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-AOCC-1.3-Compiler
|
AMD Optimizing C/C++ Compiler 1.3 Brings More Zen Tuning
|
Michael Larabel
|
Earlier this month AMD quietly released a new version of their Optimizing C/C++ compiler in the form of AOCC 1.3. This new compiler release has more Zen tuning to try to squeeze even more performance out of Ryzen/EPYC systems when using their LLVM-based compiler.
The AMD Optimizing C/C++ Compiler remains AMD's high performance compiler for Zen compared to the earlier AMD Open64 Compiler up through the Bulldozer days. AOCC is based on LLVM Clang with various patches added in. Fortunately, with time at least a lot of the AOCC patches do appear to work their way into upstream LLVM Clang. AOCC also has experimental Fortran language support using the "Flang" front-end that isn't as nearly mature as Clang.
With the AMD AOCC 1.3.0 release there is more Zen tuning, enhanced loop optimizations, better vectorization and code generation, continued work on Flang as well as the DragonEgg GCC plug-in, optimizing AMD Math library, the LLVM Linker is now used by default, and their code has been re-based to LLVM/Clang 7.0.
My benchmarking of the AOCC has been mixed. When AMD first introduced AOCC there wasn't much of a difference compared to the upstream open-source LLVM Clang compiler, but with succeeding releases it has deviated a bit more and for at least some HPC/workstation workloads building with AOCC does offer performance advantages. I'll be running some AOCC 1.3 compiler benchmarks shortly on Threadripper and EPYC to see how this new release performs compared to the Clang and GCC compilers.
Those wanting to try out the AOCC 1.3 compiler release can find it at developer.amd.com.
| 2
| 1,760,719,598.851107
|
https://www.phoronix.com/news/AMD-Next-Horizon-Event-Notes
|
AMD Next Horizon: Zen 2, 7nm Vega, AMD On Amazon EC2
|
Michael Larabel
|
AMD's "Next Horizon" investor day event is taking place right now in San Francisco. Here are the highlights from this AMD event as we approach the end of 2018.
Highlights so far from this event include:
- AMD EPYC CPUs will be coming online with Amazon AWS' Elastic Compute Cloud (EC2) starting today for the R5a, M5a, and T3a instance types.
- AMD Radeon Instinct MI60 is the first 7nm Vega GPU.
- The initial preview of the AMD Rome next-gen 7nm EPYC CPU.
- Zen 2 is sampling today while it's on track for being released still in 2019.
- 7nm slated to deliver 2x density improvement, 0.5x power, and greater than 1.25x the performance.
- Zen 3 7nm++ is "on track".
- Zen 2 will have better floating point performance.
- Hardware-enhanced Spectre mitigation for Zen 2 along with other security improvements.
- Zen 2 is using a "revolutionary" chiplet design.
- Zen 4 is in design stages already.
- The Radeon Instinct MI60 (Vega 20) supports hardware virtualization, 2x more density, 50% lower power, and greater than 1.25x higher performance. This GPU offers end-to-end ECC memory protection, and machine learning instructions for training and inference.
- The MI60 with its 32GB HBM2 memory can achieve 1TB/s of memory bandwidth.
- The MI60 supports full PCI Express 4.0 with 64GB/s bi-directional bandwidth and does support Infinity Fabric Links at 100GB/s per link between the EPYC CPU and GPU.
- The ROCm 2.0 compute stack is coming for the MI60. This is the latest open-source Radeon Compute stack for machine learning and more.
- "We are committed to open-source and the open-source community."
- ROCm 2.0 will be out before the end of 2018.
| 76
| 1,760,719,599.291769
|
https://www.phoronix.com/news/GCC9-Merges-znver2-Zen-2-CPU
|
AMD Zen 2 CPU Support Merged To GCC 9 (-march=znver2)
|
Michael Larabel
|
It was just days ago that AMD published their Zen 2 compiler patch for the GCC compiler but with the race on to merge new feature code before the feature freeze happening later this month, that "znver2" tuning patch has now been merged to mainline.
The latest GCC 9.0.0 development code as of this Sunday morning now has the "-march=znver2" support for generating optimized binaries for the yet-to-be-released AMD Zen 2 processors. As covered in the article earlier this week, the Zen 2 GCC patch is an initial first stab at supporting the new AMD CPU micro-architecture in the GNU Compiler Collection.
Most of the patch is based on the existing "Zen 1" (znver1) target but it does support at least three of the new CPU instructions to be supported by these next-gen CPUs... Zen 2 brings support for cache line write back (CLWB), read processor ID (RDPID), and write back and do not invalidate cache (WBNOINVD). It's possible there are other new instructions coming that AMD isn't yet ready to reveal so they didn't include it as part of this initial patch.
The scheduler costs table and other tuning also still needs to be firmed up for Zen 2 while this initial code is good for starters and at least having this base znver2 support in time for GCC 9, which will ship as stable around the end of Q1'2019 with the initial stable version being GCC 9.1.0. The first AMD Zen 2 CPUs expected to launch in early 2019 are the new 7nm EPYC "Rome" processors.
Stay tuned to Phoronix for more Zen 2 coverage as the Linux/open-source enablement continues.
| 0
| 1,760,719,600.673467
|
https://www.phoronix.com/news/Air-Cooling-Threadripper-2970WX
|
Noctua Air Cooling With The AMD Ryzen Threadripper 2970WX
|
Michael Larabel
|
With this week's launch of the Ryzen Threadripper 2920X and 2970WX, the new 24-core / 48-thread 2970WX has a 250 Watt TDP like the 2990WX. Fortunately, with Noctua's high-end TR4/SP3 heatsinks, it's still possible to get by with air cooling.
For those wondering about air cooling with the AMD Threadripper 2970WX, I did some fresh thermal tests this week with the various Noctua TR4-SP3 heatsinks that work for AMD EPYC and Threadripper processors. Today's tests are complementary data to our earlier Cooling AMD EPYC with Noctua heatsinks and Threadripper cooling benchmarks done in the past.
For this 2970WX testing was the Noctua NH-U14S TR4-SP3, NH-U12S TR4-SP3, NH-U9S TR4-SP3, and a second run of this smallest NH-U9 TR4-SP3 when the fans were set to run at their full fan-speed rather than the "SmartFan" settings from the BIOS. With the NH-U9 TR4-SP3 is the smallest, it's my favorite Threadripper/EPYC heatsink because it fits within 4U height requirements. I am currently running four NH-U9 TR4-SP3 heatsinks in various Threadripper/EPYC systems and they continue doing great: I haven't found a better workstation cooler for these processors that fits within 4U height for rackmount systems.
The system was running within a SilverStone CS350 chassis.
To little surprise, the largest of these heatsinks - the NH-U14S TR4-SP3 - led to the coolest results.
In compiling LLVM with 48 threads, the NH-U9S TR4-SP3 did appear to have a bit of thermal throttling given the variation in the results. Fortunately, when running the two Noctua 92mm fans at full-speed the performance was in line with the other coolers.
In compiling GCC, no thermal throttling was encountered.
The thermal results in some of the other workloads tested where no thermal throttling was encountered on any of these Noctua heatsinks.
With Blender 3D modeling, the full-speed fans on the 4U compatible cooler was also necessary to avoid reduced performance.
Lastly is a look at the data overall during the course of many benchmarks carried out. All of the individual data sets can be found via this OpenBenchmarking.org result file. These Noctua EPYC/Threadripper heatsinks can be found from the likes of NewEgg.
| 2
| 1,760,719,600.86585
|
https://www.phoronix.com/news/AMD-Zen-GCN-CodeBench-Lite
|
Mentor Graphics Releases CodeBench Lite For AMD EPYC/Ryzen & Radeon Instinct Targets
|
Michael Larabel
|
Mentor Graphics announced the availability today of their GNU toolchain based CodeBench Lite Edition that is a development environment intended for HPC application development that now supports targeting for AMD Ryzen/EPYC processors as well as Radeon Instinct GPUs.
The CodeBench Lite Edition provides C/C++ and Fortran tool-chains for compiling high-performance computing applications for multi-core heterogeneous architectures. CodeBench Lite works on Linux and is their first time now supporting the AMD GCN graphics offloading target with support for Radeon Instinct GPUs in particular.
OpenMP and OpenACC are the current multi-threading/offloading APIs being targeted with the C/C++ and Fortran language support. More details in today's press release.
This Radeon Instinct GPU targeting doesn't come as a big surprise considering we've been reporting for months about Mentor / Code Sourcery working on a new AMD GCN back-end for GCC. That GCN GPU offloading target/back-end for GCC hasn't yet been merged to mainline GNU Compiler Collection but we hope it will still happen before the GCC 9 feature branching later this month. So ultimately mainline GCC should end up having as good of coverage for AMD CPU/GPU hardware, but for now it's all shipped in an easy-to-use binary form by Mentor Graphics with a focus on HPC developers.
| 0
| 1,760,719,602.206536
|
https://www.phoronix.com/news/AMD-Zen-2-Linux-Temp-Driver
|
AMD Sends Out Linux Temperature Driver Patches For Zen 2 CPUs
|
Michael Larabel
|
It looks like my report on AMD kicking off Zen 2 Linux enablement from last month is panning out. Earlier this week they posted the initial AMD Zen 2 "znver2" support for the GCC compiler and they are ending the week back in kernel space with an updated hardware monitoring driver for being able to report the CPU core temperatures of these CPUs shipping in 2019.
The patch series sent out a short time ago add k10temp support for AMD F17h M30h. The k10temp driver is the "hwmon" subsystem driver for reading the CPU core temperature on recent generations of AMD hardware, including Zen 1 and soon Zen 2. Family 17h Model 30 and newer is Zen 2, which is also firmed up by the aforementioned znver2 compiler patches.
These four patches to k10temp allow for Zen 2 CPU core temperatures to be reported to Linux user-space. The patches are larger than normal due to the new processors reporting multiple roots per Data Fabric / SMN interface, so some basic tweaks had to be applied to the querying logic. But these kernel patches don't reveal any new surprises about Zen 2 that we didn't already know, just new PCI IDs.
It's great to see AMD getting this k10temp driver support out ahead of the processors shipping, which will begin in H1'2019 with AMD EPYC 2 "Rome" processors. With the original Zen/Ryzen launch, k10temp support didn't come until months after launch. Even for this year's Threadripper 2 launch the (quite simple) k10temp code wasn't even sent in early and I ended up submitting that patch to get the support squared away in the Linux kernel. So seeing AMD themselves sending in this Zen 2 k10temp code months before these next-gen CPUs launching is a good sign the Linux support is going to come together nicely.
These k10temp patches aren't candidates for merging in the current Linux 4.20/5.0 development cycle but will be material for merging in the follow-on kernel cycle in late December or early January, presumably with more Zen 2 patches in tow.
| 2
| 1,760,719,602.597218
|
https://www.phoronix.com/news/Threadripper-2-ECC-DDR-Fail
|
The AMD Threadripper ECC DDR4-2666 Testing That Wasn't
|
Michael Larabel
|
Recently in our forums there has been a lot of interest in Threadripper 2 builds using ECC DDR4 memory and the impact on performance, especially now with the Threadripper 2 family being rounded out by the 2920X and 2970WX. So I set out to do some DDR4-2666 ECC UDIMM testing with Threadripper 2, but that hasn't turned out well.
Due to the growing interest in Threadripper 2 + ECC memory builds by Linux users (especially among the supportive premium crowd), I ordered a quad-kit of DDR4-2666 ECC UDIMM for testing... While the latest Threadripper 2 processors support DDR4-2933, DDR4-2666MHz is the fastest that can currently be found for ECC UDIMMs. As I only have a couple older DDR4-2133 ECC UDIMMs sticks in some Xeon E3 v5 boxes and the rest of my DDR4 ECC being Registered DIMMs that are not supported by Threadripper CPUs, I ordered some.
But even finding DDR4-2666 ECC UDIMMs proved particularly difficult as most US Internet retailers listing them are marked discontinued or out-of-stock with DDR4-2400 being more common. Most of the memory compatibility lists from the different motherboard vendors with X399 products also only lists DDR4-2400 models when it comes to ECC UDIMMs.
When shopping for the memory, the only prominent Internet retailer listing DDR4-2666 ECC UDIMMs was NewEgg with several different listings for NEMIX RAM. NEMIX RAM advertises itself as a Florida-based memory product company founded in 2007 with its products backed by a lifetime warranty. I never touched NEMIX memory before but given no other viable option, I ordered a quad kit of their 8GB DDR4-2666 ECC Unbuffered memory.
The memory arrived after opting for overnight shipping. Unfortunately I have yet to get this memory working in any of my Threadripper X399 systems. I've tried the MSI X399 MEG CREATION, ASUS ROG ZENITH EXTREME, and Gigabyte X399 Aorus Gaming 7 to no avail either in quad or dual-channel configurations. Trying this memory on an MSI C236A WORKSTATION board also did not work.
I am waiting on hearing back from NEMIX if there is any magic or an RMA is hopefully in store.
With the odd over-hanging placement of their memory stickers and feeling the stickers on each of the DIMMs, it became clear that there was a second sticker behind. When carefully peaking behind, this does indeed appear to be some re-branded memory that is being sold. When looking back on NewEgg, others have also reported their memory kits having been re-labeled, RMAs going through other companies, etc.
So unfortunately it doesn't look like any ECC Threadripper 2 benchmarking will be coming up anytime soon until finding some working DDR4-2666+ ECC UDIMMs... As consolation, here are some fresh DDR4 (non-ECC) Threadripper 2990WX Linux memory benchmarks with some G.SKILL DIMMs at 2 x 8GB DDR4-3200, 4 x 8GB DDR4-2666, 4 x 8GB DDR4-2933, 4 x 8GB DDR4-3200 that I started before abandoning the tests due to the non-working ECC DIMMs.
| 31
| 1,760,719,603.640647
|
https://www.phoronix.com/news/AMD-znver2-GCC-Patch
|
AMD Publishes Zen 2 Compiler Patch "znver2" Exposing Some New Instructions
|
Michael Larabel
|
With GCC 9 feature development ending in November, AMD today sent out their first patch enabling Zen 2 support in the GNU Compiler Collection via the new "znver2" target.
This is the basic patch introducing the next-generation AMD Zen CPU to the GCC compiler collection. At this stage it's just the basic implementation and carries over the same cost tables and scheduler data from Znver1. So it doesn't reveal any major breakthrough changes, but in digging through the code, it does confirm some new CPU instructions that will be supported by these next-gen Zen CPUs... On top of the Znver1 instructions, Zen 2 is adding:
- Cache Line Write Back (CLWB)
- Read Processor ID (RDPID)
- Write Back and Do Not Invalidate Cache (WBNOINVD)
That's it in terms of new instructions, at least what's enabled by these patches. It's possible there might be some other new instructions supported by Zen 2 that AMD doesn't want to reveal at this time, just like the scheduler cost tables haven't yet been tuned, etc. This patch is basically a starting point so the GCC 9.1 stable update due out in 2019 can at least handle -march=znver2 and that march=native targeting will also work for these next-gen AMD processors.
The first AMD Zen 2 processors expected are 7nm EPYC 2 and we should be hearing more about them in early 2019... Given all the successes we've seen with Threadripper and EPYC 7000 series on Linux already, I am especially excited to see what next-gen EPYC will have in store and how fast it will be.
The patch is currently on gcc-patches but will likely be merged to mainline GCC before the feature freeze goes into effect at the middle of November. The timing of this patch does also reinforce what I reported recently with it looking like AMD has begun their upstreaming / open-sourcing of Zen 2 enablement for the Linux kernel and related components of the open-source toolchain.
| 7
| 1,760,719,604.108393
|
https://www.phoronix.com/news/LLVM-Clang-2018-Bdver2-Tuning
|
Six Years After Launch, AMD Piledriver CPU Tuning Gets Reworked In LLVM Clang
|
Michael Larabel
|
Six years after AMD introduced "Piledriver" as the successor to the original Bulldozer CPUs, the LLVM Clang compiler is seeing a revised scheduling model for these processors that can yield faster performance of generated code targeting this older class of AMD CPUs.
Piledriver cores ended up a range of CPUs from the FX-8300 series through the FX-9590, many APUs including the A10-6800K, more than a dozen mobile parts, and also some Opteron CPUs. Piledriver as a reminder was based on a 32nm SOI process, offered better IPC over the original Bulldozer microarchitecture, bumped the clock speeds, and other incremental improvements.
This 220 Watt beast might now run faster on Linux... Roman Lebedev, a Darktable software developer, took to optimizing LLVM Clang with a focus on speeding up the Piledriver CPU performance in handling the open-source RAW photography software's image decoding speed. With tweaking the Piledriver/bdver2 scheduler mode, he got the generated code performance to improve by about 1% while in the most significant test cases it was up to 7% faster.
The revised Piledriver scheduler model is now in LLVM master for the next release that will be LLVM/Clang 8.0 in early 2019.
For all the older AMD systems out there with Piledriver cores, the newer LLVM Clang compiler may generate more optimal code when using the "bdver2" targeting. What also makes this scheduling work interesting is that he was able to optimize the model by making use of new LLVM tooling like llvm-exegesis for helping to profile the host machine's instruction characteristics using the system's performance counters. This new profiling/benchmarking means is rather than traditionally relying upon the data provided by the CPU vendor about optimal CPU instruction characteristics or profiling by hand. Hopefully this approach will help optimize other CPU scheduler models for LLVM/Clang moving forward especially as LLVM's vast collection of tooling continues to mature.
| 18
| 1,760,719,605.144724
|
https://www.phoronix.com/news/AMD-EDAC-Kicks-Off-Zen-2
|
It Looks Like AMD's Linux Developers Have Begun Work On Zen 2 / EPYC 2 "Rome" Support
|
Michael Larabel
|
Ahead of the Zen 2 processors expected in 2019, it appears AMD developers have begun working on their Linux kernel support patches for these next-generation CPUs. In particular, it appears the flow of Linux kernel code for supporting EPYC 2 "Rome" processors has begun.
Sent into the mainline Linux kernel this morning were the EDAC updates for Linux 4.20~5.0. EDAC in this context is for Error Detection And Correction - error reporting functionality mostly useful for server/workstation platforms. Usually the EDAC changes aren't worth writing about on Phoronix, but I did notice one of the changes standing out: "Add Family 17h, models 10h-2fh support."
That is building upon a previously reported patch for Model 11h "Great Horned Owl" support for EDAC, but now there are all these extra models added... When checking the latest model data, the later Family 17h Models up through 2Fh (47) are indeed for Zen 2. The patch itself for the AMD64 EDAC code is just adding the new models but with no real logic changes compared to Zen 1, thus no new architectural details are revealed by this new code.
With this being to the EDAC code, this is obviously more with server CPUs in mind, i.e. EPYC 2 (Rome), than for Ryzen desktop processors.
So far I haven't seen any other Linux kernel patches pertaining to these other Family 17h models aligned for Zen 2. Well, with this Linux kernel there is also initial xGMI support on the AMDGPU graphics driver side for Vega 20. It's also believed EPYC 2 platforms will support the xGMI interconnect but at this stage I haven't seen any xGMI Linux patches outside of the AMDGPU driver code.
The timing does make sense that we should begin seeing more Zen 2 enablement code landing soon. Especially on server platforms where customers don't tend to be running bleeding edge kernels, it's important to get that driver enablement support in place well ahead of the hardware's availability so it can be back-ported where appropriate to the enterprise/LTS kernels and all around stabilized and widely available by the time the new CPUs hit the shelves given the release cadence of the Linux kernel.
