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/Radeon-GPU-Analyzer
|
Radeon GPU Analyzer Open-Sourced: Analyze OpenGL / Vulkan / OpenCL
|
Michael Larabel
|
In addition to AMD having open-sourced their UMR debugger a few days back, over in their "GPU Open" team they open-sourced the Radeon GPU Analyzer.
Radeon GPU Analyzer is GPUOpen's latest tool for helping game/application developers optimize their shaders for consumption on AMD APUs/GPUs. They describe the Radeon GPU Analyzer (RGA) as:
RGA CLI is an offline compiler and a performance analysis tool for DirectX shaders, OpenGL shaders, Vulkan shaders and OpenCL kernels. Using this product, you can compile source code for a variety of AMD GPUs and APUs, independent from the GPU/APU that is physically installed on your system, and generate AMD ISA, intermediate language and performance statistics for each target platform.
RGA CLI replaces CodeXL Analyzer CLI, and maintains backward compatibility. All of the commands and functionality that used to be supported by CodeXL Analyzer CLI are still being supported by RGA. It's great to see the OpenGL and Vulkan shader support as well as support for OpenCL kernels.
Windows is supported but also Linux with Ubuntu 14.04 and Red Hat Enterprise Linux 6.4 being listed as supported platforms. However, right now it appears this open-source tool will only work with their binary-only driver. There is a note under the system requirements, "The Radeon Software Crimson Edition or AMD Catalyst release must be installed to run this tool." Hopefully we'll be able to see this tool supported under the fully-free AMDGPU+RadeonSI driver stack and not be dependent upon AMD's binary-only user-space blobs.
Those wishing to learn more about the Radeon GPU Analyzer can visit the GitHub project site.
| 5
| 1,760,719,759.556619
|
https://www.phoronix.com/news/AMDGPU-UMR-Debugger-Opens
|
AMDGPU "UMR" Debugger Open-Sourced
|
Michael Larabel
|
AMD has just announced the release of their awaited AMDGPU open-source debugger.
This AMDGPU Debugger is initially a user-mode register debugger, which allows privileged users to read/write to GPU registers for diagnosing and debugging. The tool also supports decoding ring contents, analyzing wave fronts, viewing machine states, and other functionality. The AMDGPU Debugger works with Southern Islands through Volcanic Islands hardware currently and requires the Linux 4.10 kernel.
We've known AMD was working on a new open-source GPU debugger for Linux after a Valve developer announced his own AMD GPU debugging utility. This weekend this "AMDGPU debugger" is now public. The debugger is called UMR, short for the User-Mode Register debugger.
The AMDGPU open-source debugger is under the MIT license. They started developing it last year and is now cleaned up and public. Some features will work when using it on Linux 4.9, but more functionality is present for Linux 4.10 and then 4.11.
More details via this announcement and the code is available via FreeDesktop.org.
| 4
| 1,760,719,759.745
|
https://www.phoronix.com/news/AMDGPU-PRO-16.60
|
AMDGPU-PRO 16.60 Released
|
Michael Larabel
|
AMDGPU-PRO 16.60 is now available as the latest version of the hybrid AMDGPU-based Radeon Linux graphics driver stack.
Listed as the only highlight of the AMDGPU-PRO 16.60 driver is CentOS 7.3 support. Fixed issues are taking care of a rendering error in glxgears, launching the Steam client sometimes causing system hangs, and hard-hangs sometimes occurring when display hot-plugging.
That's it as far as the official changes go that are listed at AMD.com. Hopefully it's working with more modern kernels, but I haven't yet fired it up yet but there still might not be newer kernel support considering Ubuntu 16.10 still isn't listed as an option.
The operating systems listed with support are RHEL/CentOS 7.3, RHEL/CentOS 6.8, Ubuntu 16.04, and SLED/SLES 12 SP2. Will be firing up some Linux benchmarks on AMDGPU-PRO 16.60 momentarily to see if there's any other interesting changes.
| 121
| 1,760,719,760.936617
|
https://www.phoronix.com/news/AMD-HSA-BRIG-Still-Trying-GCC7
|
AMD HSA IL / BRIG Front-End Still Hoping To Get Into GCC 7
|
Michael Larabel
|
For many months now there's been work on an AMD HSA IL front-end for GCC with supporting the BRIG binary form of the Heterogeneous System Architecture Intermediate Language (HSA IL). It's getting late into GCC 7 development and onwards to its final development stage while this new front-end has yet to be merged.
