list as they work their way towards the mainline Linux kernel.
| 12
| 1,760,719,337.951458
|
https://www.phoronix.com/news/AMD-s2idle-One-Line-Fix
|
One-Line Linux Patch Fixes AMD s2idle Failures For Some Ryzen Laptops
|
Michael Larabel
|
Over the past year there has been a lot of work for getting AMD's suspend-to-idle "s2idle" support in order under Linux and the latest is a one-line code change expected to help at least some Ryzen laptops behave properly.
Over the past year has been a lot of AMD s2idle work for increased power-savings under Linux, which has meant a lot of fixing and code handling. The latest is a "lucky fix" with a one line change that is fixing some s2idle failures.
This patch to cut the AMD PMC driver polling delay from 100 micro-seconds to 50 micro-seconds is enough to fix some laptops from experiencing s2idle failures, potentially more common when running on battery power.
Fabrizio Bertocci who reported the issue and ultimately submitted a patch to change the default delay found this fixed his HP Pavilion Aero Laptop 13 with Ryzen 7 5800U for s2idle on battery power. The prior 100us delay worked fine on AC power, but would fail on battery power. After debugging this issue with AMD Linux developers, by chance it was figured out that simply halving this delay would workaround the problem.
The debugging adventure for this current-generation HP laptop can be found via this Gitlab issue thread. The patch changes the default value unconditionally and it's not known how many different AMD Ryzen laptops may be affected, so if you have had s2idle failures it may be worth trying out. The patch is currently in the x86 platform drivers tree as a fix and is marked for going to the stable branches once merged, so in the near future it should ultimately work its way out to supported stable Linux kernel series.
| 8
| 1,760,719,339.512865
|
https://www.phoronix.com/news/amd-pstate-v5
|
AMD-Pstate Driver Updated A 5th Time For Improving Ryzen Power Efficiency On Linux
|
Michael Larabel
|
Sent out today was the fifth revision to AMD's new "amd-pstate" kernel driver focused on providing enhanced CPU frequency controls for Linux systems.
AMD's P-State driver remains under active development for improving the Linux power efficiency for Ryzen (and EPYC) processors. AMD P-State makes use of ACPI CPPC for more informed and finer-grained frequency controls on modern (Zen 2 and newer) processors compared to what is afforded by the existing ACPI CPUFreq frequency scaling driver currently used by AMD Linux systems.
AMD has been working on this new Linux driver with partners like Valve for ensuring optimal power efficiency on the Steam Deck and other AMD-powered hardware. With today's v5 patches, there is a build warning fix when building the Linux kernel under LLVM Clang, updated documentation, and other fixes.
See the patch series for the latest work on the AMD P-State Linux driver. From last week are some power and performance tests of amd-pstate with a Ryzen laptop.
| 5
| 1,760,719,341.094618
|
https://www.phoronix.com/news/AMD-AMF-SDK-1.4.23
|
AMD AMF SDK 1.4.23 Brings Main 10 HEVC Encode, Auto LTR Encoder Mode
|
Michael Larabel
|
AMD on Friday published a new version of their Advanced Media Framework "AMF" software development kit that enhances the multimedia processing capabilities for Radeon hardware.
AMD AMF continues to support both Windows and Linux and supporting interoperability with multiple APIs including DirectX, Vulkan, OpenGL, and OpenCL. As the first AMF update since this summer, AMF 1.4.23 is rather noteworthy in now adding an Auto LTR encoder mode as well as Main 10 HEVC encoder profile.
The full set of official changes for AMD AMF 1.4.23 include:
- New Auto LTR encoder mode
- FFmpeg updated to version 4.3.2
- Additional statistics and feedback in encoder
- New encoder usage presets
- Main 10 HEVC encoder profile support
- Support for encoder instance selection AMD's Advanced Media Framework tends to not be too useful for Linux users, most of whom have their video acceleration needs fulfilled simply by using the Gallium3D video acceleration support within Mesa and thus very easy to deploy and use across distributions. But if you are making use of the AMF SDK, you can find this new v1.4.23 release via AMD's GPUOpen.com.
| 21
| 1,760,719,341.565173
|
https://www.phoronix.com/news/AMD-Zen-12-CCDs-hwmon
|
Linux Prepares For Next-Gen AMD CPUs With Up To 12 CCDs
|
Michael Larabel
|
The latest Linux kernel patches confirm that next-gen AMD Zen processors are capable of featuring up to twelve CCDs.
Currently the k10temp Linux driver for temperature monitoring of AMD processors can handle up to eight core-complex dies (CCDs) while next-gen AMD Zen processors will have configurations up to 12 CCDs to accommodate higher core counts.
The latest Linux patches explicitly say, "The newer AMD Family 19h Models 10h-1Fh and A0h-AFh can support up to 12 CCDs. Update the driver to read up to 12 CCDs."
For those new Family 19h model IDs, they debuted in Linux patches earlier this month as a "new generation". While not said specifically, this should be for next year's Zen 4 processors.
These basic patches get the k10temp support in place for handling up to 12 CCDs. This doesn't confirm the CCD to CCX layout for Zen 4, but obviously with 12 CCDs will be for next-gen EPYC and/or Threadripper processors. AMD announced earlier this month, EPYC "Genoa" processors will feature up to 96 Zen 4 cores while EPYC "Bergamo" will feature up to 128 Zen 4 "C" cores. Given the higher core counts compared to 64 cores with current EPYC/Threadripper parts, to not much surprise is the higher CCD count coming.