Anyhow, as any other Zen2/Rome patches come about to either the Linux kernel, GCC / LLVM compilers, or other software components, I'll certainly be writing about them as we look forward to the next-gen CPU launch in 2019.
| 23
| 1,760,719,605.821377
|
https://www.phoronix.com/news/AMD-Hygon-Dhyana-420-Queue
|
The AMD Zen-Based Hygon Dhyana CPU Support Landing In The Mainline Linux Kernel
|
Michael Larabel
|
Hygon's Dhyana SoC, the facsimile of the AMD Zen microarchitecture as a result of the AMD-Chinese joint venture to begin spinning up domestic x86 chips for the Chinese data center market, will be supported by the next version of the Linux kernel.
Going back a number of months I have been writing about the Linux kernel patches for Hygon Dhyana that effectively amount to copying the AMD Zen code-paths within the Linux kernel to use them as well for this x86 Chinese server SoC based on EPYC. After several rounds of patch revisions, that code is into shape and is now making its maiden voyage to the Linux kernel.
The Dhyana x86 CPU support was sent in today as part of the x86/cpu changes for the just-opened Linux 4.20~5.0 kernel merge window. It's just a few hundred lines of code of introducing the new Hygon/Dhyana identifiers and then adding them to the Zen code-paths ranging from the power management code to PCI to KVM/Xen and other areas of the Linux kernel having Zen-specific code/support. There are no other changes, showing just how close the Dhyana is to existing AMD Zen1 processors.
So with Linux 4.20~5.0 the kernel should be working on these Chinese servers or will be found out-of-the-box by the time of Ubuntu 19.04, Fedora 30 (or F29 updates), etc.
| 9
| 1,760,719,607.664177
|
https://www.phoronix.com/news/AMD-EPYC-Linux-4.19-Boosts
|
AMD EPYC Sees Some Performance Improvements With Linux 4.19
|
Michael Larabel
|
I am still finishing up work on my Linux 4.19 kernel stable benchmarks given it's been (and continues to be) a very busy month for Linux hardware testing, but of interest so far has been seeing a few EPYC performance improvements in some of the real-world workloads.
While a featured article looking at the Linux 4.19 kernel performance is on the way from a diverse selection of hardware, below are some benchmarks from the new Dell PowerEdge EPYC 2P server we began testing a few weeks ago. It was exciting to see that there are some performance improvements with the freshly minted Linux 4.19 stable kernel on top of the already very competitive (and in some instances jaw-dropping) performance.
Tests were done between Linux 4.17, 4.18, and 4.19 stable using the Ubuntu Mainline Kernel PPA.
The Parboil OpenMP performance was running slightly faster on the 4.19 kernel.
The Blender 3D modeling performance also sees a small improvement with this newest kernel.
In some cases memcached was also doing better, but on this newer kernel there was higher variance in those figures.
Aside from select multi-threaded workloads seeing small gains with Linux 4.19, most workloads though the Linux 4.19 performance was unchanged for this 64-core / 128-thread AMD Linux server. More benchmarks forthcoming while for now you can see our Linux 4.19 feature overview to learn more about the changes in this kernel that was released as stable on Sunday..
| 0
| 1,760,719,608.113692
|
https://www.phoronix.com/news/CodeXL-2.6-Released
|
AMD CodeXL 2.6 Advances GPU Profiling, Static Analysis & GPU Debugging
|
Michael Larabel
|
AMD's GPUOpen group today released CodeXL 2.6 as the newest version of their GPU developer suite.
The CodeXL cross-platform open-source developer tool is a bit different with CodeXL 2.6 as they have punted some functionality off to different utilities. They have dropped CPU/Power profiling and instead recommend developers use the AMD uProf/microprof tool, graphics frame analysis for D3D12/Vulkan can instead use RenderDoc or Radeon GPU Profiler, etc.
But what is found within CodeXL 2.6 for GPU developers are the GPU profiling features, static analysis features, and GPU debugging features.
More details on GPUOpen's CodeXL 2.6 for Windows/Linux game/application developers can learn more via today's announcement.
| 1
| 1,760,719,609.150885
|
https://www.phoronix.com/news/Threadripper-2920X-2970WX-Date
|
AMD Ryzen Threadripper 2970WX & 2920X Will Ship At The End Of October
|
Michael Larabel
|
We knew AMD was planning to release the rest of the Threadripper 2 line-up in October and now we finally know the precise date.
AMD has announced the Threadripper 2920X and 2970WX processors will begin shipping on 29 October to complement the Threadripper 2950X and Threadripper 2990WX.
The Threadripper 2920X is a 12-core / 24-thread part with 3.5GHz base frequency and 4.3GHz turbo while having a 32MB L3 cache like the 2950X. Also like the 2950X, there are 64 PCI Express 3.0 lanes and a 180 Watt TDP. This 12-core / 24-thread processor will begin shipping on 29 October at $649 USD.
The Threadripper 2970WX is a step below the 32-core / 64-thread 2990WX and is a 24-core / 48-thread part while matching the 3.0 / 4.2GHz clock frequencies of the flagship product. The 250 Watt TDP, 64MB L3 cache, and 64 PCIe lanes also match that of the other parts. The AMD 2970WX will be available at $1299 USD.
While the software is limited to Windows, AMD also announced today a Dynamic Local Mode to help force high priority workloads to CPU cores with local memory access. There are already other ways of achieving this on Linux.
We should be receiving review samples of the Threadripper 2920X and 2970WX soon for Linux benchmarking.
| 8
| 1,760,719,610.755431
|
https://www.phoronix.com/news/AMD-Picasso-VCN-DPG
|
AMD Picasso APUs To Support VCN Dynamic Power Gating
|
Michael Larabel
|
Earlier this month AMD sent out the initial Linux graphics driver patches for "Picasso" APUs and now a new patch series today sheds some light on a new capability for these new APUs reported to be similar to current Raven Ridge hardware.
While the initial AMDGPU DRM driver Picasso code drop happened earlier this month -- and it's already been queued for Linux 4.20~5.0 along with initial Raven 2 support -- as is usually the case, over weeks/months that follow are more of the new feature work for the driver gets ironed out beyond the initial hardware enablement.
The first set of patches post-enablement for Picasso hardware were quietly sent out today with not much exposure... Just ending the patch series with a "Enable DPG mode on PCO." That PCO is short for Picasso while "DPG" in this context is for Dynamic Power Gating.
Dynamic power gating is about shutting off power to unused hardware blocks depending upon the current system activity. Previous generations of AMD APUs have supported dynamic power gating while for the Picasso APUs it's being extended through to the VCN block -- Video Core Next. VCN premiered with Raven Ridge for video encode/decode/transcode as a unified block over UVD and VCE on previous hardware.
In digging through the eight patches the DPG support for Picasso with VCN is sorted out. But the patch comments don't reveal any notable details -- it will certainly be interesting to see once Picasso hardware surfaces but should end up delivering better power-savings. Improving the power efficiency is likely one of the broader goals with Picasso given these yet-to-be-released APUs are also expected to be manufactured at 12nm compared to the current 14nm FinFET Raven Ridge processors.
| 1
| 1,760,719,610.831376
|
https://www.phoronix.com/news/Hygon-Dhyana-Linux-V8
|
The "Chinese EPYC" Hygon Dhyana CPU Support Still Getting Squared Away For Linux
|
Michael Larabel
|
Back in June is when the Linux kernel patches appeared for the Hygon Dhyana, the new x86 processors based on AMD Zen/EPYC technology licensed by Chengdu Haiguang IC Design Co for use in Chinese data-centers. While the patches have been out for months, they haven't reached the mainline kernel quite yet but that might change next cycle.
The Hygon Dyhana Linux kernel patches have gone through several revisions and the code is mostly adapting existing AMD Linux kernel code paths for Zen/EPYC to do the same on these new processors. While these initial Hygon CPUs appear to basically be re-branded EPYC CPUs, the identifiers are different as rather than AMD Family 17h, it's now Family 18h and the CPU Vendor ID is "HygonGenuine" and carries a new PCI Express device vendor ID, etc. So the different areas of the kernel from CPUFreq to KVM/Xen virtualization to Spectre V2 mitigations had to be updated for the correct behavior.
The AMD EPYC Linux support itself has been in great shape since these competitive server CPUs began shipping last year, as covered in many benchmarks on several different EPYC SKUs since that point.
Published on Sunday is now the v8 patches for this CPU support. These patches have been tested on "HygonDhyana SoC silicon" and doesn't affect the kernel support for existing AMD Linux customers. The new patch revision fixes a boot issue while v7 came out recently too with various other fixes and is re-based against the latest Linux 4.19 state.
The Hygon Dyhana Linux kernel support appears to be mostly settling down, so it wouldn't be surprising if this gets queued soon for merging into the Linux 4.20~5.0 kernel.
| 3
| 1,760,719,612.387005
|
https://www.phoronix.com/news/AMD-Platform-QoS-RFC-Patches
|
AMD Publishes Platform QoS Patches For Next-Gen Processors
|
Michael Larabel
|
This afternoon AMD sent out their first Linux kernel patches for what might end up being a new feature for the "EPYC 2" / Zen 2 processors.
Hitting the Linux kernel mailing list a few minutes ago was a set of experimental / request-for-comments patches on "AMD QoS support".
This series adds support for AMD64 architectural extensions for Platform Quality of Service. These extensions are intended to provide for the monitoring of the usage of certain system resources by one or more processors and for the separate allocation and enforcement of limits on the use of certain system resources by one or more processors.
The monitoring and enforcement are not necessarily applied across the entire system, but in general apply to a QOS domain which corresponds to some shared system resource.
The initial QoS functionality is for L3 cache allocation enforcement, L3 cache occupancy monitoring, L3 code-data prioritization, and memory bandwidth enforcement/allocation. This AMD QoS sounds like -- and implied as much by the Linux code -- akin to Intel RDT (Resource Director Technology).
The AMD Linux developer notes did say that this QoS functionality is for their "next generation of processors." That would be Zen 2 or more specifically likely reserved just for the "EPYC 2" server processors as I doubt we'll end up seeing this QoS functionality enabled for desktop Ryzen CPUs.
The initial patches for this support can be found on the kernel mailing list. In my close monitoring of the Linux kernel mailing list and my various automated detectors to note potentially interesting patches/commits, I believe this is the first of seeing any what is likely next-gen EPYC patches for the Linux kernel... It will certainly be interesting to see what else is on approach for the kernel in the near future as AMD gears up for a very interesting 2019.
| 11
| 1,760,719,612.418248
|
https://www.phoronix.com/news/AMD-Picasso-RadeonSI-Mesa
|
AMD Picasso Support Comes To The RadeonSI OpenGL Driver
|
Michael Larabel
|
Last week AMD sent out initial support for yet-to-be-released "Picasso" APUs with the Linux AMDGPU kernel graphics driver. Today on the user-space side the support was merged for the OpenGL RadeonSI Gallium3D driver.
Picasso details are still fairly light but they are expected to be similar to Raven Ridge and for the AM4 processor socket as well as an edition for notebooks. On the same day as publishing the Picasso AMDGPU kernel patches, AMD also went ahead and published the Linux patches for the "Raven 2" APUs too.
From the user-space/RadeonSI side, indeed it's very similar to current Raven Ridge APUs with their graphics processor... The Picasso support from the OpenGL driver side is just adding the new PCI ID (0x15D8) and identifies it as Raven for taking the same driver code paths.
That one-liner patch for RadeonSI is now in for Mesa 18.3. The kernel bits meanwhile are coming for Linux 4.20~5.0.
| 14
| 1,760,719,613.681076
|
https://www.phoronix.com/news/AMDKFD-Vega-20-Support
|
AMD Sends Out Initial Vega 20 Support For AMDKFD Compute Kernel Driver
|
Michael Larabel
|
While AMD has been sending out Linux enablement patches for the yet-to-be-released Vega 20 for months now, what didn't see any work until today was for the AMDKFD driver support so this expected 7nm Vega GPU can work with their ROCm/OpenCL compute stack.
Out today are six patches adding the Vega 20 GPU support to the "AMD Kernel Fusion Driver" and is a fairly basic addition at just over one hundred lines of code.
It's great to see the Vega 20 support squared away ahead of launch and hopefully with Linux 4.20~5.0 it will be in quite good shape. On the Mesa side it seems to be as well with Git master.
| 3
| 1,760,719,614.272244
|
https://www.phoronix.com/news/AMDVLK-12-September-2018
|
AMD Lands Mostly Fixes In Latest Batch Of AMDVLK/XGL/PAL Code Updates
|
Michael Larabel
|
The AMD developers maintaining their "AMDVLK" Vulkan driver have pushed out their latest batch of code comprising this driver including the PAL abstraction layer, XGL Vulkan bits, and LLPC LLVM-based compiler pipeline.
This week's worth of work mostly is just bug fixes but some low-level code improvements and restructuring.
- Dropping a workaround for Dota 2 now that Valve corrected the issue upstream.
- A graphics path for scaled copies in PAL.
- PAL now supports a direct display for console mode.
- The LLVM Pipeline Compiler (LLPC) now supports a PAL new meta-data format for Vulkan.
- Various fixes and optimizations to PAL, including some CTS fixes.
- Various fixes in XGL.
Instructions as always on building this driver from source can be found at GitHub.
| 5
| 1,760,719,615.137756
|
https://www.phoronix.com/news/AMD-GCN-GPU-2018-Cauldron
|
More Details On The AMD GCN Back-End For GCC That's Expected To Merge For GCC 9
|
Michael Larabel
|
Last week I reported on Code Sourcery / Mentor Graphics posting their new AMD GCN port to the GNU Compiler Collection (GCC). This GPU back-end for the widely-used GCC compiler is hoped for merging ahead of the GCC 9 stable release expected in early 2019. At this past weekend's GNU Tools Cauldron 2018 conference was a briefing by Mentor Graphics on undertaking funded by AMD.
Andrew Stubbs who posted those latest AMD GCN patches last week for GCC was in attendance at the Cauldron in Manchester (UK) to talk about their AMD GPU back-end work for the GNU compiler toolchain. Among the highlights are:
- AMD has indeed been funding Mentor Graphics to create this new GCC port... It was pretty much assumed, but not entirely clear with AMD and SUSE being long-time partners and SUSE having originally started working on the GCN back-end in 2016. But it turns out AMD ended up hiring Mentor to work on this code since 2017 while using the SUSE code as a starting point.
- One of the explicit requirements by AMD is for this GCN back-end to handle Fortran code (GFortran) with the OpenACC and OpenMP parallel programming interfaces.
- A binary release of this AMD GCN GCC port based on GCC 7 and with only supporting C++ OpenMP/OpenACC offloading is currently available within Sourcery CodeBench.
- Mentor will be making a new binary release in November based on GCC 8 and with various OpenACC updates, expanded vector support, and other backend-specific improvements.
- They don't yet have the AMD GCN code written for Binutils coverage so for now they are relying upon LLVM's assembler and linker support.
- They indeed are hoping to get this back-end merged for GCC 9.
- They still need to finish more C++ functionality, sub-word vector operations, register sharing to support more than four threads per compute unit, and various clean-ups / optimizations.
A lot more technical details on this AMD back-end for GCC are available via this slide deck (PDF) from the GNU Tools Cauldron 2018.
| 16
| 1,760,719,616.108126
|
https://www.phoronix.com/news/AMD-Ryzen-2300X-2500X
|
AMD Officially Announces The Ryzen 3 2300X & Ryzen 5 2500X
|
Michael Larabel
|
Following weeks of leaks about these new processors targeting OEMs and system integrators, AMD today officially announced the Ryzen 3 2300X and Ryzen 5 2500X processors.
The Ryzen 3 2300X and Ryzen 5 2500X will be marketed to OEMs and system builders as great processors for "off-the shelf gaming, family, or small form factor desktop PC." Both of these models have a maximum boost clock speed of 4.0GHz and four cores while the Ryzen 5 2500X also has SMT to yield eight threads in total. These new CPUs each have a 65 Watt TDP and a 10MB smart cache and PBO support.
These new Ryzen CPUs come just days after AMD announced new Athlon and Ryzen PRO processors as well. More details at AMD on the Ryzen 5 2500X and Ryzen 3 2300X.
| 15
| 1,760,719,616.6518
|
https://www.phoronix.com/news/Vulkan-Memory-Allocator-2.1
|
AMD's GPUOpen Vulkan Memory Allocator 2.1 Released
|
Michael Larabel
|
AMD's GPUOpen group has announced a new version of their open-source Vulkan Memory Allocator project that seeks to make it easier to deal with memory allocation and management when using this graphics API.
With Vulkan Memory Allocator 2.1 there is now support for linear allocations, improvements to its memory debugging system, support for traces and replays, changes to the JSON dumping format, and other enhancements.
More details on Vulkan Memory Allocator 2.1 can be found via the release announcement posted today to the GPUOpen blog. This cross-platform, open-source Vulkan memory allocation library is available to download from GitHub.
| 2
| 1,760,719,617.530366
|
https://www.phoronix.com/news/AMD-OpenCL-2.0-ROCm-2.0-Work
|
AMD Nearing Full OpenCL 2.0 Support With ROCm 2.0 Compute Stack
|
Michael Larabel
|
AMD's fully open-source GPU compute stack in the form of ROCm "Radeon Open eCosystem" is nearing its next milestone with OpenCL 2.0 compliance.
While we have been looking forward to ROCm 1.9 as the next feature release for this Linux OpenCL/GPGPU stack with OpenCL 1.2 officially plus portions of OpenCL 2.0+, the ROCm 2.0 release is where they are squaring up for complete OpenCL 2.0 support.
AMD's Gregory Stoner commented this weekend on the ROCm roadmap when questioned about it, "We are looking to have full OpenCL 2.0 support with ROCm 2.0. Note All the 2.0 features have been in for quite a while, except for kernel enqueue and Pipes are the last two feature we need to role out to formally turn on OpenCL 2.0. I will have the team work on a feature matrix for AMDGPUpro vs ROCm. You find we even have some of the OpenCL 2.1 runtime features in ROCm OpenCL driver now. Right now we are focused on any new runtime and Kernel feature."
This is really great to hear. ROCm is coming together nicely now with OpenCL 2.0 on the horizon and the kernel-side bits in AMDKFD recently having gone mainline -- though there still does seem to be some "gotchas" on the kernel side with at least the released/packaged ROCm bits hitting problems I tried recently on Linux 4.18 mainline, but we'll see how well it works on Linux 4.19... Plus moving forward, the AMDKFD driver being folded into AMDGPU.
The Radeon Open eCosystem stack is becoming quite compelling especially with the highly anticipated 7nm "Vega 20" expected later this year and that looking like it could hold a fair amount of potential for deep learning / workstations. With xGMI support coming and that playing into HSA, it looks like AMD could be on a strong footing for GPU compute in 2019. The one final element really left to be improved upon is for better CUDA code portability with their HIP compiler or getting more frameworks running natively on their driver stack after years of these frameworks (and GPU compute experts/researchers) being almost entirely focused on NVIDIA/CUDA.Come 2019, OpenCL-Next should be released as more work for the AMD/GPUOpen crew.The only OpenCL driver currently in better shape than ROCm for Linux users is the Intel OpenCL NEO driver that exposes OpenCL 2.1+ for its latest UHD/HD/Iris Graphics. Their previous "Beignet" driver also reached OpenCL 2.0 support before being replaced by the NEO OpenCL stack. On the NVIDIA Linux driver front, they are officially at OpenCL 1.2 with some CL 2.0 extensions but in not supporting SVM, but they are waiting for OpenCL-Next to suit their needs.
| 24
| 1,760,719,618.397564
|
https://www.phoronix.com/news/RadeonSI-Zen-Tuning-Merged
|
The RadeonSI Performance Tuning For AMD Zen CPUs Has Landed In Mesa
|
Michael Larabel
|
Earlier this week I reported on the RadeonSI Gallium3D code being tuned for AMD Zen CPUs in an attempt to deliver greater gaming performance for Ryzen processors. That work has now been merged into Mesa 18.3.