Developer Pekka Jääskeläinen has been trying to get in the finishing reviews and changes for getting approval to land this BRIG front-end into the GNU Compiler Collection. It's a big addition and with GCC 7 soon just focusing on wrong-code fixes, bug fixes, and documentation fixes starting on 19 January, there would be just a few days left to land this new front-end for GCC 7 to avoid having to wait until next year for it to debut in stable with GCC 8.
Pekka Jääskeläinen who has been working on this BRIG code has been active in this mailing list thread sending out pings, making a few revisions to the patches, and addressing other questions with Richard Biener doing some of the review work. But as of this morning it's still not clear whether this front-end will be pulled into GCC mainline or not.
With already waiting years for the Heterogeneous System Architecture to really become a reality on the Linux desktop, hopefully we'll see this front-end merged in time for GCC 7 along with AMD's other open-source HSA/OpenCL/GPGPU efforts pan out in the months ahead.
| 0
| 1,760,719,761.345665
|
https://www.phoronix.com/news/No-Talos-Now-AMD-Libre-Effort
|
The POWER8 Libre System Looks Set To Fail, Now There's An AMD Libre System Effort
|
Michael Larabel
|
It doesn't look like the Talos Secure Workstation will see the light of day with it's crowdfunding campaign ending this week and it's coming up more than three million dollars short of its financing goal. Now there's another effort to offer a libre system but using off-the-shelf x86 hardware.
Even though the campaign was extended for the Talos Secure Workstation, to date this POWER8, ATX-compatible, fully-open mainboard/system only raised $457,490 USD of its $3.7 million dollar goal. There are five days left to the campaign, but it's next to possible for any crowdfunding campaign to raise three million dollars with less than one week to go in its campaign.
Libreboot developer Leah Rowe is now launching a libre system out of the ashes of the Talos Secure Workstation. She wrote in an email to Phoronix, "It's a high-end desktop/server platform, available in either configuration. It also supports virtualization and PCI passthrough, unlike older systems, so Qubes would be compatible...TALOS looks set to fail. Crowd Supply has removed it from their homepage, and Raptor Engineering is writing up an announcement that TALOS is shutting down - they are going to link to Minifree and tell people to purchase Libreboot D16 from me."
But before getting too excited, this isn't a new platform but rather an existing AMD server motherboard that simply comes pre-loaded with Libreboot to free the firmware/BIOS and then loaded with Debian GNU/Linux. The desktop and server versions make use of an AMD Opteron 6272, a.k.a. the older 32nm "Interlagos" CPUs derived from Bulldozer and released back in 2011.
The mentioned motherboard(s) planned for the desktop and server systems aren't explicitly mentioned, but based upon the supported boards and the name being "D16", it's pretty safe to assume it's the ASUS KGPE-D16. You can still find this motherboard via NewEgg.com for $415 USD and $488 from Amazon.
It's tough to find these Opteron 6272 CPUs still in stock but they appear to go used on eBay for about $25 USD. Newer AMD or Intel CPUs/motherboards aren't supported by Libreboot due to the hardware depending upon binary-only firmware.
As for the graphics cards in their "high-end desktop", they are planning a "NVidia GTX 660Ti or 670." Going with a GeForce GTX 600/700 series is understandable given that it's the last generation of NVIDIA GPUs where Nouveau is able to support it without relying upon closed-source firmware from NVIDIA. Nouveau self-generates the needed microcode and so you have 100% open-source driver support. Even on the AMD side they have been relying upon closed-source firmware/microcode for many generations now. So the NVIDIA Kepler usage is nice, but I wouldn't consider the GTX 660 Ti or GTX 670 to be worthy of a "high-end" system at all in 2017, especially in a system that costs ~$4k+ USD. As you can see from my recent Nouveau benchmarks, from Kepler for a high-end system I would only consider the GTX 680 or GTX 780 Ti if wanting a fully libre system.
As with other niche free/libre hardware projects, the Libreboot D16 setup isn't cheap at all even with using off-the-shelf components. The Libreboot D16 desktop with the D16 motherboard, one Opteron 6272, 16GB of RAM, and the NVIDIA GTX 660 Ti / 670 is marked up at £3,450.00 (~$4188 USD) while the D16 server with 1 CPU and 16GB of RAM is £3,450.00 or up to £5,600.00 ($6800 USD) if wanting two of the Interlagos CPUs and 128GB of RAM. It's a tough purchase price surely for most Phoronix readers, especially given it's based upon ~5 year old hardware. If wanting to build the system yourself and flash Libreboot on the same motherboard, you can likely build the system for around $1k USD or less: $400~500 for the KGPE-D16 at the aforementioned links, the CPUs for $16+, a couple hundred or less depending upon how much DDR3 RDIMMs you want, etc. But if you have the money to spare and can't be bothered to try flashing Libreboot on the board and installing Debian, you can find out more details via the Libreboot D16 product page.
| 43
| 1,760,719,762.361916
|
https://www.phoronix.com/news/Vega-Architecture-Details
|
AMD Reveals More Vega GPU Architecture Details
|
Michael Larabel
|
AMD isn't using CES 2017 to launch their Ryzen (Zen) processors or Vega graphics cards, but at least they have opened up more Vega architecture details for this busy week in Las Vegas.