Expect more of the exciting Zen 4 enablement patches to come to the Linux kernel as the product debuts approach in 2022.
| 87
| 1,760,719,342.682729
|
https://www.phoronix.com/news/AMD-P-State-v4
|
Updated AMD P-State Driver Posted For Improving Linux Power Efficiency
|
Michael Larabel
|
A fourth iteration of the AMD P-State CPU frequency scaling driver patches for Linux have been sent out for review and testing.
This is the amd-pstate driver aiming for better power efficiency on Linux by leveraging ACPI CPPC found with Zen 2 and newer processors. Valve collaborated with AMD on the creation of this new driver that aims to be superior to the generic ACPI CPUFreq driver currently used by AMD processors. This driver has been undergoing public review since September with aims to make it to the mainline kernel.
With today's amd-pstate v4 series, the driver has been re-based against the upstream Linux 5.15 state for the power management branch, cleaning up some of the code, introduction of an AMD CPPC support library, an option to disable shared memory support, and other low-level improvements to the driver code. The v4 series can be found via the kernel mailing list.
It's been a revision or two ago since I last tested the amd-pstate driver to mixed success, so will take these fresh patches for a spin shortly. The timing of this driver though makes it out of reach for mainlining in Linux 5.16 but we'll see if it's squared away in time for v5.17.
| 31
| 1,760,719,343.336788
|
https://www.phoronix.com/news/Linux-5.17-k10temp-New-AMD-Zen
|
Linux 5.17 To Support Temperature Monitoring For New AMD Zen Generation
|
Michael Larabel
|
The Linux 5.17 kernel next year will support temperature monitoring for a "new generation" of AMD Zen processors.
While AMD has often been late to the game in supporting CPU temperature reporting under Linux for Zen processors, it's nice to see them out in front ahead of their next launch. Even in cases where new IDs simply need to be added to the k10temp driver, unfortunately they have often not added them until post-launch or in some cases where those in the community (including cases like I when getting hands on review samples) have the hardware and find the support not working until making some trivial driver alterations.
Recently though AMD jumped ahead of the game and added Yellow Carp (Rembrandt) temperature monitoring with Linux 5.15 ahead of launch. Now patches queued in hwmon-next for Linux 5.17 are continuing in that direction by adding support for a new generation of Family 19h processors.
This patch adds the PCI IDs for Family 19h Models 10h-1Fh and A0h-AFh, noting it's for the "new generation".
This patch gets the k10temp-based CPU temperature monitoring working for those new AMD Family 19h CPU models. The new generation is using the 0x300 CCD offset that first appeared for Yellow Carp for reading the thermal information.
It's quite possible these new Family 19h models are for Zen 4 or we'll find out in due course, but for now that's all there is to report on the matter and in any case it's good to see these basic additions being made ahead of launch. The patches are in the hardware monitoring subsystem's "-next" area until the Linux 5.17 cycle kicks off around the start of the new year and that kernel will then reach stable by the end of March.
| 13
| 1,760,719,344.312
|
https://www.phoronix.com/news/AMD-Sound-Open-Firmware-Renoir
|
Sound Open Firmware For AMD Audio Hardware Arrives, Initially For Renoir ACP
|
Michael Larabel
|
Back in 2018 Intel founded Sound Open Firmware as their effort to provide an open-source audio DSP firmware and software development kit. AMD has begun supporting Sound Open Firmware too now, initially for the Renoir audio co-processor (ACP).
Sound Open Firmware as a Linux Foundation project has been maturing over the past three years and now supports a wide-range of Intel hardware with other audio hardware also becoming supported. Ultimately it's about having open-source audio DSP firmware and a SDK to better support modern audio processing. In the SOF documentation it's summed up rather broadly, "The Sound Open Firmware SDK is comprised of many ingredients that can be customized for use in the firmware/software development lifecycle. Customization allows for a “best fit” development approach where the SDK can be optimized for a particular process or environment. Some SDK ingredients are optional while there can be more than once choice for other ingredients."
With last month's Sound Open Firmware 1.9 release they began adding support for AMD Renoir hardware along with the NXP i.MX8ULR, Mediatek MT8195, and on the Intel side is Alder Lake support. There has also been work underway for supporting Sound Open Firmware on Microsoft Windows.
This week the patch series was sent out providing the SOF kernel driver support for the Renoir audio co-processor hardware.
With this being the first AMD audio DSP platform supporting Sound Open Firmware, it also adds the necessary build infrastructure and the new Kconfig switches. AMD is putting their Sound Open Firmware kernel driver code for Renoir out under a GPLv2-only or BSD 3-clause license.
This is just the latest of a fair number of AMD audio patches we've seen for Linux from new hardware support to making improvements such as this SOF bring-up. With Google using Sound Open Firmware for Chromebooks, this is possibly among the reasons why AMD is investing now in SOF audio platform support. Given the timing of these kernel patches, the soonest we'll see this Renoir ACP SOF support land would be in the v5.17 cycle.
| 9
| 1,760,719,344.964855
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.