As explained when the patches surfaced for tuning RadeonSI/Gallium3D for Ryzen CPUs, the work by AMD's Marek Olsak is due to the CCX (core complexes) design of Zen, it will try to pin the application thread and driver execution threads to the same CCX so that they can share the same L3 cache. With these patches, Marek found in some OpenGL benchmarks that there was an increase by 25~32% in some micro-benchmarks, but real-world gaming performance has yet to be analyzed -- I'll be running some benchmarks shortly.
Over the course of several patches, the work was merged into Mesa 18.3 Git on Friday.
Marek confirmed in our forums that this tuning was needed to RadeonSI's design. "There is no per-driver work needed if a driver doesn't have any threads internally. Most of the logic is in the shared code and is applied to the GL and gallium threads. RadeonSI is an exception, because it has 1 internal thread."
Stay tuned for some benchmarks as soon as time allows, it should likely provide at least a small benefit for Radeon + Ryzen setups.
| 16
| 1,760,719,619.026697
|
https://www.phoronix.com/news/AMD-Athlon-200GE-New-PRO
|
AMD Announces The Athlon 200GE With Vega 3 Graphics, 2nd Gen Ryzen/Athlon PRO
|
Michael Larabel
|
AMD has a few processor announcements this morning that are no longer under embargo.
First up is the AMD Athlon 200GE SoC. The Athlon 200GE offers two-cores / four-threads (SMT) at 3.2GHz with Vega 3 onboard graphics. This SoC has a 35 Watt TDP and will be available at $55 USD. The Athlon 200GE is intended for basic desktop use-cases.
AMD is also announcing their second-generation Athlon PRO and Ryzen PRO processors. These Zen+ CPUs range from the Athlon PRO 200GE to the Ryzen 7 PRO 2700X. These PRO CPUs are intended to offer greater reliability and security for enterprise-focused workloads.
More details on these new AMD PRO processors via today's announcement.
| 43
| 1,760,719,619.928111
|
https://www.phoronix.com/news/AMD-RadeonSI-Ryzen-Tuned
|
AMD's RadeonSI/Gallium3D Linux Graphics Driver Gets Optimized For Ryzen CPUs
|
Michael Larabel
|
It's arguably a bit late, but patches are now pending for optimizing the RadeonSI Gallium3D open-source Linux graphics driver for the AMD Ryzen CPU microarchitecture.
What this set of patches do to the Mesa Gallium3D and RadeonSI driver code is optimize it for the AMD Zen architecture with its multiple core complexes (CCX). The new code allows for the pinning of the application thread and driver execution threads to the same CCX where they are able to share the same L3 cache.
By ensuring all of the driver threads for a given application/game are pinned to the same L3 cache / CCX as the application, there can be a significant performance benefit. In OpenGL micro-benchmarks there was an increase in performance by 32% for a draw elements operation or 25% for DrawArrays when using an AMD Ryzen Zen 1 CPU.
For real-world games/benchmarks, the performance benefit is expected to be much less than those micro-benchmarks but still should be noticeable. I'll be firing up some benchmarks in the next few days with these Mesa patches.
For now the patches are under review but will hopefully be mainlined for Mesa 18.3-dev in short order. Stay tuned for the benchmarks on these RadeonSI+Ryzen optimizations.
| 19
| 1,760,719,620.536832
|
https://www.phoronix.com/news/AMD-Excavator-More-CPU-Temps
|
Linux Will Finally Report Temperatures For Certain AMD Excavator CPUs
|
Michael Larabel
|
Besides the Linux "k10temp" AMD CPU temperature reporting driver recently seeing support for Threadripper 2 temperature monitoring, much older Excavator (Bulldozer 4th Gen) processors will now see working CPU temperature reporting for select models.
While AMD Excavator CPUs came out in 2015, particularly in the APU form like Carrizo, not all of these models have had working CPU temperature reporting even with the latest Linux kernel code. Excavator CPUs in the Family 15h (Bulldozer) have a CPUID model between 60h and 6Fh and newer revisions between 70h and 7Fh. But the k10temp Linux driver up to now has just been checking for 60h and 70h, not any of the other model numbers.
While 60h/70h seems to cover most of the Excavator parts, there have been some APUs like the A10-9620P that have a Model 65h number and thus haven't been supported by k10temp due to failing this simple check.
But now there's a patch pending to correct the behavior of the k10temp driver so it properly checks for the entire range of Excavator models. Hopefully this will work its way into Linux 4.19 and for back-porting as a "fix" for those still relying upon this last generation of Bulldozer processors.
It seems temperature reporting under Linux hasn't been the highest priority for AMD with the missing Threadripper support for recent CPUs but also this Excavator incident and just with Linux 4.18 came temperature support for Stoney and Bristol Ridge, more older AMD APUs. These patches as well haven't been coming from AMD but rather Guenter Roeck and other hwmon subsystem developers.
| 6
| 1,760,719,621.36263
|
https://www.phoronix.com/news/AMD-ROCm-1.8.3-Released
|
AMD ROCm 1.8.3 Released To Fix Breakage With Latest Ubuntu 18.04 Kernel
|
Michael Larabel
|
While still waiting on the ROCm 1.9 release to happen, version 1.8.3 of the Radeon Open eCosystem stack was released for Linux systems.
This latest point release to ROCm 1.8 comes just to fix a build regression against the latest Ubuntu 18.04 kernel update.
ROCm 1.8.2 brought early support for Ubuntu 18.04 LTS but the ROCm kernel bits ended up breaking with Ubuntu 18.04's 4.15.0-33-generic update. ROCm 1.8.3 has been released to fix that breakage against the latest point release in Ubuntu 18.04 Bionic Beaver.
No other changes in ROCm 1.8.3 have been noted by AMD. Instructions on setting up this open-source OpenCL/compute stack for newer Radeon graphics cards can be found via ROCm on GitHub.
| 12
| 1,760,719,621.987443
|
https://www.phoronix.com/news/Linux-4.18.6-k10temp-Correct
|
Linux 4.18.6 Kernel To Properly Report AMD Threadripper 2 CPU Temperature
|
Michael Larabel
|
The soon-to-be-released Linux 4.18.6 stable kernel will correctly report the CPU core temperatures of the new AMD Threadripper 2950X and 2990WX processors.
With the new high-core-count AMD processors that launched earlier this month, the 16-core / 32-thread Threadripper 2950X and 32-core / 64-thread Threadripper 2990WX, the only real Linux shortcoming to report had been the lack of correct temperature reporting on the stock Linux kernel at the time... With the kernels up to this point, the reported CPU core temperature on these Threadripper 2 CPUs has been +27 degrees (Celsius) higher than it should be due to a missing Tctl offset.
The 27 degree difference is potentially nerve-wracking considering the 2950X has a 180 Watt TDP while the 2990WX has a 250 Watt TDP... If you are air cooling, chances are you'd like to accurately spot check the CPU temperature under load. Fortunately, I spotted the missing Tctl offset in time and that patch was merged into Linux 4.19.
Now that Linux 4.19-rc1 is out the door as the first development release of Linux 4.19, Greg Kroah-Hartman has pulled this stable-marked patch for back-porting. As of a short time ago, it's in Greg's 4.18 stable queue. Those patches will soon be released as Linux 4.18.6 and thereby offering correct Threadripper 2950X/2990WX temperature reporting.
The k10temp patch isn't being back-ported to earlier LTS kernels since it was only around Linux 4.15 when the Zen temperature reporting code was added. If you want to bring it back to Linux 4.16~4.17 kernels, the patch should apply cleanly to older kernels. Aside from the temperature reporting caveat, the Threadripper 2 experience on Linux should be smooth-sailing and in fact much better performance-wise than under Windows as already shown in the many Threadripper 2 benchmarks.
| 14
| 1,760,719,622.935123
|
https://www.phoronix.com/news/AMD-VCN-PSP-Firmware
|
More AMDGPU Work For Linux 4.19 Has VCN + PSP Firmware Hookup For Future Hardware
|
Michael Larabel
|
The good news is that the open-source AMD graphics team continues working on support for upcoming hardware, but the bad news is that it looks like their VCN video hardware might be a bit more locked down than it is now.
With current Raven Ridge APUs there is VCN as "Video Core Next" as a replacement to UVD and VCE for video decoding and encoding, respectively. This dedicated hardware core for video encode/decode has been supported well now for some months on the open-source Linux graphics driver stack. The latest patches hitting the mailing list for hopeful integration to Linux 4.19 are a bit interesting and reveal a change for future hardware.
With Thursday's amdgpu drm-next-4.19 has just a few changes but the main work is adding support for VCN into the PSP driver and allowing the VCN booting with firmware that is loaded by the PSP. It's said that "this is required on upcoming parts."
PSP in this context is the controversial Platform Security Processor. Basically the PSP is going to be responsible for setting up the firmware for the VCN video encode/decode hardware using a "Trust Memory Region" mac address.
There's nothing immediate that should cause concern, but the change may be of interest to those free software individuals that don't like hearing "PSP" and "firmware" talked about often by open-source drivers. But let's hope these future AMD parts won't be faced by any other restrictions. From my personal perspective, I'm just glad to see more confirmation of their continual and punctual work on supporting upcoming/future hardware.
Besides this firmware loading change, this pull request also contains a few display fixes for Linux 4.19.
| 18
| 1,760,719,623.387128
|
https://www.phoronix.com/news/Threadripper-2900-Benchmarks-Co
|
The Linux Benchmarking Continues On The Threadripper 2950X & 2990WX
|
Michael Larabel
|
While I haven't posted any new Threadripper 2950X/2990WX benchmarks since the embargo expired on Monday with the Threadripper 2 Linux review and some Windows 10 vs. Linux benchmarks, tests have continued under Linux -- as well as FreeBSD.
I should have my initial BSD vs. Linux findings on Threadripper 2 out later today. There were about 24 hours worth of FreeBSD-based 2990WX tests going well albeit DragonFlyBSD currently bites the gun with my Threadripper 2 test platforms. More on that in the upcoming article as the rest of those tests finish. It's also been a madhouse with simultaneously benchmarking the new Level 1 Terminal Fault (L1TF) vulnerability and the performance impact of those Linux mitigations on Intel hardware will start to be published in the next few hours.
Anyhow, for those that have been requesting particular benchmarks, I have been satisfying many reader requests directly on the Phoronix Test Suite's OpenBenchmarking.org.
If searching for the Threadripper 2950X or Threadripper 2990WX are various result files with different benchmarks ranging from more code compilation tests to CPU crypto currency mining performance to other reader requests. Check it out while waiting for the next featured articles on Phoronix. Phoronix Premium readers especially, continue sending in your Threadripper 2 benchmark requests.
| 5
| 1,760,719,624.493247
|
https://www.phoronix.com/news/Threadripper-2900-Hwmon-PR
|
Threadripper 2900 Series Temperature Monitoring Sent In For Linux 4.19 Then Backported
|
Michael Larabel
|
As expected, the CPU temperature monitoring support within the "k10temp" hwmon driver has seen the patches sent in today to be updated for the AMD Threadripper 2900 series CPU support. These patches are going into the Linux 4.19 kernel merge window but slated to be back-ported to the currently supported stable kernel series.
The k10temp driver will report a CPU core temperature 27 degrees higher than it actually is, on unpatched kernels, for the new Threadripper 2 hardware. As covered, that's expected to be the lone compatibility issue with the new Threadripper processors on Linux. So just be aware of this temperature offset if soon buying a Threadripper 2990WX or 2950X given the high wattage of these CPUs as to not be startled if initially seeing a CPU temperature that is higher than expected.
These patches will work not only for the Threadripper 2950X and 2990WX but also with the 2920X and 2970WX launching later in the year with the same Tctl offset. The patches were sent in overnight as hwmon fixes for Linux 4.19 but as stated are copied to the Linux stable list so they should be back-ported for 4.18 point releases and earlier in the very near future.
Stay tuned for our Threadripper 2950X and 2990WX Linux benchmarks coming up very shortly with that embargo soon expiring.
| 1
| 1,760,719,624.890853
|
https://www.phoronix.com/news/Linux-EDAC-Great-Horned-Owl
|
Linux EDAC Support For AMD's Great Horned Owl
|
Michael Larabel
|
The latest Linux kernel patch is for supporting ECC error detection via the Error Detection And Correction (EDAC) code with AMD's Great Horned Owl.
Great Horned Owl is the codename for Family 17h Model 11h hardware, which is what launched back in February as the Ryzen Embedded V1000 series chips.
This appears to be the first time I'm seeing any Model 11h / "Great Horned Owl" Linux kernel patches and no hits in Git. Granted, the V1000 series is based upon the Zen micro-architecture and should leverage pretty much all existing AMD code-paths except for where any specific ID entries are needed.
Likewise, in the case of this EDAC addition, it's just a matter of adding the IDs.
I don't have any Ryzen Embedded V1000 hardware nor have I heard of any reports from Phoronix readers trying to run this embedded hardware on Linux, but presumably it should be in good shape given the Zen CPU cores, etc, sans any quirks like this EDAC driver support. If you happen to have any other information to share on the Great Horned Owl Linux support, feel free to ping us.
| 18
| 1,760,719,626.083659
|
https://www.phoronix.com/news/Threadripper-2-Kernel-Bulletin
|
Linux Kernel Expectations For AMD Threadripper 2
|
Michael Larabel
|
If you have already pre-ordered your AMD Threadripper 2990WX processor or just planning to be an early customer of that high-end desktop processor or the Threadripper 2950X, you may be wondering about Linux requirements from these new high-end AMD CPU offerings. Here's the gist of the Linux support state of AMD Zen+ CPUs for those wanting to get ready for Threadripper 2.
There's still a few days to go until the global embargo expires for sharing Threadripper 2990WX/2950X performance benchmarks and reviews, at which time you'll be able to see my full Linux analysis of these new processors with plenty of tests... But for today we're just talking in general terms about Linux support requirements.
Fortunately, there isn't much in terms of heightened Linux requirements compared to some past CPU launches or particularly recent APU product debuts... With the Threadripper 2 processors being based on Zen+ like the Ryzen 7 2700 series from earlier this year and no new chipset requirements (X399 still working great), the Linux support is largely as expected and should work with most any Linux distribution over the past year.
Of course, as is generally the case with newer hardware from major vendors, the newer the kernel and other components will generally lead to a better experience. But basically something with a recent kernel (let's say Linux 4.15+ given it's found in Ubuntu 18.04 and has much of the AMD Zen optimizations we've seen to date) and so beyond Ubuntu, Fedora Workstation 28, openSUSE Tumbleweed, Arch Linux, and others should all be fair game.
Another general important factor for Zen/Zen+ hardware if you often finding yourself compiling your own software would be a newer compiler release. Both LLVM Clang 6.0 and GCC 8.1 (now 8.2) carry various Zen optimizations if compiling using the znver1/native target. So having GCC8 or Clang 6 (soon to be Clang 7) should be beneficial for squeezing the maximum performance out of your generated code for AMD Ryzen/Threadripper CPUs.
Once the embargo passes, I'll certainly have more information and benchmarks on several of the leading Linux distributions to see how the performance compares. Long story short though, I would expect AMD Threadripper 2 to play nicely with most currently released Linux distributions. But there is one exception: like was an issue with Ryzen and Threadripper1, the thermal reporting in the current Linux code isn't for the time being taking into account a Tctl offset.
It's not that you won't find the CPU temperature reporting working for new Threadripper CPUs, it's just that it will be incorrect: the reported temperature will be 27 degrees higher than the actual temperature. As of the current Linux 4.18 Git code, that offset isn't in place for the Threadripper2 CPUs. So keep this in mind before checking on the temperature of your 180 Watt (or 250 Watt, in the case of the 2990WX) CPU and become concerned that it may seem like it's operating very warm... Wait for either a patched kernel or just be sure to subtract 27 degrees. The patch fixing the offset for the 2950X and 2990WX is currently queued in hwmon-next for landing in next week's Linux 4.19 kernel merge window. From there the patch is already tagged as a candidate for back-porting to existing and currently supported stable kernel versions... So within a few weeks, point releases to Linux 4.18 and friends should bring the correct temperature reporting. Just making this point very clear and now given the high power requirements of these CPUs as to not get the wrong impression over your cooling setup for the time being. (Obviously Windows users shouldn't have to worry about this with the AMD software and drivers being distributed independent of the Windows kernel.)
That's about it. As far as X399 motherboards, they continue working out generally well under Linux. About the only caveat with those motherboards is along a similar line and that is most current Intel/AMD motherboards not playing well with the sensor thermal/voltage/fan-speed monitoring under Linux, now even more so due to the out-of-tree it87 driver going unmaintained. But in terms of all core functionality from X399 motherboards, they continue playing well on Linux with no major worries over Linux hardware compatibility.
That's it for now and stay tuned for the AMD Threadripper 2 performance numbers under Linux once that embargo passes.
| 11
| 1,760,719,626.403554
|
https://www.phoronix.com/news/AMD-Zen-PMU-Events-Linux
|
Linux Kernel Gets Patch To Support AMD Zen's Performance Monitoring Unit Events
|
Michael Larabel
|
SUSE developer Martin Liška has published a patch wiring in support for AMD PMU events on the AMD Family 17h "Zen" processors.
This is for recognizing the events generated by the CPU's performance monitoring unit (PMU) within the Linux's "perf" performance monitoring subsystem.
Those PMU events pertain to the Ryzen/EPYC processors cache/core/FP/memory and other characteristics for tracking things like cache misses, retired instructions, reload requests, etc. This PMU perf support is based upon documentation AMD had published last year when Zen CPUs began shipping.
The patch for now can be found on the mailing list but should soon be on its way to a kernel near you for those of you wanting to do perf-based profiling on Zen hardware.
| 0
| 1,760,719,627.614253
|
https://www.phoronix.com/news/Windows-Linux-AMD-Power
|
A Look At The Linux vs. Windows Power Use For A Ryzen 7 + Radeon RX Vega Desktop
|
Michael Larabel
|
Recently I have been posting a number of Linux laptop battery benchmarks including how the power consumption compares to Windows 10. If you are curious how these numbers play out on the desktop side and when using AMD hardware, here are some results for your viewing pleasure with a Ryzen 7 2700X and Radeon RX Vega 64 desktop system.
While working on some recent Windows/Linux benchmarks from the AMD side given the number of recent Intel operating system benchmarks, I took the opportunity to also run some fresh power consumption tests. The system under test was an AMD Ryzen 7 2700X with an ASUS ROG CROSHAIR VII HERO motherboard, 2 x 8GB DDR4-3400 memory, 256GB Samsung 950 PRO 256GB NVMe SSD, and Radeon RX Vega 64 graphics.
The operating systems tested out-of-the-box on this system were Microsoft Windows 10 Pro, Fedora Workstation 28, Antergos 18.7-Rolling, Clear Linux 23830, and Ubuntu 18.04 LTS.
The power consumption was being monitored via a WattsUp Pro connected to a different PC for being able to analyze the AC system power consumption from a neutral environment to rule out any polling differences. The Phoronix Test Suite was recording the WattsUp Pro AC system power draw for this Ryzen + Vega system.
First up was a look at the power use over the course of a cold boot to the logged in system, launching the default web browser and navigating to Phoronix.com, opening up the file manager, opening up the terminal/PowerShell, and other light desktop tasks.