AMD's embargo on the Vega architecture expired this morning. Unfortunately didn't get much information in advance and thus just a short article this morning. Some of the highlights from the Vega architecture details sent to the press include:
- HBM2 backs the "High Bandwidth Cache" (HBC; what is basically the vRAM) and delivers 256GB/s per stack, twice that of HBM1.
- Vega comes with a Draw-Stream Binning Rasterizer as similar to a tile-based approach to rendering, leading to more efficient shading of pixels.
- The GPU has 512 TB of virtual address space.
- Vega will be able to interface with NAND Flash or 3D X-Point SSDs over PCI-E.
- Render back-ends are now clients of the L2 cache.
- Significantly improved compute units.
- Rapid Packed Math is a new feature for clumping multiple 16-bit operations between 32-bit registers.
- Vega when running DOOM on Windows is estimated to perform between a GTX 1070 and GTX 1080.
Vega is still on track for release later in H1'2017. The many changes with the Vega GPU architecture are exciting as finally having a big step forward since GCN, though all of the changes due raise some concerns. Vega is launching within a few months while AMD has yet to publish any of their AMDGPU or RadeonSI Gallium3D changes for bringing up Vega. Hopefully they have a lot of it queued up internally and just waiting on legal review / clearance to publish; just really hoping it won't be a bumpy road for Vega on Linux when the launch happens. Given no code is out yet, you'll almost certainly need to use third-party repos and unofficial kernels if you aren't on a rolling-release distribution, but hopefully the code will be in good shape and that AMDGPU-PRO will also be in good standing similar to NVIDIA's binary Linux driver on launch days. Of course, once Vega GPUs are available there will definitely be Linux benchmarks on Phoronix.Ve.ga is also now the promotional site for the new architecture.
| 28
| 1,760,719,762.731146
|
https://www.phoronix.com/news/AMD-Zen-Ryzen17h-Patches
|
Linux 4.10 Is Hopefully In Good Shape For AMD Zen / Ryzen Processors
|
Michael Larabel
|
AMD's upcoming Ryzen (Zen) processors appear to be in good enough shape that they are working on the current mainline kernel as far as I can tell based upon limited information available prior to getting my hands on the CPUs or getting any official announcement from AMD, but some Linux kernel patches have yet to be mainlined. The yet-to-be-merged work appears to be more for non-core features and Zen server functionality with those CPUs shipping later than the upcoming Ryzen desktop CPUs.
As covered in the Linux 4.10 kernel feature overview. the 4.10 cycle is bringing some Zen/Ryzen/17h kernel patches. But still on the kernel mailing list there still are yet-to-be-merged patches mentioning AMD Zeppelin / Family 17h / Zen.
Among the work still baking appears to be multi-IOMMU support with AMD Zen servers will be having multiple IOMMUs and this ~200 lines of code enables said support. The code is now up to its sixth revision but wasn't ready for Linux 4.10. There is also support for perf on 17h with the patch for perf core PMU support on Family 17h CPUs didn't land for 4.10. But again, non-critical functionality for most users. And then just some other small, random patches.
Unfortunately a bit difficult to get a firm understanding if any critical patches are missing due to the less than ideal searching of Patchwork and LKML with relying upon Google and various Zen / 17h queries. But at least with the code in the kernel now it does appear that Zen / Ryzen CPUs should be working, but we should be able to find out in hopefully a few weeks if that really holds up once we're able to test this big AMD CPU advancement and see how well they work out-of-the-box on different released distributions.
| 9
| 1,760,719,763.734979
|
https://www.phoronix.com/news/AMDGPU-Polaris-12
|
AMD Publishes "Polaris 12" Support For AMDGPU+RadeonSI Linux Driver
|
Michael Larabel
|
Patches were published today for supporting "Polaris 12" graphics cards within the AMDGPU DRM kernel driver on Linux.
Polaris 12 just came to light earlier this month when some people discovered references to it in a macOS driver update. Polaris 12 is not the forthcoming high-end Vega hardware. But Polaris 12 details beyond that are rather light.
Now AMD developer Alex Deucher has published patches by AMD's Junwei Zhang that implement Polaris 12 support. Supporting Polaris 12 is just shy of 100 lines of code needed on top of the current AMDGPU DRM driver.
The series was published moments ago. Most of the ~100 lines are just for adding new "Polaris12" cases to various areas that are shared with Polaris 10/11 and then just different firmware file names. It's rather all straight-forward changes in digging through the patches with no breakthrough details, due to using all of the common Polaris 10/11 code-paths. The only new real information are the PCI IDs with there being six listed for Polaris 12: 0x6980, 0x6981, 0x6985, 0x6986, 0x6987, and 0x699F.
While writing this article, the RadeonSI Gallium3D patches for Polaris 12 were just published and again are non-exciting but basically adding the PCI IDs and the new Polaris 12 class. Should be interesting to see what Polaris 12 is about in 2017. Given the patches aren't invasive, it's not clear yet if AMD will try getting them into Linux 4.10 or if this material will wait until Linux 4.11.
| 18
| 1,760,719,764.055053
|
https://www.phoronix.com/news/ROCm-1.4-Available
|
AMD's ROCm 1.4 Now Available With OpenCL Support
|
Michael Larabel
|
The Radeon Open eCosystem platform has been updated and quietly released prior to the weekend. The ROCm 1.4 release comes with preliminary OpenCL support.
We've been looking forward to ROCm with OpenCL while now it's available in preliminary form with ROCm 1.4. ROCm 1.4 supports OpenCL 2.0 compatible kernel language support with OpenCL 1.2 compatible runtime. Details here.
But keep in mind it's still very early, the range of hardware support is quite limited, etc. Nevertheless, I'll be trying out ROCm 1.4 to see if it's in shape for running my OpenCL benchmarks. Unfortunately the complete v1.4 release notes do not appear to be available yet. Those wanting to try out the official Debian/RPM packages can find the details here.
| 22
| 1,760,719,765.136867
|
https://www.phoronix.com/news/R9-290-Linux-4.9-Update
|
The Strange Behavior Of My Radeon R9 290 Is Still There
|
Michael Larabel
|
In recent days there have been a few Phoronix readers inquiring why I am not testing with my Radeon R9 290 graphics card in all our frequent comparisons and driver benchmarks. The short story is that the regression since Linux 4.7 remains and for my Radeon R9 290 and others with select Hawaii graphics cards, there still is a performance regression. Though over Christmas I hope to finally find the time to bisect it.
So for those wondering but haven't asked why the R9 290 hasn't been used, it's since there is still that pesky regression... While there was a fix for some, my HIS Radeon R9 290 and that of other select users still are having issues, likely due to differing video BIOS. AMD, meanwhile, reportedly hasn't been able to reproduce this issue with their hardware.
It's been a rather strange issue but admittedly low priority on my end when busy with testing other (and newer) graphics cards, and having a never-ending TODO list of more pressing work to pay the bills. The Phoronix Test Suite should be able to bisect it automatically, at least this gives me the time to make sure those bisecting code-paths are still in good shape, but it still takes lots of system time and unfortunately few systems with easy access to open from all of them being rackmounted.
Affected cards see very low performance, such as these latest results with Linux 4.9 and Mesa 13.1-dev still being affected:
Basically if your card is affected, the performance is unplayable. I had hoped to find the time to run the bisecting tests over Thanksgiving, but that didn't happen. My hope is now that I'll have extra time once the ad revenues drop off around Christmas to sit back, have PTS bisect it, while enjoying some Augustiner to pass the time.
When running these tests over night to confirm the regression was still in place, some workloads like OpenArena were not affected:
But what's rather strange is that after either a certain operation with GpuTest, length of time, or other internal change, the Radeon R9 290 was back to running fast for the remainder of the automated OpenGL benchmarks:
The R9 290 was performing better in the remainder of the tests. What also makes this issue strange is that the Radeon R9 290 continues working fine with the AMDGPU-PRO stack.
Anyhow, hopefully I'll find the extra time over Christmas to be able to bisect this longstanding issue.
Those wanting to dig through this data further can see this OpenBenchmarking.org result file for the metrics as of last night.
| 24
| 1,760,719,766.587481
|
https://www.phoronix.com/news/AMDGPU-Virtualization-Patches
|
AMD MxGPU Virtualization For The AMDGPU Driver
|
Michael Larabel
|
Well this weekend is exciting for AMDGPU users and open-source AMD fans. Yesterday was the news we published about Valve looking to improve AMDGPU/RADV for their Vulkan-based VR experience while the latest is work from AMD that implements GPU virtualization support within the AMDGPU driver.
NVIDIA has been working on their own vGPU support for their binary driver, Intel's GVT is coming together in mainline, and now AMD has joined the party with publishing some patches for their DRM driver.