To much surprise, Windows 10 had the highest power use on average at 88.4 Watts for this Ryzen 7 + RX Vega desktop while having the lowest average power use was actually Intel's Clear Linux platform at 79 Watts. Coming in at 81~82 Watts were Fedora, Antergos, and Ubuntu. When idling, Windows 10 appeared to be consuming several more Watts than the Linux operating systems under test. But under the desktop use, the Linux power use was peaking higher. The maximum Windows 10 power draw during boot and the start of these light desktop tasks was 188 Watts while Fedora Workstation came in at 191 Watts, Clear Linux at 187 Watts, and Ubuntu 18.04 LTS having the highest power draw at 205 Watts.
Next up was a look at the system power consumption when under load. The load was the Phoronix Test Suite running OpenArena, Crafty chess benchmark, x264 video encoding, Stockfish, C-Ray ray-tracing, and OSBench.
Most interesting from the Windows vs. Linux metrics was Windows 10 having a significantly lower peak power use at 198 Watts compared to 214~226 Watts on the four Linux distributions tested. But interestingly Windows had a slightly higher minimum power draw 79 Watts versus 69~72 Watts. The average power draw during this load benchmarking was 154 Watts. Meanwhile the lowest average power use of the Linux distributions was Fedora Workstation 28 at 150 Watts, which recently went through a round of power optimizations. Consuming the most power on average were Ubuntu 18.04 LTS and Antergos 18.7-Rolling.
Overall, the Linux desktop power use for this AMD Ryzen 7 2700X + Radeon RX Vega 64 desktop was largely on-par with Windows 10. Fedora Workstation 28 and Clear Linux tended to be the most energy efficient along with Antergos 18.7 while Ubuntu 18.04 LTS did tend to be a bit more power hungry -- similar to what we've found in the Intel/laptop benchmarks recently.
| 13
| 1,760,719,627.904702
|
https://www.phoronix.com/news/AMD-AOCC-1.2.1-Released
|
AMD AOCC 1.2.1 Compiler Flings Flang Fixes
|
Michael Larabel
|
AMD released a minor update to their AMD Optimizing C/C++ Compiler.
AOCC is the company's downstream of LLVM/Clang with optimizations for their Zen CPU microarchitecture with compiler optimizations/improvements before they work their way into upstream LLVM. AOCC is the replacement for AMD's Open64 compiler used years ago with earlier micro-architectures.
AOCC has been able to offer some performance advantages for Ryzen/EPYC CPUs and this month they quietly pushed out the AOCC 1.2.1 update. The only changes mentioned for AOCC 1.2.1 are fixes for their Flang compiler to enable better alpha testing. Flang is the LLVM-based Fortran compiler that has been in development but is still considered about alpha quality.
AOCC 1.2.1 with these latest Flang Fortran compiler improvements can be found via developer.amd.com.
| 4
| 1,760,719,629.473627
|
https://www.phoronix.com/news/AMDGPU-Has-More-For-Linux-4.19
|
AMDGPU Gets More Features For Linux 4.19 Kernel
|
Michael Larabel
|
On top of AMDGPU improvements/features already staged for Linux 4.19, the AMD folks on Thursday sent in their seemingly last set of feature updates to DRM-Next ahead of the Linux 4.19 kernel merge window.
There is certainly a lot of new DRM material queuing for Linux 4.19: if you are behind on your Phoronix reading, there will be a DRM recap next week or so on Phoronix with the cutoff for new DRM-Next material hitting its end for the upcoming 4.19 window. Thursday's Radeon/AMDGPU update just adds to this big list of changes.
This latest batch of work for AMDGPU in Linux 4.19 includes fixes for Raven Ridge GFXOFF support to be able to turn off the graphics engine when not needed for these newer APUs, more DC display code improvements, CRC support for DCN "Display Core Next" as currently present for Raven Ridge, a rework of the scheduler code in preparation for new load balancing code coming in Linux 5.0+, TTM memory management clean-ups, mapping processes to VM IDs for debugging GPUVM faults, and a range of other code clean-ups and improvements/fixes.
The complete list of AMDGPU changes for this latest DRM-Next pull request ahead of Linux 4.19 can be found on amd-gfx.
| 21
| 1,760,719,629.483362
|
https://www.phoronix.com/news/Linux-4.18-AMD-SSBD-Spectre-V4
|
Linux 4.18 Gets AMD Patches For Speculative Store Bypass / Spectre V4
|
Michael Larabel
|
Linux 4.17 landed the initial Spectre V4 mitigation as "Speculative Store Bypass Disable" (SSBD) while primarily focused on Intel CPUs and for Linux 4.18 the SSBD code has been updated for AMD processors.
The in-development Linux 4.18 kernel will receive the patches for making use of the SPEC_CTRL / VIRT_SPEC MSRs to be provided by future AMD CPUs / firmware updates. The AMD SSBD work done by Konrad Rzeszutek Wilk of Oracle was previously covered here. The AMD SSBD whitepaper can be read here.
The AMD speculative store bypass patches are the only Spectre/Meltdown patches in the x86 space currently queued up for Linux 4.18 via the x86/pti pull request. Granted, we could still see more Spectre/Meltdown improvements land for Linux 4.18 given that this work generally gets accepted outside of the merge window since it's a security issue.
Outside of the x86 space for Spectre there is also Spectre V4 SSBD for ARM64 and finally 32-bit ARM Spectre V1/V2 patches coming with Linux 4.18.
| 5
| 1,760,719,630.991955
|
https://www.phoronix.com/news/FreeBSD-Ryzen-Errata-Kernel
|
FreeBSD Kernel Patch Posted For Addressing Ryzen Errata
|
Michael Larabel
|
A few days back I wrote about workarounds for getting FreeBSD running stable on AMD Ryzen via a script to adjust some of the CPU's MSRs based upon a recently-updated AMD revision guide. That script, which was making use of FreeBSD's cpucontrol utility for adjusting the bits, has now morphed into a kernel patch.
Konstantin Belousov who worked on the script based upon the official AMD Ryzen errata guide has now turned it into a kernel patch that will hopefully be accepted upstream in the not too distant future.
For now the kernel patch just adjusts the MSR registers for Ryzen 1, but some of the workarounds are necessary for Ryzen 2, but have yet to be tested and thus not addressed by this current patch.
If you want to patch your FreeBSD kernel while waiting for the work to be merged and not having to worry about the user-space script, the kernel patch can be found via this mailing list post.
| 9
| 1,760,719,631.028698
|
https://www.phoronix.com/news/Hygon-Dhyana-AMD-China-CPUs
|
Hygon Dhyana: Chinese x86 Server CPUs Based On AMD Zen
|
Michael Larabel
|
While there are the VIA/Centaur-based Zhaoxin desktop CPUs targeted for the Chinese market, it turns out there is another x86 Chinese CPU effort but this time is a collaboration with AMD.
Chengdu Haiguang IC Design Co is a new x86 CPU vendor that is a joint venture between AMD and Haiguang Information Technology Co. This joint venture is aiming to make x86 CPUs for the Chinese server market. This deal was announced back in May and paid out $293 million to AMD for using their intellectual property.
The first Hygon CPU offerings is the "Dhyana" family that is based upon AMD technology and is derived from the AMD Family 17h "Zen" micro-architecture.
We know this as the new company today sent out Linux kernel patches for adding in the new x86 CPU vendor ID and for allowing the Dhyana Family 18h processors to use the AMD Family 17h code paths within the kernel. The Dhyana appears to be a SoC-based solution and not socketed.
With these patches, the Hygon Dhyana CPUs should now work under Linux... The patches are quite straight-forward and amount to less than 200 lines of new kernel code, so indeed at this point seems to be mostly a re-branded Zen CPU for the Chinese server market.We'll pass along any other details as we work them out.
| 44
| 1,760,719,632.798441
|
https://www.phoronix.com/news/ATI-RAGE-128-DRM-In-2018
|
A Revived Linux Driver To Be Attempted For The ATI RAGE 128
|
Michael Larabel
|
The ATI Rage 128 series was introduced in 1998 while now twenty years later a renewed DDX driver and potentially DRM/KMS kernel driver is going to be attempted for these AGP/PCI graphics cards from the days of OpenGL 1.2.
The ATI RAGE 128 family was great back in 1998 when it was going up against the Voodoo 3 and RIVA TNT while now for those Linux users who are fans of vintage hardware, there's going to be a renewed driver push. That's hardware back from the days when GPUs could have less than 10 million transistors, were manufactured on a 250nm process, and just over 100MHz clock speeds with SDR memory.
Kevin Brace who dived into open-source Linux graphics drivers two years ago while being the sole remaining developer working on the OpenChrome DRM/KMS driver has decided to take up work on the RAGE 128 graphics stack.
Recently he talked about taking a break from VIA graphics driver work to focusing upon a reusable DRM driver that could be used by other under-served/vintage/old graphics hardware. Today he announced his plans to work on the ATI RAGE 128 driver support.
At this stage he has just begun looking at the DDX user-space X.Org driver for "R128" and posted some basic patches so far. Details on his initial thoughts about RAGE 128 development via this blog post.
He also followed up with another post highlighting the existing issues with RAGE 128 open-source driver support in 2018.
| 49
| 1,760,719,633.661991
|
https://www.phoronix.com/news/AMD-Computex-2018
|
AMD Showed Off New Threadrippers, 7nm Vega At Computex 2018
|
Michael Larabel
|
Overnight was the AMD press conference at Computex 2018. Here are the highlights.
The key items to learn from the AMD Computex 2018 press conference included:
- AMD's new 12nm Zen+ Threadripper processors are on track to launch in the third quarter of this year. These new Threadrippers will come into a configuration up to 32 cores / 64 threads, remain compatible with existing X399 motherboards once receiving a firmware update, and will carry other updated features in line with Pinnacle Ridge.
- AMD also announced the rumored 7nm Radeon RX Vega Instinct deep learning product. This 7nm Vega GPU will be paired with 32GB of HBM2 memory. The accelerator was shown off during the event and it's now sampling, but no timeframe for launch was provided. This 7nm Vega GPU is just being intended for server and workstation compute.
- The Radeon RX Vega 56 Nano was announced as the Vega 56 on a very small PCB akin to the R9 Nano Fiji days. The Vega 56 Nano will begin shipping soon.
- AMD will introduce 7nm EPYC processors first followed by consumer parts next year.
Embedded below is their press conference stream.
| 46
| 1,760,719,634.157331
|
https://www.phoronix.com/news/Linux-4.18-HWMON-Updates
|
Linux 4.18 To Report CPU Temps Finally On Stoney & Bristol Ridge
|
Michael Larabel
|
The hardware monitoring "hwmon" updates have been sent in for the just-opened Linux 4.18 kernel merge window while what's interesting this time around are the k10temp driver updates for AMD CPU temperature reporting.
With Linux 4.18, the k10temp kernel driver is now able to report temperatures on Stoney Ridge and Bristol Ridge processors. Stoney Ridge are the 2016 ultra-mobile APUs with Excavator v2 cores and GCN 1.2 graphics. Bristol Ridge as a refresher is the 2016 desktop/mobile APUs like the Athlon X4 970, A12-9800, A10-9700, etc. Only now with the Linux 4.18 kernel is there CPU temperature reporting under Linux for these two year old processors.
Separately, another change to the k10temp driver in Linux 4.18 is support for reporting both the Tctl and Tdie values. Tdie is the die temperature while Tctl is the reported temperature. The Tctl value is used for fan control while Tdie is the real CPU die temperature. On capable CPUs, the driver now reports both values.
Those are the main changes to find this cycle for the HWMON subsystem via this pull request.
| 4
| 1,760,719,635.105775
|
https://www.phoronix.com/news/AMD-K8-Coreboot-Removal
|
AMD K8 Support Stripped Out Of Coreboot
|
Michael Larabel
|
Support for AMD K8 "Hammer" processors including the original Athlon 64 processors and original AMD64 Opterons has been dropped from Coreboot.
Following the recent Coreboot 4.8 release they have decided to remove AMD K8 CPU support and the relevant motherboards from the tree for this open-source BIOS/firmware project. This is being done since the code relies upon the deprecated LATE_CBMEM_INIT functionality and the code has not yet been ported to the newer Coreboot interfaces.
So the K8 CPU support has been removed as well as the associated motherboards. This drops support for dozens of original AMD Athlon motherboards from ASUS, Gigabyte, MSI, Kontron, Tyan, and others. Removing these AMD K8 motherboards lightens the Coreboot source tree by nearly seventy-thousand lines of code while the CPU and northbridge/southbridge removal is another thirty-thousand lines.
Following the AMD K8 support removal was removing VIA C3 and VIA C7 CPU support from Coreboot as well as the associated CN700/CX700/VX800 chipsets and the few VIA supported motherboards.
| 14
| 1,760,719,635.771776
|
https://www.phoronix.com/news/AMDGPU-VCN1-JPEG-Patches
|
AMDGPU Patches Prepping JPEG Support For "Video Core Next"
|
Michael Larabel
|
AMD's Boyuan Zhang has sent out an initial set of 18 patches adding JPEG handling to the AMDGPU kernel driver for VCN "Video Core Next" as the new media encode/decode block found with Raven Ridge APUs for media decode/encode.
The 18 patches are out there with setting up AMDGPU for VCN JPEG ring handling. As of writing there hasn't been any (M)JPEG patches on the Mesa side, but that might be squared away already with the existing Gallium3D state trackers, but we'll see.
Though most of you don't care about JPEG capabilities, VCN 1.0's most notable addition with Raven Ridge was full hardware decoding for VP9. AMD has been getting their VCN Linux support in good shape for over a year now and it's mostly all squared away for Raven Ridge, albeit we keep seeing new occasional feature patches out of AMD for extending this open-source support.
As covered earlier this month, the AMD Raven Ridge Linux support is mostly in good shape now with the latest driver components though some systems may be problematic as illustrated. There is much more commentary about the Raven Ridge Linux support state via the comments in the forums.
| 8
| 1,760,719,636.883347
|
https://www.phoronix.com/news/AMD-Zen-Linux-Firmware-Add
|
AMD Zen CPU Microcode Added To Linux-Firmware Tree, Bulldozer Updated
|
Michael Larabel
|
Ensuring your CPU microcode is kept up-to-date for Zen processors is now a little bit easier with the microcode files being added to the linux-firmware.git collection.
When the Linux Firmware tree was updated on Friday with the newest AMDGPU firmware files for the graphics processors, the Family 17h "Zen" CPU microcode files also made their debut.
AMD has previously made the Ryzen/EPYC CPU microcode files available to Linux users albeit not via the linux-firmware tree that is easily packaged up by most Linux distributions while now it's going through this de facto archive of Linux firmware blobs.
As well as landing the latest Zen CPU microcode files, the Family 15h "Bulldozer" CPU microcode files were updated too.
The latest AMD CPU microcode files are available via the Git tree and should soon be appearing in various Linux distributions when they update their linux-firmware package.
| 12
| 1,760,719,637.772591
|
https://www.phoronix.com/news/AMDGPU-Vega-20-Patches
|
AMD Publishes Open-Source Driver Support For Vega 20
|
Michael Larabel
|
AMD today published their big set of patches bringing open-source Linux kernel support for the "Vega 20" graphics processor.
Vega 20 is the rumored 7nm AMD graphics processor that is said to be up to 70% faster than the current leading RX Vega 64 graphics card, according to some reported leaks. Vega 20 is expected to offer up to 32GB of HBM2 memory and be announced this calendar year, but there is some belief that it might just be a deep learning accelerator and not focused as a gaming graphics card or at least not initially.
There were 57 patches just made public today adding the Vega 20 support to the AMDGPU DRM kernel driver. This new support amounts to almost 13,000 lines of new code, much of which is the addition of emu_soc code -- this is for getting Vega 20 running on emulator code.
With these patches, Vega 20 is taking many of the same code paths in the driver as Vega 10.
The patches enable a new version of UVD video decoding as well, version 7.2, for this Vega 20 hardware.
There are six PCI IDs currently associated as being Vega 20 and they include 0x66A0, 0x66A1, 0x66A2, 0x66A3, 0x66A7, and 0x66AF. Though that doesn't mean there will be six different SKUs as often the hardware vendor reserves extra IDs in case of future revisions/products or for engineering models as well.
For now the Vega 20 Linux support is hidden behind the amdgpu.exp_hw_support=1 kernel module parameter flag until it further stabilizes. Given the timing of these patches, it's likely this experimental Vega 20 hardware support will land in the upcoming Linux 4.18 kernel cycle.
The timing of this Vega 20 Linux kernel driver support doesn't signify too much with regards to when we might see the hardware announced, just that now it met AMD's internal legal review to publish the code. There is code for Vega 20 on the emulator as part of these patches. Additionally, ZFB patches were posted just prior to the Vega 20 unveiling. ZFB is for Zero Frame-Buffer support to allow the AMDGPU driver to use system memory as its video memory for when used for emulators and early silicon models. Given the maturity of other parts of the Radeon Linux graphics stack, hopefully this indicative of AMD now in a position where they are moving forward with their open-source Linux driver support much earlier than many past generations for where they were battling just to get the code open-sourced in time for launch day.The patches out today are all on the kernel driver side while we have our eyes out for support to RadeonSI Gallium3D and the other user-space driver components. Usually those other open-source patches are just a short distance behind the kernel code unveiling.
| 30
| 1,760,719,638.888256
|
https://www.phoronix.com/news/GPUOpen-Compressonator-3.0
|
AMD's Compressonator 3.0 Brings Better Texture Compression
|
Michael Larabel
|
AMD's GPUOpen team has released Compressonator 3.0, the latest major update to this tools collection for dealing with texture and 3D model compression and optimizations for Linux, macOS, and Windows.
The Compressonator 3.0 release brings improved texture compression, mesh optimizations, mesh compression support, and other enhancements.
Coming up for Compressonator 3.1 they are working on full support for viewing glTF 2.0 models inside OpenGL, Vulkan, and Direct3D 12. For the 3.1 release they are also working on Radeon Powered Compression with Compute and Packed Math Libraries, extended support for ETC compression, and other improvements.
More details on Compressonator 3.0 can be found via GPUOpen.com while the cross-platform code for these tools are hosted on GitHub.
| 0
| 1,760,719,639.233648
|
https://www.phoronix.com/news/Linux-4.17-AMD-Power-Fix
|
Linux 4.17 Getting A Fix That May Help AMD Systems Conserve Power When Idling
|
Michael Larabel
|
Besides other promising Linux 4.17 power saving improvements, a separate fix was queued today for potentially helping AMD systems conserve power.
An AMD engineer noticed that with the existing Linux kernel code, using the MWAIT instruction is supported and used but on AMD CPUs but does not allow deeper c-states than C1 with current-generation hardware. The MWAIT x86 instruction is used as a hint for letting the processor enter a CPU-specific optimized state.
So with the kernel code up until now, on AMD CPUs it could forgo entering some of the deeper power-saving states. But with this simple patch, that's no longer the case.
Recent AMD systems support using MWAIT for C1 state. However, MWAIT will not allow deeper cstates than C1 on current systems.
play_dead() expects to use the deepest state available. The deepest state available on AMD systems is reached through SystemIO or HALT. If MWAIT is available, it is preferred over the other methods, so the CPU never reaches the deepest possible state.
Don't try to use MWAIT to play_dead() on AMD systems. Instead, use CPUIDLE to enter the deepest state advertised by firmware. If CPUIDLE is not available then fallback to HALT. The good news is this patch is also copied for the Linux kernel stable series too so should be appearing in the various maintained branches soon rather than having to wait for Linux 4.17.
It will be interesting to see the impact on power-savings for AMD Zen systems as a result of this change. The pull request of the x86 patches today notes it does "prevent excessive power consumption" but does not provide any specifics, but I'll be testing it shortly on Linux Git.
| 22
| 1,760,719,640.911112
|
https://www.phoronix.com/news/AMDGPU-DDX-Color-Management
|
Color Management Support Updated For The AMDGPU X.Org Driver
|
Michael Larabel
|
A bit more than one month ago I wrote about AMD developers working on updated color management support for their AMDGPU X.Org driver. Today a significantly updated patch-set is available.
Leo Sunpeng Li of AMD sent out the second round of these color management patches that are significantly reworked to address some earlier limitations and items brought up during the initial review. This work also depends upon an updated xrandr support eventually too.
If color management is of interest to you, the updated xf86-video-amdgpu patches can be found on amd-gfx.
| 2
| 1,760,719,641.604755
|
https://www.phoronix.com/news/AMD-AOCC-1.2-Released
|
AMD AOCC 1.2 Compiler Released For Zen Systems, Brings FLANG & Retpolines
|
Michael Larabel
|
AMD has released a new update to their AMD Optimizing C/C++ Compiler (AOCC).
AOCC 1.2 is their second major update since debuting this LLVM Clang downstream compiler one year ago following the launch of the Ryzen/EPYC processors. AMD AOCC continues carrying various patches atop the LLVM/Clang compiler tool-chain to cater towards the performance of these "znver1" CPUs.
With AOCC 1.2, they have re-based their compiler support against LLVM Clang 6.0 that was released earlier this year. With that Clang 6.0 release, among other changes, comes Retpolines support for dealing with Spectre V2 mitigation on Linux systems.
AOCC 1.2 also is significant in that it adds FLANG support. FLANG is the experimental Fortran language front-end to LLVM.
More details on the AOCC 1.2 compiler release via community.amd.com. Benchmarks coming up soon.
| 7
| 1,760,719,642.188309
|
https://www.phoronix.com/news/Ryzen-7-2700X-Scaling-CPUFreq
|
Ryzen 7 2700X CPUFreq Scaling Governor Benchmarks On Ubuntu Linux
|
Michael Larabel
|
With this week's Ryzen 5 2600X + Ryzen 7 2700X benchmarks some thought the CPUFreq scaling driver or rather its governors may have been limiting the performance of these Zen+ CPUs, so I ran some additional benchmarks this weekend.
Those launch-day Ryzen 5 2600X / Ryzen 7 2700X Ubuntu Linux benchmarks were using the "performance" governor, but some have alleged that the performance governor may now actually hurt AMD systems... Ondemand, of course, is the default CPUFreq governor on Ubuntu and most other Linux distributions. Some also have said the "schedutil" governor that makes use of the kernel's scheduler utilization data may do better on AMD. So I ran some extra benchmarks while changing between CPUFreq's ondemand (default), performance (normally the best for performance, and what was used in our CPU tests), schedutil (the newest option), and powersave (if you really just care about conserving power).
Tests were done via the Phoronix Test Suite and OpenBenchmarking.org.
In the few graphical tests, the performance governor performed the same as ondemand and schedutil... While obviously powersave led to the slowest performance.
And in the CPU tests, performance tended to be the best or at worst was in line with ondemand and schedutil.
I did not find any situation where using the performance governor led to worse performance.
Nor did Schedutil end up being a magic bullet.
More AMD Ryzen 7 2700X Linux benchmarks via this OpenBenchmarking.org result file.
| 24
| 1,760,719,643.477796
|
https://www.phoronix.com/news/AMDGPU-GFXOFF-Patches
|
AMDGPU DRM Gets "GFXOFF" Patches To Turn Off Graphics Engine
|
Michael Larabel
|
AMD's Huang Rui has posted a set of 20 patches providing "GFXOFF" support for the AMDGPU Direct Rendering Manager Linux kernel driver.
GFXOFF is a new graphics processor feature that allows for powering off the graphics engine when it would otherwise be idle with no graphics workload. Obviously, this would equate to a potentially significant power savings with that engine being able to be shut-off.
But before getting too excited, on the hardware side GFXOFF is only supported currently by Raven Ridge APUs but hopefully we will see it on Radeon GPUs moving forward.
The patches for those interested in testing this new GFXOFF feature can find them on the mailing list.
With the AMD Linux driver performance getting squared away, it's great to see the recent focus by AMD developers on power-saving features such as the recently introduced WattMan-like functionality with the upcoming Linux 4.17 kernel now to the "GFXOFF" functionality.
| 7
| 1,760,719,644.042197
|
https://www.phoronix.com/news/AMDKFD-GFX9-Vega-Patches
|
AMD Posts KFD Support For GFX9/Vega
|
Michael Larabel
|
With the in-development Linux 4.17 kernel there is the long-awaited discrete GPU support in good shape at least for hardware like Polaris and Fiji. While the latest and greatest AMD GPUs are the Vega family, more work has been needed for AMDKFD support. Unfortunately those Vega changes didn't make it in for Linux 4.17, but those patches are now available.
Felix Kuehling of AMD today sent out the patch series adding GFX9/Vega GPU support to the AMDKFD "Fusion Kernel Driver" that is needed for ROCm/OpenCL compute support. This gets the support squared away for "Vega 10" GPUs. There is some prep work for Raven Ridge APU support, but additional patches are coming soon to get that code into shape.
This includes close to five thousand lines of new code for getting the latest-generation AMD graphics architecture into shape for this compute kernel driver while also now attempting to enable atomics for all GPUs. The list of 21 patches for GFX9/Vega10 support can be found for now on amd-gfx.
It's too late for Linux 4.17, but hopefully this code and the pending Raven Ridge patches will be in great shape for premiering in Linux 4.18 later in the summer.
| 8
| 1,760,719,644.837297
|
https://www.phoronix.com/news/LLVM-CUDA-To-AMD-HIP
|
Work Is Underway To Upstream LLVM Clang's CUDA Toolchain For AMDGPU/HIP
|
Michael Larabel
|
A long available tool has been AMD's ROCm HIP that allows converting CUDA code to portable C++ code that in turn can be executed on Radeon GPUs. There is now work on getting the upstream LLVM Clang compiler's CUDA toolchain support to also support HIP.
HIP's hipify tool can convert CUDA code to HIP for execution on NVIDIA/AMD GPUs. HIP also consists of a portable C++ language for execution across GPU vendors. Those not familiar with HIP can learn more via its ROCm tool repository.
What initially got me digging into the LLVM/Clang upstreaming work was seeing on Thursday: [CUDA] Add amdgpu sub archs. Clang's CUDA code now not only listing NVIDIA GPU micro-architectures but also the Radeon GPU generations backed by the LLVM AMDGPU compiler back-end.
HIP/HIP-ify has made use of LLVM/Clang already, but has relied upon out-of-tree changes.
Tracing things back, it's part of an ongoing effort to allow LLVM Clang's CUDA toolchain to support the HIP language mode and AMDGPU device support. There's this code review part of the process. Its summary:
This patch let CUDA toolchain support HIP language mode. It includes:
Add file type hip for toolchain.
Create specific compiler jobs for HIP.
Choose specific header/libraries for HIP.
Use clang-offload-bindler to create binary for device ISA.Hopefully this will help the adoption of HIP with getting this support upstreamed into the LLVM Clang CUDA code, albeit it's not all merged yet. This is official AMD work with the LLVM patches being led by AMD's Yaxun Liu.
It's great seeing more of the AMD/ROCm code being upstreamed, especially now on the Clang side. With Linux 4.17 is also the big piece for initial dGPU support in AMDKFD kernel driver to start allowing ROCm -- including ROCm OpenCL -- to begin working off an unmodified, upstream kernel. With more pieces in upstream, deploying ROCm across different Linux distributions will become much easier.
| 12
| 1,760,719,645.306853
|
https://www.phoronix.com/news/Raven-Ridge-March-Update
|
Latest AMDGPU DC Plays Nicer With Raven Ridge But Still Linux Gaming Stability Issues
|
Michael Larabel
|
Back in February was the exciting AMD Raven Ridge desktop APU launch with the Zen CPU cores and Vega graphics. Sadly, however, the Raven Ridge Linux support still appears to be a bit problematic but there have been improvements in recent weeks.
Going back to our original testing of Raven Ridge with the Ryzen 3 2200G and Ryzen 5 2400G APUs, there were some display issues as well as some game stability issues.
Fortunately, with the latest Linux 4.16 Git kernel there have been some Raven fixes trickling in this month. Additionally, there are more improvements for Linux 4.17 as the next kernel cycle.
This week I spent some time trying the Ryzen 3 2200G and Ryzen 5 2400G retail APUs with retail motherboards on their latest BIOS and using Ubuntu 18.04 LTS with both the Linux 4.16 and DRM-Next kernels. Additionally, trying out Mesa 18.0-rc5 and Mesa 18.1-devel from the Padoka PPAs.
It's good to report now that the display issues appear fixed. I am no longer running into any mode-setting issues or other display problems for the Vega 8 graphics.
But it's not all smooth sailing as when trying to run some fresh Linux gaming benchmarks, I routinely am encountering system hangs. It seems to be semi-random but generally can't get a successful run out of a complete set of our usual Linux gaming benchmarks without hitting a hang on either systems. But occasionally when running pure CPU/system workloads, I am able to occasionally hang the system as well, so it might not be an issue isolated to just the AMDGPU DRM driver...
So for now the Raven Ridge Linux support doesn't appear fully baked yet. For what it's worth besides the two different Raven APUs, I have been testing with two different motherboards as well to the same behavior. If any other Raven Ridge Linux users have a different experience, be sure to share in the forums as well as with any tricks you may have learned along the way.
| 16
| 1,760,719,646.175055
|
https://www.phoronix.com/news/AMD-CTS-Firmware-Updates
|
AMD To Issue PSP/BIOS Firmware Updates For Recent Vulnerabilities
|
Michael Larabel
|
Last week was the controversial publishing of the "AMD Flaws" CPU vulnerabilities for Ryzen and EPYC processors. AMD has now issued their first public update on the matter and have said they will be issuing PSP firmware and BIOS updates for mitigation.
AMD on Tuesday night published their initial assessment into these MASTERKEY/RYZENFALL/FALLOUT/CHIMERA vulnerabilities disclosed last week by CTS Labs. While the motives of CTS Labs are still under the microscope and questionable, AMD will be working to address the issues brought up. AMD has also reaffirmed in their statement that these vulnerabilities are only prone to exploit if the bad actor has already administrator/root privileges to the system -- severely limiting the possible impact of these vulnerabilities and making them much less severe than say Spectre/Meltdown.
AMD though does plan to issue PSP firmware updates and BIOS updates in the coming weeks for mitigating these issues. They do not believe their mitigations of these issues will lead to any performance impact.
More details coming in the weeks ahead but for now their public statement can be found at community.amd.com.
| 64
| 1,760,719,646.665506
|
https://www.phoronix.com/news/AMDVLK-Sub-Group-Multi-View
|
AMDVLK Vulkan Driver Updated With Improvements For Sub-Groups & Multi-View
|
Michael Larabel
|
The AMD developers working on their official cross-platform "AMDVLK" Vulkan driver have updated their open-source code-base for Linux users.
On Friday the AMD developers pushed to the open-source repository their latest work, their first update since introducing Vulkan 1.1 support back on launch day earlier this month.
There are many changes in this latest code push including making use of instance and device specific dispatch tables, handling unaligned memory to image and image to memory copies with the DMA queue, complete support for the VK_EXT_sampler_filter_minmax extension, updates to KHR_descriptor_update_template, an assert fix while running DOOM, and several LLPC (LLVM Pipeline Compiler) fixes.
This code update also has significant updates to the VK_KHR_subgroup and VK_KHR_multiview support. The sub-group support as introduced in Vulkan 1.1 has been refined with this latest code drop to adding the sub-group built-ins to compute shaders, support for shufflexor/shuffleup/shuffledown functions, and more. On the Vulkan multi-view front is a more complete implementation, some rewritten functions, and interaction support between VK_KHR_multiview and VK_KHR_device_group.
Instructions for building this latest AMDVLK Vulkan Linux driver from source can be found via the GitHub guide.
| 1
| 1,760,719,647.992742
|
https://www.phoronix.com/news/AMDFLAWS-Vulnerability-Outted
|
AMD Secure Processor & Ryzen Chipsets Reportedly Vulnerable To Exploit
|
Michael Larabel
|
Just two months after the big Spectre and Meltdown CPU vulnerabilities were disclosed, Israeli security researchers have published 13 security vulnerabilities claiming to affect AMD Ryzen and EPYC product lines.
These vulnerabilities are being called "AMDFLAWS" and the vulnerabilities have names like MASTERKEY, RYZENFALL, FALLOUT, CHIMERA, and the PSP PRIVILEGE escalation amounting to 13 vulnerabilities in total.
These vulnerabilities could allow the AMD Secure Processor to be overrode with malicious code, read/write from protected memory areas, backdoors were found within the AMD Ryzen chipset, and other vulnerabilities around the AMD Secure Processor and AMD's current chipsets. But the actual technical information on these vulnerabilities is rather light at the moment with some questioning their merit. It also appears a BIOS flash or local root access would also be needed for at least making some of these vulnerabilities a reality.
Some of these vulnerabilities may be fixed ahead via new firmware updates while CHIMERA claims to be a hardware vulnerability that cannot be fixed, at least according to these security researchers. CTS Labs though only provided AMD with a 24 hour lead time before making this public disclosure, so it may be a while before these issues are fully investigated and more details from AMD... Sadly this disclosure is not well structured and raises more questions than answers, so stay tuned.
More details via AMDFLAWS.com.Update: Some security researchers are indeed questioning the claims raised by CTS Labs about these supposed AMD Zen vulnerabilities... So fortunately it might not end up being too bad, but we will wait and see what is published by AMD.
| 82
| 1,760,719,648.051894
|
https://www.phoronix.com/news/GPUOpen-VMA-Nearly-2.0
|
AMD's Vulkan Memory Allocator Nears Version 2.0
|
Michael Larabel
|
Last year AMD's GPUOpen group posted the Vulkan Memory Allocator while coming soon is version 2.0 of this code-base.
The GPUOpen Vulkan Memory Allocator is an easy-to-use memory allocation library that tries to rid your Vulkan code-base of the necessary boilerplate code for memory setup and other routine tasks under Vulkan while making the memory easier to manage across drivers, handle out-of-memory conditions, and is cross-platform. Version 1.0 was out last year but v2.0 appears to be quickly approaching with their current Git activity referencing 2.0 alpha versions.
Version 2.0 Alpha 8 was marked in the VulkanMemoryAllocator Git this weekend with the addition of several new functions for finding memory types and more. Besides extending the API, the GPUOpen developers for v2.0 have added Android support, there is initial Apple/macOS support via MoltenVK support, a number of documentation improvements, and various fixes.
Those wanting to learn more about this Vulkan memory allocation library can do so via GitHub.
NVIDIA meanwhile has posted some updates to VkHLF, their high-level Vulkan framework. VkHLF also now has initial Apple/macOS support thanks to the MoltenVK support and various other updates outlined here.
| 1
| 1,760,719,649.550663
|
https://www.phoronix.com/news/Raven-Display-Improvements-4.17
|
It Looks Like Raven Ridge Desktop APUs Will Work Better With Linux 4.17
|
Michael Larabel
|
Sadly right now with the highly-anticipated Vega+Zen Raven Ridge desktop APUs is in fairly rough shape with some hangs, display corruption, etc. Fortunately it looks like Linux 4.17 support will be in better shape.
Phoronix reader Piotr was hitting similar display issues on his new Raven Ridge desktop to what I had been encountering. He tested out a new patch and found his screen now works.
Unfortunately though it doesn't appear at the moment like these changes to AMDGPU DC will be queued up for the upcoming Linux 4.16.0 stable release but rather are queuing for Linux 4.17.
After that initial patch discussion on the mailing list, there was a new AMDGPU DC patch series posted this morning. This latest round of display code updates has the fix for a corrupt screen when booting on Raven Ridge hardware as well as a "bunch of Raven patches and fixes all around." For non-Raven users this code also has backlight support for pre-DCE11 hardware.
These 28 patches are queuing for the Linux 4.17 kernel cycle. As soon as time allows I will try them out on the Ryzen 5 2400G and Ryzen 3 2200G to see if these modern APUs now are working nicely across the board under Linux. But for those sticking strictly to Linux kernel stable releases, unless we find AMD trying to backport some of the Raven work to 4.16, the Linux 4.17.0 kernel won't be out until late June or early to mid July.
| 32
| 1,760,719,649.780784
|
https://www.phoronix.com/news/Sun4i-A83T-HDMI-Linux-4.17
|
Allwinner A83T Will Support HDMI With Linux 4.17
|
Michael Larabel
|
The Sun4i DRM driver work has been progressing a lot since its mainline introduction two years ago with Linux 4.7. With the Linux 4.17 cycle, the A83T SoC will have initial HDMI output support.
If you happen to have a tablet or other device powered by the Allwinner A83T, it should finally have working HDMI out support when using the Sun4i DRM driver with the kernel update coming later this year.
The A83T SoC makes use of Cortex-A7 CPU cores with a PowerVR SGX544 GPU albeit this DRM driver is just for the Allwinner display side of things. The HDMI support was queued up this week in drm-misc-next.
Other work in drm-misc-next includes documentation improvements, new backlight helpers, and various miscellaneous improvements for the Direct Rendering Manager subsystem.
| 5
| 1,760,719,651.282911
|
https://www.phoronix.com/news/AMD-EPYC-3000-Ryzen-V1000
|
AMD Launches EPYC Embedded 3000 & Ryzen Embedded V1000 Series
|
Michael Larabel
|
AMD is taking their Zen microarchitecture to the embedded space now with the announcement of the AMD Launches EPYC Embedded 3000 and Ryzen Embedded V1000 series.
Ryzen Embedded V1000 is targeting medical imaging / industrial systems / digital gaming / thin clients while the EPYC Embedded 3000 chips are intended for networking, storage, and edge computing.
The EPYC Embedded 3000 products will feature between four and 16 CPU cores, TDPs from 30 to 100 Watts, up to 64 PCI Express lanes, and up to a 32MB L3 cache.
The Ryzen Embedded V1000 offerings will offer up to four cores / eight threads and up to 11 compute units from an onboard Vega GPU. The TDPs on those Ryzen Embedded offerings are from 12W to 54W and up to 16 PCI-E lanes.
More details via this morning's press release.
| 22
| 1,760,719,652.350252
|
https://www.phoronix.com/news/Ryzen-3-2200G-vRAM-Size
|
Ryzen 3 2200G Video Memory Size Testing On Linux
|
Michael Larabel
|
One of the discussion items in the forums this week was about the video memory allowance for the Vega graphics on Raven Ridge APUs as well as efficiences or inefficiencies around the TTM memory manager as used by the AMDGPU kernel driver. Here are some vRAM size tests with the Ryzen 3 2200G.
This comparison is quite small and took more than one day to complete due to the multitude of Raven Ridge stability issues encountered and often needing to restart a round of tests due to hangs. In the Vega 8 / Vega 11 tests delivered last week, all of them were done with the 2GB of video memory allotted, the maximum supported by Raven Ridge, as configured from the BIOS.
For this round of testing the Linux OpenGL/Vulkan game tests were run in 256MB, 512MB, 1GB, and 2GB vRAM sizes of the 2 x 4GB DDR4-3200 memory. In the days ahead will also be some reference numbers for those wondering the precise Linux impact with different RAM speeds for Raven Ridge.
But onto the few results today for those curious on the vRAM size impact...
With DiRT Rally there was no measurable difference between 512MB and 2GB. At 256MB there are no results due to repeated hangs, but I'm not sure if that is due to the low vRAM budget or rather the broader Raven Ridge Linux issues I've been encountered. For the larger vRAM sizes I had to restart testing a few times as well due to hangs.
Dota 2 with Vulkan dipped going from 256MB to 512MB but was noticeably faster with a 1GB and 2GB allowance.