The patches by AMD's Xiangliang Yu work to implement CSA and KIQ along with mailbox communication with the GPU hypervisor. CSA is the Context Save Area. KIQ in this context is the Kernel Interface Queue, as described in one of the patches, "KIQ is queue-memory based initialization method: setup KIQ queue firstly, then send command to KIQ to setup other queues, without accessing registers. For virtualization, need KIQ to access virtual function registers when running on guest mode."
This AMD GPU virtualization support is initially implemented for Tonga and Fiji hardware. This AMD virtualization code is for implementing their virtualization support they advertise under the MxGPU branding. AMD announced their first hardware-virtualized GPU products at the beginning of 2016 while finally we are seeing this MxGPU support come for AMDGPU, but too late for Linux 4.10. Their MxGPU virtualization is described in more detail via this press release.
The code is initially going to be disabled at compile-time via hiding behind the DRM_AMD_MXGPU Kconfig switch. The DRM_AMD_MXGPU description, "This adds AMD GPU virtualization driver and wires it up into the amdgpu drivers. User can load the driver in guest OS and run graphics applications on AMD hardware in guest mode." The 23 patches add over two thousand lines of code to the kernel and can be found for now via amd-gfx.
| 19
| 1,760,719,766.849051
|
https://www.phoronix.com/news/AMD-Ryzen-New-Horizon
|
AMD Reveals More Zen CPU Details, Officially Known As Ryzen, No Linux Details Yet
|
Michael Larabel
|
AMD's Zen New Horizon event is going on right now. For those missing out on the livestream, here are my live details so far on Zen, or now officially known as Ryzen.
Here are my highlights, check back for updates.
- They have "met or exceeded" their goals at AMD for Zen. Beating performance goals of 40%+ instruction per clock improvement.
- AMD's Zen desktop "Summit Ridge" processors are indeed named Ryzen.
- AMD Ryzen processors will be out in Q1'2017. The Ryzen CPUs will have 8 cores, 16 threads via SMT, and the highest-end Ryzen Summit Ridge CPU will have a 3.4GHz clock speed or above. There will be a 4MB L2 cache and 16MB L3 cache. Boost clock speeds are not yet revealed for these AM4 CPUs.
- AMD Zen server processors will be out in Q2'2017 while the notebook APU (Raven Ridge) isn't coming until the second half of the year.
- AMD claims the 3.4+ GHz Ryzen can run slightly faster than a Core i7 6900K while having lower power consumption (95 Watt TDP of the Ryzen part shown during the demo).
- AMD SenseMI technology includes "true machine intelligence" and smart prefetch with Ryzen. There are also "Precision Boost" and "Pure Power" features. (I would anticipate there would be some OS integration needed, haven't seen any Linux kernel patches in this area yet for Zen.)
No mention yet of Linux during the presentation. But at least they are using some open-source apps for demonstration like Handbrake and Blender, albeit from Windows 10. While we have seen some more Zen patches queued up for Linux 4.10, we haven't yet seen any patches yet about the "SenseMI" features for Linux (particularly concerned about boosting and the power-savings functionality under Linux given no patches there yet that I've seen), working to get some comment from AMD about that for getting a better idea for the Linux support of Ryzen at launch come next quarter.Update: The event is over. No mention of Linux, but Ryzen looks super exciting anyhow, assuming it will play nicely on Linux. Stay tuned for AMD Ryzen benchmarks on Linux at launch in Q1.
| 92
| 1,760,719,768.176559
|
https://www.phoronix.com/news/AMD-Zen-Event-Today
|
AMD's Zen Livestream Event Is Today, More Zen Code Lines Up For Linux 4.10
|
Michael Larabel
|
For those interested in the upcoming Zen processors, a quick reminder that later today is AMD's livestream event where they will be giving a "sneak preview" of the upcoming Zen CPU.
Head on over to AMD.com for the "New Horizon" livestream event taking place at 3PM CST / 21.00 UTC. AMD will be showing off some (Windows) gaming with it as well as allowing the public to try a Zen system for the first time, along with some giveaways and "special guests."
Don't worry if you aren't able to watch it live, I'll post updates on Phoronix if any interesting details are revealed.
Meanwhile, for Linux users, more AMD Zen code is getting lined up for Linux 4.10. Yesterday I commented on some pull requests referencing Zen support and more PRs since then have carried additional Zen code. One of the latest pull requests is Zen in amd64_edac. It ended up being a few hundred lines of code adding Zen / 17h support to the Error Detection and Correction code.
| 19
| 1,760,719,768.644137
|
https://www.phoronix.com/news/Radeon-AMDGPU-Fixes-4.10-Soon
|
More Radeon/AMDGPU Fixes For Linux 4.10 Ahead Of The Merge Window
|
Michael Larabel
|
With Linux 4.9 expected for release this weekend and the 4.10 merge window to then immediately open, Alex Deucher of AMD sent in an early batch of fixes atop the earlier feature material of AMDGPU/Radeon DRM changes for Linux 4.10.