Mad Max was faster with 1GB+.
Serious Sam 3 BFE is another game with frequent hangs for Raven Ridge on Linux. There were small performance improvements going from 512MB to 2GB.
And not much of a difference for the open-source OpenGL games. Unfortunately, just this short one-page comparison due to repeated hangs in the other game tests with this MSI B350 system.
| 32
| 1,760,719,653.712304
|
https://www.phoronix.com/news/AMD-Raven-Ridge-Mobo-Linux
|
AMD's Raven Ridge Botchy Linux Support Appears Worse With Some Motherboards/BIOS
|
Michael Larabel
|
With my launch testing of the Raven Ridge desktop APUs with the Ryzen 5 2400G and Ryzen 3 2200G there were some stability issues to report and some hangs within games and mode-setting issues. It appears those issues are exacerbated with some motherboards: the past few days with two different AMD B350 motherboards have been a real pain getting the current AMDGPU driver stack working -- and even Linux 4.17 AMDGPU WIP code -- on either of these Raven Ridge APUs.
It's been three days since last having any new Raven Ridge Linux tests not out of running out of benchmarking ideas for these interesting Zen+Vega chips or taking a break, but because I've simply been struggling to get the systems working well and reliably. With my initial testing I was using an MSI X370 XPOWER GAMING TITANIUM motherboard out of having it available for testing, but I doubt most people would be using a such a $260~270 USD motherboard to go with a $99~169 APU. The motherboards I picked up for the Raven Ridge benchmarking systems in the longer term were a MSI B350M GAMING PRO and ASUS PRIME B350M-E motherboards. They use the B350 chipset, are micro-ATX, and cost less than $100 USD -- certainly more appropriate for pairing with the Ryzen 3 2200G that retails for $99 or the Ryzen 5 2400G that goes for $169, rather than the X370 motherboards that can cost around $300 USD.
With both of these motherboards I first had to flash their BIOS to their latest versions when using a Ryzen 7 1800X CPU in order to enable the Raven Ridge APU support. After that BIOS update with that CPU and an external graphics card, I tried the two Raven Ridge APUs in each of these boards. When installing Ubuntu 17.10 and Fedora 27 with their pre-4.15 kernels, the displays were working albeit without hardware acceleration since the Raven Ridge AMDGPU kernel driver support relies upon Linux 4.15+ due to its dependence on the DC display code stack.
On both Ubuntu 17.10 and Fedora 27 as testing on both systems, I then proceeded to run all available system updates, fetch the latest AMDGPU firmware files from linux-firmware.git, tried with the latest Mesa patches, and also using 4.15+ kernels. Sadly, with both systems it's often difficult to get a working display on Linux 4.15+ short of booting with "nomodeset" to disable the AMDGPU driver. Occasionally I get lucky with the mode-setting going alright, but when beginning the tests, hangs are still very common. It does also seem to have a better success rate when doing a cold boot rather than a restart on problems.
I've tried out the Linux 4.15.4 stable kernel, Linux 4.16 Git, and Alex Deucher's 4.17 work-in-progress AMDGPU development branch. With all of those different kernels, the system often gets stuck at boot with varying behavior: either no display at all but can SSH in only to find nothing helpful from the dmesg output, a hard hang with no ability to SSH into the system. When trying the APUs back in the MSI X370 XPOWER GAMING TITANIUM board, it's back to working albeit with the spotty support. But when I do get lucky and hit the desktop, when engaging in common OpenGL/Vulkan Linux games, hangs are still an issue.
I have tried different RAM modules and enabled/disabled the memory profiles from the BIOS, adjusting the amount of exposed RAM for the graphics, trying different DVI / HDMI / DP outputs and monitors, other kernel command line options, varying versions of Mesa, and changing around other basic tunables for trying to get the Vega APU graphics working on these two APUs and two B350 motherboards on their latest firmware. Unfortunately rolling back the BIOS isn't an option since only their latest BIOS support the Raven Ridge APUs. When putting in a discrete GPU, the systems boot up fine with the AMDGPU driver. Trying out AMDGPU-PRO 17.50 on Ubuntu 16.04 LTS also hadn't worked out nor did an Antergos 18.2 live image.
Whenever seeing new AMDGPU DC code drops in recent months, they have often mentioned "Raven" fixes, but I really didn't expect the support to be this bad at launch. Open-source AMD developers have been working on the Raven Ridge support going back to last year and all. Thus I'm thinking some motherboard vendors with their BIOS updates changed some behavior adversely affecting the Linux driver support and/or the pre-production bring-up of Raven with the Linux driver stack is a lot different than what ended up being with the production hardware/BIOS... Perhaps this is why AMD hadn't offered any review samples this time around for our Linux testing or even briefed on the launch.
Disappointing and very frustrating after buying the hardware in order to deliver this Linux testing. For now back to toying around with this Raven Ridge hardware to see if I get lucky and can deal with enough hangs or failed boots to get some more benchmarks completed today.
| 116
| 1,760,719,653.953164
|
https://www.phoronix.com/news/Ryzen-Raven-Ridge-Tuesday
|
Ryzen 3 2200G + Ryzen 5 2400G Linux Benchmarks Coming Tomorrow
|
Michael Larabel
|
Tomorrow I will be posting our initial benchmarks of the Ryzen 3 2200G and Ryzen 5 2400G "Raven Ridge" APUs with the Zen CPU cores plus Vega graphics.
The embargo just lifted for media outlets to publish their reviews of Raven Ridge, At least under Windows, the reviews seem to be positive with decent CPU performance and good graphics performance for an APU. Under Linux, as mentioned many times now including in the Linux expectations for Raven Ridge from yesterday, you will need a very updated Linux graphics driver stack.
But even with the Linux 4.15 kernel, it looks like some issues remain. Golem (German) notes that even with their latest Raven Ridge firmware and Linux 4.15 kernel, they were running into issues: basically only half the display was working correctly. Though as also mentioned in various Phoronix articles, there are many AMDGPU DC improvements for Raven Ridge in the in-development Linux 4.16.
Beginning tomorrow will be my stream of extensive Linux testing for the Ryzen 3 2200G and Ryzen 5 2400G. Unfortunately AMD didn't send out any Raven Ridge review samples for Linux testing at Phoronix. This morning I ended up buying retail these new desktop APUs and they will be arriving on Tuesday and have already begun working on other fresh comparison data points for this testing. So tomorrow you will get the initial results here while plenty other graphics tests and more will be coming in the days ahead.
XDA Developers did upload some initial 2200G and 2400G CPU benchmark results. You can find their initial data on these APUs via this OpenBenchmarking.org result file. Or simply install the Phoronix Test Suite and run phoronix-test-suite benchmark 1802105-GARW-RYZENVE64 to see how your own Linux CPU performance would compare to those numbers.
Stay tuned for more information. As a reminder if you too are ordering Raven Ridge hardware today, you will want to be using Linux 4.15 but it's looking like Linux 4.16 Git might even be needed for correct display, Mesa 18.0 or 18.1-dev built against LLVM 6.0 or 7.0 SVN, and the latest Raven Ridge binary firmware files. For anyone with special test requests for Raven Ridge, feel free to contact me or post in the forums -- of course, first priority goes to the premium supporters in honoring test requests.
| 28
| 1,760,719,655.363011
|
https://www.phoronix.com/news/Raven-Ridge-Desktop-Tomorrow
|
Raven Ridge Desktop APUs Come Out Tomorrow, The Likely Linux Requirements
|
Michael Larabel
|
For those that haven't been paying attention or have lost track of time, the first two Raven Ridge desktop APUs are expected to become available tomorrow with their Zen CPU cores and Vega graphics.
12 February is the launch date for the Ryzen 3 2200G and Ryzen 5 2400G. The Ryzen 3 2200G is the $99 USD CPU that is quad-core without SMT, 3.5GHz base frequency and 3.7GHz turbo frequency. The 2200G has Vega 8 graphics with eight compute units and a 1.1GHz clock frequency. The Ryzen 5 2400G meanwhile is a quad-core / eight thread design with 3.6GHz base frequency, 3.9GHz turbo frequency, and Vega 11 graphics meaning 11 compute units and a 1250MHz clock speed. Both of these desktop APU parts are rated for a 65 Watt TDP.
While these are the first Ryzen 2000 series processors, they are not Zen+ 12nm CPUs like those expected to launch in the months ahead but still manufactured on a 14nm process.
Unfortunately, tomorrow we will not have Linux benchmarks to share of the Ryzen 3 2200G / Ryzen 5 2400G. While we have been having great experiences with Ryzen / Threadripper / EPYC ever since AMD got the "performance marginality problem" under control, for reasons unknown we didn't receive any review samples of these Raven Ridge desktop APUs. Quite a shame really as we've been really looking forward to Raven Ridge and these APUs have a lot of potential for Linux customers given the good open-source Radeon graphics driver stack and the Zen Linux support now in order. But a result, tomorrow when these CPUs become available through retail channels I will be buying one or both of them depending upon availability and actual pricing... So assuming they are in good supply tomorrow, by the middle of the week we should be able to begin delivering plenty of Raven Ridge Linux desktop benchmarks.
In terms of likely Linux requirements, you will need to be using at least the Linux 4.15 kernel that was released in January. Raven Ridge display support like the RX Vega GPUs is implemented only with AMDGPU DC, so it places a requirement at least of Linux 4.15 for having mainline AMDGPU DC support. By using Linux 4.15 you can also have Zen temperature monitoring support (assuming no offsets or other driver adjustments are needed for these new desktop APUs) and all around good support plus Spectre protection.
With the in-development Linux 4.16 kernel there are a number of Raven Ridge fixes, but without being able to test any actual hardware yet, I don't know if any of those fixes/changes are required good 2200G/2400G APU support. But generally speaking, the newer the kernel will yield the better AMDGPU experience. Alternatively, if on Ubuntu 16.04 LTS or another enterprise Linux distribution, there will presumably be support with the AMDGPU-PRO hybrid driver.
Besides needing Linux 4.15~4.16, it's also recommended to be using Mesa 18.0 or 18.1-dev Git built against LLVM 6.0/7.0 for having the best and fastest possible RadeonSI/RADV graphics support. Also using GCC 7 as the default system compiler for at least having "znver1" tuning support but the Zen tuning is even better with the upcoming GCC 8 stable release.
Long story short though, I will have much more information to share in the days ahead once getting my hands on the Ryzen 3 2200G and/or Ryzen 5 2400G for Linux testing and eventual BSD tests too. If you enjoy all my Linux hardware reviews and benchmarks, you can show your support by joining Phoronix Premium or making a PayPal tip. Stay tuned for the long-awaited Zen+Vega APU Linux benchmarks!
| 39
| 1,760,719,655.450828
|
https://www.phoronix.com/news/Core-i9-Threadripper-Linux-4.15
|
Core i9 7980XE vs. Threadripper 1950X On Linux 4.15 With Ubuntu 18.04
|
Michael Larabel
|
With more than one hundred different benchmarks, here are some fresh tests of the Core i9 7980XE and Ryzen Threadripper 1950X boxes when running on the Linux 4.15.2 stable kernel atop a daily snapshot of Ubuntu 18.04 LTS.
For those curious about how the CPU performance is on Linux 4.15 following the Spectre and Meltdown mitigation work, in the days ahead will be a larger CPU comparison using the latest kernel release. For those wondering about the Linux 4.15 + Ubuntu 18.04 performance for high-end desktop platforms, I ran a larger comparison of tests just on the Core i9 7980XE and Threadripper 1950X systems.
Both boxes were using 4 x 4GB DDR4-3600 memory, NVMe SSD storage (different models, but these tests were pretty much isolated anyways to just being CPU tests), and running on Ubuntu 18.04 x86_64 with the Linux 4.15.2 kernel, the latest stable point release as of testing.
As you can see from this massive selection of Linux CPU benchmarks, the Threadripper 1950X and Core i9 7980XE do trade blows in a number of benchmarks. For a closer look at this massive selection of open-source Linux benchmarks, see this OpenBenchmarking.org result file. Of course keep in mind the price difference between these CPUs (the i9-7980XE being more expensive), but these weekend benchmarks are mostly being put out for reference purposes while more analysis is coming to our larger CPU comparison with a more concise set of benchmarks. Enjoy.
| 13
| 1,760,719,656.723508
|
https://www.phoronix.com/news/AMD-AOCC-1.1-Released
|
AMD AOCC Compiler 1.1 Released For Zen CPUs
|
Michael Larabel
|
AOCC 1.1 is the second public release of the AMD Optimizing C/C++ Compiler designed for Ryzen/Threadripper/EPYC processors.
Back in May AMD released AOCC 1.0 as their optimized compiler stack for Ryzen CPUs. AOCC is the replacement to the company's older AMD Open64 compiler designed for older CPUs. With Open64 sadly being a relic now of the past, AOCC is based upon LLVM/Clang.
AOCC 1.1 was quietly released back in December to not much fanfare at all. I only heard of this new release over the weekend thanks to a Phoronix reader discovering the update.
AOCC 1.1 contains further tuning for Zen CPU cores, enhanced loop optimizations / inlining / other high-level optimizations, improved vectorization / code generation, and more. AOCC 1.0 is derived from the LLVM Clang 6.0 code-base released at the end of last year.
AOCC 1.1 also now ships optimized AMD libraries such as the libM Math Library, uses the LLVM "LLD" linker by default, and other improvements.
With AOCC 1.1 offering more changes compared to the AOCC 1.0 release, I plan on running some fresh EPYC/Ryzen compiler benchmarks this week. AOCC 1.0 didn't offer much at the time compared to LLVM/Clang trunk, but given more changes now, it will be interesting to see how the performance compares to Clang 6.0 and Clang 7.0 SVN.
The AOCC 1.1 binary for Linux x86_64 can be downloaded from developer.amd.com. Stay tuned for fresh benchmark results.
| 8
| 1,760,719,657.048049
|
https://www.phoronix.com/news/TR-1900X-Temperature-Fix
|
Ryzen Threadripper 1900X Should Report The Correct Temperature With Linux 4.16
|
Michael Larabel
|
While the just-released Linux 4.15 kernel brings AMD Zen CPU temperature reporting support for Ryzen/Threadripper/EPYC processors, an oversight in the k10temp driver code is yielding an incorrect temperature for the Threadripper 1900X.
Several Ryzen 5/7 and Threadripper processors need a temperature offset for correct temperature reporting. The Ryzen Threadripper 1920X/1950X have a 27 degree offset with the original code landed in Linux 4.15, but the 1900X did not.
Now with the Linux 4.16 kernel the k10temp driver has a 27 degree offset in place for the Ryzen Threadripper 1900X to allow it to report the correct CPU temperature.
This change has already been submitted for the Linux 4.16 merge window, we'll see if the one-liner change gets backported to a Linux 4.15.x point release. The change was sent in as part of hwmon-next. This tree also adds a new W83773G driver and fan control support for PMBus drivers, besides other fixes and improvements to other temperature kernel drivers.
| 1
| 1,760,719,658.6516
|
https://www.phoronix.com/news/Radeon-VCN-Video-Encode
|
Radeon VCN Gets Mesa Patches For HEVC Main Video Encode
|
Michael Larabel
|
For those planning to pick up a Raven Ridge laptop or the forthcoming desktop APUs, the Mesa driver now has patches for enabling H.265/HEVC video encode support for VCN 1.0 on Raven hardware.
AMD developers today sent out a set of 12 patches for adding HEVC encode support to the Gallium3D VL interface, Radeon VCN specific HEVC encode bits, and added HEVC encoding support to the Gallium3D VA "video acceleration" state tracker.
This HEVC main profile support is just enabled for Raven Ridge, which introduces the "Video Core Next" block that offers both video encode/decode compared to the prior Unified Video Decoding (UVD) and Video Coding Engine (VCE) hardware. Up to now only MPEG4 AVC video encoding was supported by the Radeon Gallium3D implementation.
All of Mesa's HEVC code up to now was focused just on video decoding support and not encoding. Though for reference the separate project VA-API has supported HEVC encode with its APIs since 2015. Polaris and Vega GPUs do support HEVC encode as well, but there hasn't been any patches yet for wiring that support up.
| 14
| 1,760,719,658.674969
|
https://www.phoronix.com/news/AMD-Ryzen-CES-2018
|
AMD Cuts Ryzen Prices, Confirms New Hardware, New Ryzen CPUs With Vega
|
Michael Larabel
|
While Intel announced their new CPUs with Radeon Vega M graphics, AMD had a host of announcements on their own for getting CES 2018 started with some excitement.
- First up, AMD is slashing prices on many of their Ryzen desktop processors. For example, the Threadripper 1900X is droping from $549 to $449, the Ryzen 7 1800X from $499 to $349, Ryzen 7 1700X from $399 to $309, and smaller discounts on the lower-end Ryzen 5 CPUs. Especially at the high-end, these new permanent price cuts are quite significant.
- AMD also announced the Ryzen 3 2200G and Ryzen 5 2400G that will begin shipping 12 February. These 65 Watt Ryzen desktop CPUs also feature onboard Radeon graphics. The 2200G has Vega graphics with 8 compute units while the Ryzen 5 2400G has 11 compute units and its Vega GPU clock speeds up to 1250MHz.
- AMD confirms Zen 12nm processors are now sampling.
- The "Zen 2" design is considered complete.
- AMD has a working 7nm AMD product that is a GPU built for machine learning.
- AMD X470 and B450 motherboards will begin hitting the market in April.
- AMD will be issuing a Radeon Software driver update to introduce HDMI 2.1 Variable Refresh Rate support for Radeon RX graphics cards. (Hopefully Linux driver support too via AMDGPU DC.)
More information at AMD.com.
| 67
| 1,760,719,660.08615
|
https://www.phoronix.com/news/AMD-Retpoline-Linux-4.15-FX-Zen
|
AMD Retpoline Benchmarks From FX To Threadripper & EPYC
|
Michael Larabel
|
For those curious about the performance impact of the Retpoline patches as found in the latest Linux 4.15 kernel, here are some benchmarks on an assortment of old and new AMD Linux systems.
With the latest Linux 4.15 kernel there is not only the original Retpoline code but also an AMD-specific minimal thunk that is optimized for AMD's processor and should perform better for this Sepctre Variant Two mitigation. The Retpoline mode can be configured via the spectre_v2= kernel parameter and those interested can find more details via the kernel documentation.
Tests were done on an AMD FX-8370, A10-7870K, Ryzen 7 1800X, Ryzen Threadripper 1950X, and EPYC 7601. The systems were a mix of Debian and Ubuntu. This testing isn't meant for comparing the performance of the different processors but just looking at a range of the Retpoline performance impact on an array of different hardware with different CPUs, storage, RAM, OS, etc.
With each system the Linux 4.15 Git kernel twas tested using no Retpoline, the minimal generic Retpoline, and the minimal AMD-specific Retpoline.
The Threadripper system was more affected by the random read performance than the other AMD boxes with Retpoline.
For the most part, the Retpoline impact on AMD systems with the minimal thunk isn't too bad.
Even in the I/O benchmarks, the impact isn't too severe with the possible exception of the Threadripper box.
In CPU benchmarks, there was little change in the results.
The Linux kernel build speed wasn't impaired by the AMD Retpoline.
Those wishing to dig through more of this data can do so via OpenBenchmarking.org.
| 12
| 1,760,719,660.288175
|
https://www.phoronix.com/news/AMD-Branch-Prediction-Still
|
AMD Did NOT Disable Branch Prediction With A Zen Microcode Update
|
Michael Larabel
|
With the plethora of software security updates coming out over the past few days in the wake of the Meltdown and Spectre disclosure, released by SUSE was a Family 17h "Zen" CPU microcode update that we have yet to see elsewhere... It claims to disables branch prediction, but I've confirmed with AMD that is not actually the case.
AMD did post a processor security notice where they noted their hardware was not vulnerable to variant threee / rogue data cache load, for the "branch target injection" variant that there was "near zero risk" for exploiting, and with the bounds check bypass it would be resolved by software/OS updates.
Along with the Linux kernel patches for enabling KPTI (Page Table Isolation), SUSE issued a security bulletin where they added an AMD microcode update. The bulletin mentions, "This new firmware disables branch prediction on AMD family 17h processor to mitigate a attack on the branch predictor that could lead to information disclosure from e.g. kernel memory." The AMD change-log does note this AMD microcode update is indeed for CVE-2017-5715, a.k.a. SPECTRE.
But surprisingly I have yet to see any other Linux distribution vendors promoting this new microcode_amd_fam17h.bin microcode file for disabling branch prediction on these latest AMD Ryzen/Threadripper/EPYC processors. This new Family 17h microcode file also hasn't been added as of writing to the linux-firmware.git tree.
I reached out to AMD and on Friday heard back. They wrote in an email to Phoronix that this Zen/17h microcode update does not disable branch prediction. They'll be working with SUSE to re-clarify this microcode update description... But as far as what this microcode update does in the wake of SPECTRE they have yet to clarify or why this microcode binary has yet to make it to other Linux distributions. If/when I hear anything more, I'll certainly post about it but doesn't appear to be anything as dramatic as disabling branch prediction, which could have slaughtered their CPU performance.
| 27
| 1,760,719,661.770106
|
https://www.phoronix.com/news/AMD-PSP-2018-Vulnerability
|
AMD PSP Affected By Vulnerability
|
Michael Larabel
|
While all eyes have been on Intel this week with the Spectre and Meltdown vulnerabilities, a disclosure was publicly made this week surrounding AMD's PSP Secure Processor in an unrelated security bulletin.
AMD's Secure Processor / Platform Security Processor (PSP) that is akin to Intel's Management Engine (ME) is reportedly vulnerable to attack.
A member of Google's Cloud Security Team discovered through static analysis that a function in PSP's firmware TPM code is vulnerable to a stack-based overflow due to missing bounds checks. Submitting a specially-crafted certificate to the fTPM trustlet code can lead to an overflow and then full control on the program counter.
Google reported this issue to the AMD Security Team in September and then in December began rolling out a software fix. Following the 90-day disclosure process, the information was made public here.Update: Contrary to the original security notice, AMD has now confirmed to us this vulnerability isn't subject to remote code execution.
| 83
| 1,760,719,662.221352
|
https://www.phoronix.com/news/x86-PTI-EPYC-Linux-4.15-Test
|
For Now At Least AMD CPUs Are Also Reported As "Insecure"
|
Michael Larabel
|
Right now with the big mysterious security vulnerability causing the rush of the x86 Page Table Isolation work that landed in the Linux kernel days ago, it's believed to be a problem only affecting Intel CPUs. But at least for now the mainline kernel is still treating AMD CPUs as "insecure" and is too taking a performance hit.
Besides my initial benchmarks of the performance impact as a result of this x86 workaround in the Linux 4.15 kernel, I've been working on various other tests since yesterday and one of them was just seeing what happens on AMD hardware.
Back on 26 December is when Tom Lendacky of AMD posted a patch to confirm this PTI problem shouldn't affect the company's processors -- at least with what information is currently known. Lendacky wrote, "AMD processors are not subject to the types of attacks that the kernel page table isolation feature protects against. The AMD microarchitecture does not allow memory references, including speculative references, that access higher privileged data when running in a lesser privileged mode when that access would result in a page fault."
But over one week later, that patch has yet to be merged to the mainline kernel. When booting the Linux 4.15 kernel on an AMD EPYC box, indeed, for now the AMD CPU is still treated with a bug of "insecure_cpu."
An immediate workaround at least until the AMD patch lands where PTI isn't applied to AMD CPUs is by booting the kernel with the nopti kernel command-line parameter. This can also be applied to Intel systems too on a patched kernel if wanting to regain the performance and are not too concerned about this vulnerability.
In affected benchmarks (those making use of a lot of system calls, context switches, etc), indeed AMD EPYC faces a performance penalty similar to Intel. I'll have more test data to share on Wednesday. Hopefully more details on the underlying vulnerability come to light soon to really know if AMD CPUs have any chance of being affected and other details.Update: Linux Will End Up Disabling x86 PTI For AMD Processors
| 70
| 1,760,719,663.217815
|
https://www.phoronix.com/news/Radeon-Whats-Left-2018
|
There's One Big Feature Left For The Radeon Linux Driver Left To Tackle In 2018
|
Michael Larabel
|
AMD/Radeon had a stellar 2017 for Linux most notably with delivering working Radeon RX Vega open-source driver support at launch, AMDGPU DC finally being merged to the mainline Linux kernel, and the official "AMDVLK" Vulkan driver now being open-source. Besides never-ending performance tuning, there's really just one major feature/area where the Radeon Linux graphics driver support is missing.
AMD managed to deliver two of My Three Hopes For AMD's Open-Source Stack The Rest Of 2017... They got DC merged and their Vulkan driver is now open. With features like FreeSync getting squared away and other minor work items, one of the last remaining big ticket items is getting Radeon Software Settings to Linux, basically their driver settings GUI.
Depending upon your use-cases this may or may not be needed for you, but there are many Radeon Linux gamers that would prefer a full-featured GUI configuration utility, its likely what Windows gamers transitioning to Linux would expect rather than learning command-line controls, and it's not as big of a hurdle as merging DC or the like. AMD previously offered a GUI driver panel (back then, known as the AMD Control Center Linux Edition) during the fglrx days that then disappeared during the transition to AMDGPU-PRO and AMDCCCLE was never supported for the open-source drivers.
The former AMDCCCLE. For the open-source AMD Linux driver has been a few different utilities over time from Radeon-Profile to the generic DriConf to a few other half-useful GUI tools and the desktop-specific monitor configuration modules but no single solution that offers tuning the full capabilities of the Radeon drivers, nothing officially supported by AMD, and not anything close to the quality of AMDCCCLE / Radeon Software Settings. Also nothing that comes close to the features and information provided by the NVIDIA Settings panel for Linux.
2016 version of Radeon Software Settings. It's been about two years since talking with AMD about having "Radeon Software Settings" for Linux. Back then they said they were investigating the possibility of open-sourcing their GUI control panel, which at the time had recently been rewritten in Qt5. They were optimistic about open-sourcing it in the future and in the mean time were working on exposing more of the driver's tunables and monitor-able information via kernel ioctls and other interfaces. Much of that information is now exposed nicely to user-space, but no Radeon Software Settings yet. During the past round of media briefings I inquired about any Linux update on the matter, but unfortunately those questions went unanswered.
I remain optimistic we could see something on this front in 2018 now that the more pressing issues like AMDGPU DC and AMDVLK availability have been addressed, but no official signs yet. Feel free to express your thoughts in the forums if you would like to see "Radeon Software Settings" made available for Linux.
| 40
| 1,760,719,663.713026
|
https://www.phoronix.com/news/AMDVLK-XGL-First-Code-Update
|
AMD Pushes Out Their First Post-Release Update Of AMDVLK/XGL
|
Michael Larabel
|
AMD developers working on the newly open-sourced AMDVLK Vulkan driver have pushed out their first post-release code update synced against the latest changes in their internal AMD driver tree.
The changes to the AMDVLK driver via the XGL repository that incorporates all of the changes made to their internal tree since the original open-source code drop.
Hopefully it will be improved upon in the future as for now there is no documented change-log of differences made with the revised code, looking at the changes is difficult with this particular commit as there is a lot of churn of code license changes (mostly capitalization changes), and still being introduced to the code-base are warnings at tops of files for "Trade secret of Advanced Micro Devices, Inc." But there is also some actual code changes that made its way in with this update.
You can checkout this commit for the updated driver code.
David Mao of AMD's driver team has also responded to David Airlie with roughly how they plan to develop this driver moving forward, but overall no breaking details compared to our original reporting on AMDVLK.
| 28
| 1,760,719,664.599184
|
https://www.phoronix.com/news/AMD-Radeon-Tops-2017
|
The Most-Viewed AMD/Radeon Linux Stories Of 2017
|
Michael Larabel
|
Here's a look at our most-viewed original AMD/Radeon Linux and open-source news stories of 2017.
Of the more than 300 original AMD/Radeon stories on Phoronix this year about their Linux and open-source support, below is a look at the top twenty. In marking their return to competitive CPU offerings with Zen, a majority of the most-viewed articles this year are about the AMD Ryzen CPU offerings rather than just the Radeon graphics that usually dominate the top Linux lists.
It was certainly a great year for AMD with the successful launch of the Ryzen desktop processors, EPYC being very competitive for servers, the Radeon Linux graphics driver stack maturing a great deal, AMDGPU DC being merged to mainline, and most recently the release of the AMDVLK Vulkan driver.
Let us know in the forums what you hope to see of AMD/Radeon in 2018 for Linux.
AMD's Ryzen Will Really Like A Newer Linux Kernel AMD's Ryzen CPU is finally shipping in a few days! If you are planning to be an early adopter of AMD Ryzen processors, you will really want to be running a newer Linux kernel release for proper support and performance. AMD Confirms Linux Performance Marginality Problem Affecting Some, Doesn't Affect Epyc / TR This morning I was on a call with AMD and they are now able to confirm they have reproduced the Ryzen "segmentation fault issue" and are working with affected customers. Some Ryzen Linux Users Are Facing Issues With Heavy Compilation Loads I haven't encountered this issue myself on any of my Ryzen Linux boxes, but it seems there are a number of Ryzen Linux users who are facing segmentation faults and sometimes crashes when running concurrent compilation loads on these Zen CPUs. Ryzen-Test & Stress-Run Make It Easy To Cause Segmentation Faults On Zen CPUs With running a number of new Ryzen Linux tests lately, a number of readers requested I take a fresh look at the reported Ryzen segmentation fault issues / bugs affecting a number of many Linux users. I did and still am able to reproduce the problem. Linux 4.11 Doesn't Change The Game For AMD's Ryzen Linux 4.11 is worthwhile in that it's bringing ALC1220 audio support, the codec used by many Ryzen (and Intel Kabylake) motherboards, but this next kernel version doesn't appear to change Ryzen's performance. AMD Reportedly Allows Disabling PSP Secure Processor With Latest AGESA With the latest AGESA update for Ryzen-based systems, AMD is reportedly allowing the Platform Security Processor (PSP) to be disabled. The AMD PSP akin to Intel's Management Engine. Google Kahlee: The First AMD-Powered Chromebook After years of many Intel and ARM Chromebooks, the first AMD-powered Chromebook appears to be gearing up for release. Extra AMD Ryzen 7 1800X Linux Benchmarks Assuming you have already checked out this morning's Ryzen 7 1800X Linux benchmarks, here are some more data points while putting the finishing touches on the Ryzen 7 Linux gaming benchmarks being published later today. AMDGPU-PRO 17.10 Released With Ubuntu 16.04.2 Support AMD has released the AMDGPU-PRO 17.10 hybrid Linux graphics driver. AMD Sends Out 100 Patches, Enabling Vega Support In AMDGPU DRM 100 patches amounting to over fourty thousand lines of code was sent out today for review in order to provide "Vega 10" support within the AMDGPU DRM driver. AGESA 1.0.0.6b Might Fix The Ryzen Linux Performance Marginality Problem Motherboard vendors have begun pushing out BIOS updates for Ryzen motherboards using the AMD AGESA 1.0.0.6b revision and it's reported that it does resolve the "Performance Marginality Problem" affecting early Ryzen Linux customers. Hammering The AMD Ryzen 7 1800X With An Intense, Threaded Workload Today I got around to running a very heavy/demanding, very real-world workload on the AMD Ryzen 7 1800X that I've been meaning to test with this Zen CPU. Linux 4.15 Is A Huge Update For Both AMD CPU & Radeon GPU Owners Linux 4.15 is shaping up to be a massive kernel release and we are just half-way through its merge window period. But for AMD Linux users especially, the 4.15 kernel release is going to be rocking. AMDGPU-PRO 16.60 Released AMDGPU-PRO 16.60 is now available as the latest version of the hybrid AMDGPU-based Radeon Linux graphics driver stack. AMDGPU DC Code Lands For Linux 4.15 Kernel Linus Torvalds has accepted the AMDGPU DC display code pull request for the Linux 4.15 kernel. AMD Linux users can now rejoice! How The Ryzen 7 1800X Compares To The Performance Of Systems By Phoronix Readers Yesterday on top of the main Ryzen 7 1800X Linux benchmarks and the follow-up Linux gaming benchmarks, I also posted some extra Ryzen benchmark results and encouraged Phoronix readers to compare their own system's performance to our data using our open-source, automated benchmarking framework. AMD Ryzen 7 1700 + B350 DDR4 Memory Speed Tests Earlier this week I posted some Ryzen 7 1800X DDR4 memory scaling Linux tests now that MSI pushed out an updated BIOS for that X370 motherboard that allows running the system at higher -- but still rather limited -- DDR4 memory frequencies. Here are some similar tests with my Ryzen 7 1700 and a B350 motherboard. AMD Begins Cutting Prices On Ryzen CPUs If you have been waiting to pick up an AMD Ryzen CPU until the prices drop, they are beginning to do so. AMD Releases Optimizing C/C++ Compiler For Ryzen Longtime Phoronix readers and AMD Linux enthusiasts probably remember the AMD Open64 compiler for past CPU launches with various compiler optimizations for AMD processors. With Open64 being dead and all the compiler rage these days about LLVM/Clang, AMD has announced the "AMD Optimizing C/C++ Compiler" (AOCC) that's based upon Clang and optimized for Ryzen/Zen processors. AMD Sends Out Prep AMDGPU Patches For New GPUs In the early hours of today AMD posted a set of 23 AMDGPU patches as "prep patches for new ASICs", which given the timing, is presumably prepping for the Radeon RX VEGA.
| 3
| 1,760,719,665.77301
|
https://www.phoronix.com/news/AMDKFD-More-For-Linux-4.16
|
AMD Queues More AMDKFD HSA Kernel Driver Changes For Linux 4.16
|
Michael Larabel
|
More AMDKFD changes are being queued for the upcoming Linux 4.16 kernel merge window with this being the kernel HSA driver for ROCm support, etc.
The big work ongoing is getting the discrete GPU support upstreamed so that the stock mainline Linux kernel could work with the user-space ROCm open-source packages for OpenCL support, etc. Unfortunately this latest AMDKFD pull request still doesn't have all the dGPU changes as it's still waiting on a patch for the PCI subsystem that introduces the needed PCI-E atomics support.
Among the work found in this pull request, however, though is a lot of other feature work particularly for AMD APUs. Carrizo with AMDKFD in Linux 4.16 will now support "compute wave save restore" CSWR and SDMA user-mode queues. Kaveri also will support SDMA user-mode queues too with Linux 4.16.
The AMD Kernel Fusion Driver update also now allows hardware scheduling to schedule multiple processes concurrently, debugfs is now supported, process locking and lock dependencies have been simplified, and refactoring topology code to prepare for the long-awaited dGPU support. There are also code clean-ups and other fixes.
More details on these AMDKFD updates for Linux 4.16 can be found via amd-gfx.
| 2
| 1,760,719,666.721607
|
https://www.phoronix.com/news/Endless-AMDGPU-DC
|
Endless Computer Is Looking Forward To Using AMDGPU DC
|
Michael Larabel
|
Endless Mobile, the company behind the Linux-based Flatpak-using Endless OS and that has sold several different low-cost computers around the world, is looking forward to AMDGPU DC.
The Endless developers are interested in AMDGPU DC primarily now for allowing HDMI audio to work on some of their computers using this open-source driver. AMDGPU DC, of course, needs no introduction around Phoronix unless you are well behind on your reading.
While the AMDGPU DC display code infrastructure is now present in the Linux 4.15 mainline kernel, their current testing has revealed some regressions. While HDMI audio is now working thanks to this massive new display code implementation, they found AMDGPU DC as what's found in mainline currently can cause system hangs when resuming from S3 suspend, display corruption when using multiple displays with DC, and the HDMI audio device still appearing as present when disconnecting the HDMI cable.
But fortunately the Endless team has provided some fairly thorough test results to the AMDGPU developers and they are optimistic the code is in better shape for Linux 4.16. They are hoping to help back-port some of those DC patches to the Linux 4.15 stable kernel in order to get regression-free, HDMI audio support sooner. Those wanting to check out Endless' AMDGPU DC test results can find them on dri-devel.
| 5
| 1,760,719,667.22393
|
https://www.phoronix.com/news/Radeon-GPU-Profiler-1.1-1.10
|
Radeon GPU Profiler Updated For Better Profiling Of Vulkan Games
|
Michael Larabel
|
Following yesterday's excitement around the Radeon Software Adrenalin Driver as well as word of AMD open-sourcing their Linux driver and making other Linux driver changes, AMD's GPUOpen team has announced the release of a new version of Radeon GPU Profiler.
Radeon GPU Profiler is the company's Windows/Linux application for GPU performance profiling of Direct3D 12 and Vulkan applications/games. It's advanced quite a bit but while it's hosted on GitHub, the source code currently still remains unavailable with the repository hosting just the documentation and then the releases are of the binaries.
The release announcement at GPUOpen.com refers to the new version as 1.1.0 while the GitHub release page also calls it version 1.10. Whatever version you want to call it, there are improvements in this feature update.
The updated Radeon GPU Profiler now has support for PIX3 user markers, a GPU-only view has been added to the profiler, various UI/usability improvements, actionable context rolls that show the state changes causing them, improvements to the Radeon Developer Panel, and other changes to help in profiling of DX12/Vulkan apps/games on AMD hardware.
Using the Radeon GPU Profiler on Linux can work with AMDGPU-PRO or the soon-to-be-opened Vulkan driver but the RADV driver doesn't currently have the necessary AMD-developed extensions for compatibility.
| 0
| 1,760,719,668.086439
|
https://www.phoronix.com/news/AMD-PSP-Disable-Option
|
AMD Reportedly Allows Disabling PSP Secure Processor With Latest AGESA
|
Michael Larabel
|
With the latest AGESA update for Ryzen-based systems, AMD is reportedly allowing the Platform Security Processor (PSP) to be disabled. The AMD PSP akin to Intel's Management Engine.
This built-in AMD Secure Processor has been criticized by some as another possible attack vector, closed-source software running on the system and locking it down, etc. PSP makes use of ARM TrustZone.
In light of the recent Intel ME vulnerabilities, AMD appears to be exposing the support in their latest AGESA update to allow the PSP to be disabled by the user through their UEFI/BIOS area.
This was brought up today on Reddit with some users reporting to see a "BIOS PSP Support - Disabled" option when updating their BIOS.Unfortunately, none of my Ryzen motherboards have seen vendor BIOS updates since September, so not yet able to confirm this feature on any of my motherboards.
| 70
| 1,760,719,669.011273
|
https://www.phoronix.com/news/Raven-Ridge-Firmware-Added
|
AMD Raven Ridge APU Firmware Added To Linux-Firmware.Git
|
Michael Larabel
|
For those who already picked up a Raven Ridge laptop or looking to when more of these Zen+Vega devices surface in the weeks ahead, the Raven Ridge firmware is now living within linux-firmware.git.
This will make it easier to deploy the accelerated graphics of Raven Ridge with now having the binary-only firmware blobs easily accessible via this firmware repository and in turn will get picked up by the various Linux distributions in time.
Ten files were introduced for the Raven Ridge firmware support.