The AMDGPU DRM driver in Linux 4.10 will introduce a new VM manager and better power management, support for multiple virtual displays, RPM fan information is now exposed, clock/power-gating improvements, and other improvements.
With this Thursday's pull request are fixes for getting the Radeon and AMDGPU DRM drivers better positioned for 4.10. As usual, most of the work is focused on AMDGPU rather than the mature Radeon DRM code.
Some of these changes in this latest pull request include cursor fixes, DPM fixes for some new Southern Islands GPU variants, PowerPlay fixes, and clock/power-gating fixes. This latest pull has 39 changes and touches just about 500 lines of code.
| 10
| 1,760,719,769.597557
|
https://www.phoronix.com/news/AMD-ARM-Where-Are-They
|
AMD's ARM Efforts Appear Stalled, But At Least Zen Should Be Great
|
Michael Larabel
|
This week's news of Qualcomm sampling a 10nm 48-core ARMv8 SoC for servers made me wonder where are AMD's ARM SoCs -- and the long-awaited development boards -- and thus been following up with a few sources this week.
I last wrote about the AMD ARM efforts in September when mentioning It Doesn't Look Like We'll See AMD ARM Development Boards This Year. The AMD ARM developer boards are long overdue but the promising LeMaker Cello and HuskyBoard products still aren't available and I hear from sources that AMD ARM has been having some PCI-E / SATA issues as what caused the original delays while now AMD seems to be losing interest in ARM, especially with Zen being right around the corner.
The HuskyBoard is M.I.A. and the newer Cello Board was supposed to ship in Q2'2016 but clearly missed that. LeMaker and other resources on the Cello haven't even bothered to be updated past their "The LeMaker Cello main board is scheduled to ship in Q2 of 2016" text to provide any update on the matter.
It looks like we might never see these low-cost AMD ARM development boards. SoftIron appears to be the only company selling an AMD Opteron A1100 system so far, and claims to be in stock, but will set you back $2495 USD.
So from those I communicated with this week, the unofficial explanation is that AMD all but dropped these ARM efforts to focus on Zen and according to one there doesn't seem to be a future with AMD and ARM for the short term. Pity, as I've been very excited to see the AMD ARM Linux experience and benchmarks... But if there's a consolation prize, AMD Zen does indeed appear to be shaping up very nicely.
| 39
| 1,760,719,770.18635
|
https://www.phoronix.com/news/AMDGPU-PRO-16.50-Download
|
Here Is The AMDGPU-PRO 16.50 Download Link
|
Michael Larabel
|
AMD ran into a snag getting out the updated proprietary hybrid Linux driver stack this morning, but it's now available for download from AMD.
This page has the 16.50 Linux x86/x86_64 driver available for download.
The listed highlights are FreeSync, R7 M350/M370/M465X support, install scripts for RHEL/CentOS 7.3, CentOS 6.8, SLED/SLES 12 SP2, and DirectGMA for OpenGL. It doesn't mention the Southern Islands GCN 1.0 support as part of the changes while it's in there unofficially.
See this morning's article for more details: AMDGPU-PRO 16.50 Rolls Out With GCN 1.0 Support, FreeSync.
| 15
| 1,760,719,771.15773
|
https://www.phoronix.com/news/AMD-DC-Display-Code-Slimmer
|
New AMD DC/DAL Patches Continue To Slim Down The Codebase
|
Michael Larabel
|
The latest patches for the AMDGPU DRM driver's DC code -- what was previously known as DAL -- have been published and they reduce the size of the code-base some more.
The newest set of 12 patches for DC (DAL) are now on the amd-gfx list. They provide some bug fixes but there is also a rework of the display clock code for allowing more code sharing between hardware generations. This greater code sharing strips out more than two thousand lines of code: 920 insertions, 3478 deletions. Great to see what started out as a massive code-base continuing to get slimmed.
This massive display rework to the AMDGPU driver is slowly but surely getting into shape, hopefully we'll see it ready for Linux 4.11 but that's yet to be confirmed.
| 28
| 1,760,719,771.717258
|
https://www.phoronix.com/news/AMD-DAL-Renamed-DC
|
AMD "DAL" Being Renamed To "DC"
|
Michael Larabel
|
AMD's big display abstraction layer (DAL) code-base that's used by AMDGPU-PRO but not yet mainlined in the Linux kernel for providing HDMI 2.0, future FreeSync/Adaptive-Sync, HDMI/DP audio, and other modern display features is DAL no more.