Besides needing these firmware bits, to get Raven Ridge graphics working on Linux with the open-source stack you will need Mesa 17.3 (or ideally Mesa Git) built against LLVM 6.0 SVN as well as the Linux 4.15 kernel RCs or newer.
AMD today also sent into this firmware repository updated Vega VCE firmware for the video encode block.
| 2
| 1,760,719,669.488604
|
https://www.phoronix.com/news/AMD-2017-Driver-Name
|
AMD Announces The Radeon Software Adrenalin Driver
|
Michael Larabel
|
AMD's embargo has just expired over the name of their new driver.
This shouldn't come as a big surprise, but AMD has been pushing out big annual updates to their "Radeon Software" graphics driver the past few years. In December they will be shipping the successor to Radeon Software Crimson ReLive Edition.
Today they aren't allowing driver details to be shared or anything of that like, but this embargo expiry is just concerning the name: Radeon Software Adrenalin Edition. And this picture was also provided under the now-expired NDA:
Not to be confused with the adrenaline hormone, the Radeon Software Adrenalin Driver is said to be inspired by the Adrenalin Rose flower.
What does this mean for AMD Linux users? Likely not too much. The open-source Radeon Linux driver is obviously its own game and the AMDGPU-PRO driver is where the code sharing takes place with their Windows driver. So it will just be a matter of if there are any big OpenGL/Vulkan improvements to find in Radeon Software Adrenalin Edition that will then be shared to -PRO or if it's just about updating the Windows software settings area and other cosmetic work or Direct3D 12 optimizations. But not many are using AMDGPU-PRO now on Linux given the maturity of AMDGPU+RadeonSI with AMD recommending the hybrid driver for workstation customers and those wanting the official Vulkan driver.
With the last Radeon Software Crimson ReLive briefing they also used it to talk about the open-source driver performance enhancements, so perhaps they'll do that again for the press, especially with all the open-source Radeon 2017 driver milestones.
But at least the Radeon Software Adrenalin Edition briefing allows me to inquire to the right folks about the state of open-sourcing their official Vulkan driver and the state of bringing their Radeon Software Settings UI control panel to Linux among other topics... (Any other questions you'd like answered? Post your questions in the comments in the forums.) But for today it's just the Adrenalin driver name that's being announced.
| 38
| 1,760,719,670.569112
|
https://www.phoronix.com/news/RadeonSI-Huge-Cleanup
|
Marek Baking A "Huge Cleanup" For RadeonSI Gallium3D
|
Michael Larabel
|
Marek has volleyed onto the Mesa mailing list a set of patches providing RadeonSI with a "huge cleanup" for this Gallium3D OpenGL driver used by Radeon HD 7000 series "GCN" / "Southern Islands" graphics cards and newer.
At this point the cleanup consists of 24 patches mostly moving around code. This clean-up comes in result as a part of the decision in September of R600 Gallium3D To End Code Sharing With RadeonSI Driver.
There is also one minor optimization too: "radeonsi: don't use fast color clear for small surfaces - This removes 35+ clear eliminate passes from DOTA 2."
These patches for now can be found on Mesa-dev.
It will be interesting to see what else this prolific Radeon Mesa contributor is cooking up ahead of the holidays given the massive improvements RadeonSI has already made in 2017.
| 14
| 1,760,719,671.067362
|
https://www.phoronix.com/news/AMDKFD-Early-4.16-Patches
|
AMD Preps To Upstream More AMDKFD HSA Kernel Driver Changes
|
Michael Larabel
|
AMD has sent out 14 new patches today for the AMDKFD HSA kernel driver in material that should be targeting Linux 4.16.
As reported previously, they are working to upstream more of their AMDKFD changes currently living within their hybrid driver stack and not yet the mainline kernel tree. They've been trying to get more of their APU changes upstreamed followed by proper discrete GPU support. With Linux 4.15 they made some progress in Carrizo/Kaveri era changes.
The 14 patches put out today unfortunately don't bring any new hardware improvements, but there is some new functionality. The 14 patches today provide some fixes, allow HWS to schedule multiple processes concurrently, debugfs support, and simplified process locking and lock dependencies.
The good news is that AMD's Felix Kuehling mentioned in the patch series, "After these patches I'm ready to start upstreaming dGPU support." Here's to hoping ROCm OpenCL and friends begins playing nicely on mainline components for Radeon GPUs in 2018 to ease the open-source OpenCL adoption and usage.
| 3
| 1,760,719,672.464021
|
https://www.phoronix.com/news/AMDGPU-DC-Post-4.15-RC1-Fixes
|
More AMDGPU DC Fixes Are Being Queued For Linux 4.15
|
Michael Larabel
|
A few of you within the forums have talked of regressions and other bugs when trying out the new AMDGPU DC display stack in the Linux 4.15 kernel, particularly on pre-Vega GPUs where it's disabled by default. The good news is that more fixes are on the way.
Harry Wentland at AMD who has been doing much of the DC/DAL display code wrangling over the months sent out a set of 26 patches he's looking to have Alex Deucher pull into his -fixes branch for eventual integration into the mainline Linux 4.15 kernel.
Of these cherry-picked fixes for back-porting, they should fix DPMS (display power management signaling) and gamma on at least Polaris graphics cards. The other fixes also help the emerging AMD Raven Ridge APUs. There's also a potential fix for dealing with S3 suspend.
More details on amd-gfx.
My testing of Linux Git with AMDGPU DC continues going well, albeit mostly with RX Vega hardware so far. Those trying out Linux 4.15 and want to try the new display code with pre-Vega GPUs can boot the kernel with amdgpu.dc=1.
Some fresh Windows 10 vs. Linux Radeon gaming tests also coming up this week on Phoronix with Linux 4.15 and Mesa Git.
| 10
| 1,760,719,672.607132
|
https://www.phoronix.com/news/Compressonator-2.7-Released
|
AMD/GPUOpen Compressonator 2.7 Brings Linux Builds, glTF 2.0 Support
|
Michael Larabel
|
AMD's GPUOpen team has announced the release of Compressonator 2.7, the newest version of their tools for dealing with compressed assets and for testing the impact of different compression techniques.
Compressonator 2.7 finally brings official Linux builds, with offering the SDK and command line utilities for Linux. The Compressonator GUI currently isn't offered on Linux.
Compressonator's GUI application now supports loading glTF 2.0 3D models and rendering them with Direct3D 12. But they are working on OpenGL and Vulkan support, so hopefully at that time we will then see the Linux support.
Compressonator 2.7 also has support now for viewing the difference between 3D models and other enhancements. More details at GPUOpen.com and via the project's GitHub repository.
| 0
| 1,760,719,673.953651
|
https://www.phoronix.com/news/LLVM-6.0-3DNow-Improvements
|
LLVM Picks Up 3DNow! Improvements In 2017
|
Michael Larabel
|
As a flashback to the past, hitting the LLVM Git/SVN code today were improvements for those still running with processors supporting AMD's 3DNow! extensions.
3DNow! as a refresher for those not in the processor game too long were AMD's x86 instruction set addition for SIMD data support for vector processing. 3D Now had been introduced during the K6 days and supported until the release of the Bulldozer microarchitecture. 3DNow had been of some benefit during the early days of 3D computer gaming, but it's been a long time (years) since seeing any news about it.
Thus it came as a surprise this afternoon when seeing this commit to LLVM. An independent contributor to LLVM added the 3DNow instruction itinerary and scheduling class data to this compiler stack.
So if you're still running an AMD K6 up until Bulldozer CPU, with the upcoming LLVM 6.0 release you may find better performance. Unfortunately I no longer have any pre-Bulldozer systems still setup for benchmarking.
| 15
| 1,760,719,674.92965
|
https://www.phoronix.com/news/Raven-Ridge-APU-Not-HBM2
|
Report: Ryzen "Raven Ridge" APU Not Using HBM2 Memory
|
Michael Larabel
|
While the Radeon RX Vega discrete graphics cards are making use of the ultra-fast HBM2 memory, it appears the newly-launched AMD "Raven Ridge" APU featuring Zen CPU cores and Vega graphics is not using HBM2 memory.
Instead of the Vega graphics on Raven Ridge using HBM2 memory, it appears at least for some models they are just using onboard DDR4 memory. FUDZilla is reporting today that there is just 256MB of onboard DDR4 memory being used by the new APU, at least for the Ryzen 5 APU found on the HP Envy x360 that was the first Raven APU system to market.
The Radeon Vega mobile graphics performance is a huge letdown in their initial tests shared. In fact, this APU was "barely beating NVIDIA's 840m GPU."
This is a big letdown if all Raven Ridge APUs indeed are just using DDR4 memory onboard rather than HBM2. The Envy x360 is also a bit of an odd-ball in just using a single memory channel, odd for being the launch laptop of Raven Ridge.
Also making it a bit more of a sticky situation for Linux users is the Raven Ridge APU with its new "DCN" display engine is still seeing a lot of code churn for AMDGPU DC and not all of the DCN 1.0 code merged for Linux 4.15. Thus the Raven display support might not even be in good shape until Linux 4.16.
Hopefully we'll learn more soon.
| 64
| 1,760,719,675.371184
|
https://www.phoronix.com/news/AMD-EPYC-Linux-4.15-First-Test
|
AMD EPYC Is Running Well On Linux 4.15
|
Michael Larabel
|
Of the many changes coming for Linux 4.15, as detailed this weekend Radeon GPU and AMD CPU customers have a lot to be thankful for with this new kernel update currently in development. Here are some initial benchmarks of the Linux 4.15 development kernel using an AMD EPYC 7601 32-core / 64-thread setup.
When it comes to EPYC in Linux 4.15, the kernel side-bits have landed for Secure Encrypted Virtualization (SEV), CPU temperature monitoring support now working, and improved NUMA node balancing.
With the AMD-specific improvements as well as general enhancements that have landed so far during the Linux 4.15 kernel, the AMD EPYC 7601 is working out well so far on the Tyan 2U EPYC test server.
I ran some quick comparison tests today between Linux 4.14 stable and the current Linux 4.15 Git code...
There are some minor to modest performance improvements to note with 4.15 on this AMD server.
More results via this OpenBenchmarking.org result file.
The AMD EPYC CPU temperature monitoring is also working with Linux 4.15 thanks to the Family 17h support added to the k10temp kernel driver.
The first release candidate of Linux 4.15 should be out next weekend while the stable Linux 4.15 should be out around mid-January and is the kernel slated for use in Ubuntu 18.04 LTS. Stay tuned for more AMD EPYC Linux benchmarks and 4.15 kernel tests in general at Phoronix.
| 13
| 1,760,719,676.443316
|
https://www.phoronix.com/news/Ryzen-Threadripper-Price-Drop
|
Ryzen/Threadripper Prices Have Been Dropping Ahead Of The Holidays
|
Michael Larabel
|
If you have been wanting to build a new system before the end of the year, AMD Ryzen CPU prices -- including the high-end Threadripper -- have been dropping in recent days in at least the US and EU.
The top-end Threadripper 1950X is no longer at $999 USD but currently just $800 for its 16 core / 32 thread configuration at 4.0+GHz and 40MB of cache and quad-channel DDR4 memory support.
Meanwhile, the Threadripper 1900X in its 8 core / 16 thread configuration is down to $500 USD.
Ryzen desktop prices are also trending lower with the Ryzen 5 1600 for $190, Ryzen 7 1700 at $270, or the flagship Ryzen 7 1800X for $320. The Ryzen 7 1800X originally launched a few months back at $499 USD.
These lower prices can currently be found at Amazon (including in the EU) and NewEgg. And, yes, if you make your Internet purchases using our referral links the sales go on to support our site operations.
AMD hasn't communicated yet if these price drops are just temporary for the holidays or are a permanent reduction.
| 44
| 1,760,719,676.83073
|
https://www.phoronix.com/news/Linux-4.15-AMD-Mega
|
Linux 4.15 Is A Huge Update For Both AMD CPU & Radeon GPU Owners
|
Michael Larabel
|
Linux 4.15 is shaping up to be a massive kernel release and we are just half-way through its merge window period. But for AMD Linux users especially, the 4.15 kernel release is going to be rocking.
Whether you are using AMD processors and/or AMD Radeon graphics cards, Linux 4.15 is a terrific way to end of the year. There are a number of improvements to make this release great for AMD customers.
First and foremost, AMDGPU DC was merged Friday night. This addition of over 130,000 lines of code to the kernel provides Radeon RX Vega display support, the initial Raven Ridge APU display support, HDMI/DP audio for recent generations of Radeon hardware, atomic mode-setting, initial FreeSync prep support, and other modern display features that takes it close to parity with the Radeon Software Windows driver. It was a long time coming and finally happened for Linux 4.15. See my dozens of AMDGPU DC articles for more on the topic.
In addition to AMDGPU DC display code landing, there is AMDGPU priority scheduling support as developed by Valve for ensuring Steam VR on Linux can run at a higher priority to reduce the risk of slowdowns causing motion sickness and the like. This priority scheduler was in development for a year and finally landed. For the RADV Vulkan driver this support is exposed via VK_EXT_global_priority extension in Git for Mesa 17.3. I hope to find the time in the days ahead to deliver some fresh Steam VR Linux testing with the HTC Vive on different graphics cards / drivers.
Other AMDGPU DRM work that's found in Linux 4.15 includes GPU reset support for RX Vega graphics cards in the event of hangs or other issues, GCN 1.1 Sea Islands PowerPlay support, TTM memory management updates, PRIME mmap support, more prep work for Raven Ridge APUs, and other improvements.
There's also much to be thankful for if you are running a new AMD CPU... Benefiting all AMD Zen CPUs whether it be Ryzen, Threadripper, or EPYC, there is finally AMD Zen CPU temperature monitoring support that's extended within the k10temp hwmon driver. It's silly it has taken so long, but with Linux 4.15 you can now read the CPU package temperatures of these new AMD processors under Linux.
The Linux 4.15 PCI improvements include 64-bit BAR support for AMD Family 15h (Bulldozer) processors.
Exciting for the EPYC server processors after Secure Memory Encryption (SME) landed in Linux 4.14 is taking it further now with SEV. Secure Encrypted Virtualization landed in Linux 4.15 for allowing guest VMs to have their memory encrypted by the hardware whereby only the guest VM can view the unencrypted data -- protecting against any rogue processes on the host trying to view a VM's memory, a fellow VM trying to exploit memory access to another VM, etc.
Also benefiting AMD EPYC customers with Linux 4.15 is improved scheduler behavior for EPYC with regard to better balancing of NUMA nodes.
The merge window for Linux 4.15 will end next weekend (26 November) while the official stable release of it should be out ~8 weeks later; roughly putting the stable Linux 4.15 release around the middle of January. Linux 4.15 is what's tentatively slated as the default kernel for Ubuntu 18.04 LTS, among other distribution releases in H1'2018. But, of course, nothing is stopping you from building this kernel yourself in its latest daily/Git state if you want these features now.
If I have missed any other interesting AMD/Radeon changes for Linux 4.15 so far, feel free to point out in the forums. Of course, our monitoring of Linux 4.15's development in general will continue as well as our benchmarking of this kernel in the days ahead.
As always, if you appreciate all of my original reporting, benchmarking, and Linux hardware reviews, please consider showing your support by joining Phoronix Premium to enjoy the site ad-free, multi-page articles on a single page, and other benefits. Alternatively PayPal tips are also supported or at the very least to please not view this website with any ad-blocker.
| 37
| 1,760,719,678.000239
|
https://www.phoronix.com/news/AMD-Stoney-Linux-4.15-Audio
|
AMD Stoney Ridge Audio Supported By Linux 4.15
|
Michael Larabel
|
The sound driver changes have been submitted for the Linux 4.15 kernel and includes finally supporting AMD Stoney Ridge hardware.
Takashi Iwai of SUSE today sent in the sound updates for the Linux 4.15 kernel window. The noteworthy mentions are a new AC97 bus implementation and AMD Stoney platform support. There was also some hardening work of USB audio drivers, cleanups to the Intel ASoC platform code, and a variety of other low-level changes.
AMD Stoney Ridge audio platform support is finally enabled with Linux 4.15. Stoney Ridge was AMD's 2016 APU architecture with Excavator CPU cores and Volcanic Islands class graphics. These APUs were a step above Carrizo and included the A6/A9-9200/9400 line-up as well as the E2-9000 series. These Stoney audio patches have been floating around on the mailing lists for a while now but with Linux 4.15 appear to be all wired up. I haven't yet seen any sound patches regarding Raven Ridge or if all of that will be handled via AMDGPU DC.
Details on all the sound subsystem changes for Linux 4.15 via this pull request.
| 5
| 1,760,719,678.537705
|
https://www.phoronix.com/news/AMD-ROCm-1.7-Released
|
AMD Rolls Out ROCm 1.7 Platform For Supercomputing 17
|
Michael Larabel
|
AMD has unveiled the Radeon Open eCosystem platform (ROCm) 1.7 release as part of their wares at this week's Supercomputing 17 (SC17) conference in Denver.
The ROCm 1.7 update introduces multi-GPU support for "the latest Radeon GPU hardware" (presumably referring to Vega) while also supporting TensorFlow and Caffe via AMD's MIOpen libraries.
It also looks like there may be some more performance optimizations in ROCm 1.7, but their press email is light on technical details.
As of writing, the ROCm documentation hasn't been updated for ROCm 1.7 but that should shed some more light on this new release when it surfaces, probably later in the day. If the changes look interesting enough, I'll be working on some fresh ROCm OpenCL benchmarks. For now at least through Linux 4.15, ROCm continues to depend on kernel changes not yet mainlined and in user-space against LLVM/Clang/LLD.
AMD is also using the SC17 conference for announcing broader availability of AMD EPYC systems by a variety of companies.
| 44
| 1,760,719,679.40352
|
https://www.phoronix.com/news/AMD-Zen-Temperature-Hwmon-4.15
|
AMD Zen Temperature Monitoring Queued For Linux 4.15
|
Michael Larabel
|
We've been expecting it to happen for weeks while indeed the hwmon pull request was indeed sent in today exposing AMD Ryzen / Threadripper / EPYC temperature reporting on Linux.
The patch to the existing k10temp Linux hwmon driver has been floating around since September for AMD Zen / Family 17h temperature reporting finally being in place. It was staged in hwmon-next and is now called for pulling into the just-opened Linux 4.15 merge window.
Guenter Roeck's hwmon 4.15 updates includes this long-awaited addition along with some other support additions and various clean-ups to this hardware monitoring kernel code. It's unfortunate to come only so many months after the original "Zen" processor debut, but great to finally see it in place for those concerned about CPU core temperature reporting under Linux.
| 25
| 1,760,719,680.036358
|
https://www.phoronix.com/news/Raja-Koduri-Intel
|
AMD/RTG's Raja Koduri Joins Intel
|
Michael Larabel
|
On Monday Intel announced their upcoming CPU with integrated AMD Vega-class graphics backed by HBM2 memory, on Tuesday Radeon Technologies Group (RTG) head Raja Koduri announced he would be resigning from the company, and now today it's announced he is joining Intel.
Raja Koduri the former head of RTG who recently went on sabbatical following the Vega launch and ATI and Apple veteran will now be joining Intel.
Raja will be joining Intel as a chief architect and senior vice president of their new Core and Visual Computing Group. This is quite interesting given his involvement with Vega and next year's AMD Navi architecture design he was heavily involved in albeit now parting ways with AMD.
It will certainly be interesting to see what this highly skilled engineer achieves at Intel and what their next-gen graphics solutions will look like whether continuing to pursue AMD graphics on Intel CPUs or new, original designs. Next year's Cannonlake processors will feature Intel "Gen 10" graphics capabilities. It would certainly be interesting if Intel returns to working on a Larrabee-like architecture.
More details at Intel.com.
| 36
| 1,760,719,680.827446
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.