When DAL was presented earlier this year it was a monstrous code-base as it's long been the display code used by their Windows and Linux proprietary drivers and was initially filled with lots of abstractions to fit into the DRM space. Over the past several months AMD developers have been working to significant slim down DAL and make it designed for the Linux kernel, in what is basically "DAL 3.0", according to past presentations. For those new to reading Phoronix or don't recall our dozens of articles on the topic, there is the DAL presentation from XDC2016 to provide more background information.
AMD has been working hard to get DAL ready for the mainline Linux kernel since they plan to use it for upcoming hardware releases and don't want to continue implementing a non-DAL code-path. In November we saw a lot of new DAL code published and it looks like that will continue until they are ready to merge the code, likely for Linux 4.11.
The latest is that DAL is being renamed. With this patch DAL directories are being renamed to display while the rest of the patches are renaming "DAL" to "DC".
| 47
| 1,760,719,772.453397
|
https://www.phoronix.com/news/AMD-CCC-Linux-Doing-The-Pre-Req
|
It Looks Like We'll Still See A GUI Control Panel For AMD Linux
|
Michael Larabel
|
Earlier this year I exclusively reported on the "Radeon Settings" GUI control panel may be open-sourced for AMD Linux users but since then I hadn't heard anything publicly or privately about getting this graphics driver control panel on Linux for AMDGPU-PRO and the fully-open AMDGPU stack. But it looks like that it's still being worked on internally at AMD.
The Radeon Settings that AMD rolled out earlier this year to Windows users is written using Qt and is the successor to the Catalyst Control Center that had been present for many years. Under Linux, there is no good "GUI control panel" for the open-source drivers with options like DriConf being really dated and limited. With the old AMD proprietary stack there was AMDCCCLE, AMD Catalyst Control Center Linux Edition, but since the premiere of AMDGPU-PRO there hasn't been any working GUI control panel there either. All the tunables for now are left to either hand-editing files, setting environment variables, and interacting with sysfs/debugfs manually to alter the driver's behavior.
Fortunately, we finally hear new information this week on the fate of having an AMD GUI control panel on Linux. From our veteran AMD Linux communicator, in the forums this week was the following comment: "Yep, there was some discussion of this a couple of months ago - porting CCC will become feasible, but all the plumbing needs to go into the driver first."
Great to see it's "feasible", but first they need to take care of all the prerequisites. This comment came after this week AMDGPU in Linux 4.10 will finally have RPM-based fan information via hwmon, a common attribute that is displayed via the control center on Windows and a common tunable/metric for enthusiasts/gamers. Hopefully we'll see the GUI driver control panel ported over to AMDGPU not too far into 2017.
| 52
| 1,760,719,773.232099
|
https://www.phoronix.com/news/AMDGPU-PRO-OpenCL-Advantage
|
OpenCL Remains One Of AMDGPU-PRO's Main Advantages, Can't Wait For It To Be Open
|
Michael Larabel
|
While many in our forums and other Linux communities want to see "AMDGPU-PRO die" or for AMD to stop supporting the hybrid/proprietary driver given the pace of RadeonSI development for OpenGL and the emerging RADV for (unofficial) Vulkan support, OpenCL remains one of AMDGPU-PRO's strongholds. AMD has been working on opening up their proprietary compute stack, but for now it's there. Here are some fresh AMDGPU-PRO 16.40 benchmarks versus NVIDIA in LuxMark, one of the real-world OpenCL workloads where the AMD blob does very well.
With the OpenCL Beignet support being stagnate, there isn't much of good open-source OpenCL support for AMD right now, but they are working on delivering that through ROCm. Our AMD veteran communicator in the forums recently clarified unofficially about their work/plans:
Clover is not dead in the water (unless everyone in the community decides that it should be), it's just that our efforts shifted a year or so ago from clover to rebuilding our proprietary OpenCL stack on top of ROCM and open sourcing the remaining closed-source bits.
The first step along the way was replacing the proprietary/third-party C-language parser with clang (done); remaining tasks include:
(a) modifying OpenCL to run over ROCM rather than the current back-end (which is based on the proprietary OpenGL driver), and
(b) using the LLVM-based native compiler (also used in ROCM/HCC and the open source Mesa drivers) rather than the proprietary shader compiler.
We are aiming to have a developer preview of the modified OpenCL stack (still closed source runtime but using a lot more open source code) around mid-December. In terms of their short-lived love with Clover Gallium3D for OpenCL, he clarified, "When we started work on clover the amount of work required to open source our existing implementation was too high to be practical and HSA/ROC work was just starting, so going with clover and hoping other vendors would do the same made sense (although the "other vendors" part didn't really happen). Today, on the other hand, we have an open source compute stack (ROC) and open source native compiler so going the rest of the way to an open source OpenCL stack built on ROC and the associated compiler work is manageable. It's still a lot of work but less than re-implementing in clover... and even if we did take the clover approach we would still have to plumb clover back into the rest of our compute stack in order to avoid a lot of duplicated effort going forward."
Anyhow, in heavy OpenCL workloads, the AMDGPU-PRO 16.40 performance can be quite good. LuxMark in particular is one of the workloads that can show of AMD's compute potential compared to NVIDIA. AMD says they perform well in it since it's a real-world workload as opposed to favoring red GPUs, but it does appear to have some less than ideal performance with NVIDIA GPUs -- as you'll see on the graphs, with e.g. the GTX 1070 being faster than the GTX 1080.
And as compliments to the AMDGPU-PRO stack, there is OpenCL 2.0 as opposed to OpenCL 1.2 still with NVIDIA's proprietary driver.
Tests via this OpenBenchmarking.org result file. You can see many of our other, older OpenCL benchmarks with other workloads for those interested.
So it will sure be nice once the proprietary OpenCL stack is fully-opened, but will still likely be a few months of work before it's really in shape and then potentially a few more months after that before it will be found out-of-the-box in the (non-rolling) distribution of your choice. At least it's something to look forward to for 2017 and is a main advantage of AMDGPU-PRO for the interim.
| 54
| 1,760,719,773.707858
|
https://www.phoronix.com/news/AMDGPU-Intel-4.9-PRIME-Ex-Patch
|
Linux 4.9 Has Tear-Free PRIME Offloading Between Intel & AMDGPU
|
Michael Larabel
|
Quietly landing last week into the mainline Linux kernel as part of the AMDGPU fixes is support for tear-free PRIME offloading between Intel and AMDGPU.
The drm/amdgpu: Attach exclusive fence to prime exported bo's. (v5) patch was merged fairly late into the Linux 4.9 kernel merge window.
The patch by Mario Kleiner explains, "External clients which import our bo's [buffer objects] wait only for exclusive dmabuf-fences, not on shared ones, ditto for bo's which we import from external providers and write to. Therefore attach exclusive fences on prime shared buffers if our exported buffer gets imported by an external client, or if we import a buffer from an external exporter... Prime export tested on Intel iGPU + AMD Tonga dGPU as DRI3/Present Prime render offload, and with the Tonga standalone as primary gpu."
This should allow for tear-free PRIME offloading on the multi-GPU systems with the Intel DRM driver and an AMDGPU-supported Radeon GPU. Further details via this bug report.
| 21
| 1,760,719,774.671617
|
https://www.phoronix.com/news/AMD-ROC-1.3-SC16
|
Radeon Open eCosystem 1.3 Platform Brings Polaris & Other Features
|
Michael Larabel
|
AMD used the SC16 super-computing conference today announce version 1.3 of the Radeon Open eCosystem platform.
They call this new version of ROCm the "Most Versatile Open Source Platform for GPU Computing." With ROCm 1.3 there is now official support for Polaris GPUs and they announced planned support for AMD Zen, Cavium ThunderX, and IBM POWER8 CPUs. AMD also announced their plan to support OpenCL atop upcoming ROCm releases.
The Radeon Open eCosystem 1.3 release also adds support for Docker containers and GPU hardware virtualization with KVM pass-through, support for new math libraries, and other changes.
Will be real nice to see OpenCL support atop this open-source Radeon stack and it's slated to be available for testing in December.
Those interested in the latest Radeon Open eCosystem release can find all of the details on today's update at AMD.com.
| 0
| 1,760,719,775.11209
|
https://www.phoronix.com/news/AMDGPU-PRO-Linux-4.8-DKMS
|
A Word Of Warning When Using AMDGPU-PRO On An Unsupported Kernel
|
Michael Larabel
|
The recent slowdowns seen with AMDGPU-PRO 16.40 on my test systems may be attributed to the Linux 4.8 kernel being not properly supported by this hybrid kernel driver.
After talking with John Bridgman last week, AMDGPU-PRO 16.40 is only intended to work with Linux 4.7 kernels and older while the official focus is on Linux 4.4 of Ubuntu 16.04. I had tested with Linux 4.8 since that's what I was using during the AMDGPU open-source testing as well as during my NVIDIA testing with its proprietary driver, which quickly supports new kernel versions.
But the strange thing is when testing on my 4.8 configuration, I don't recall seeing any DKMS errors when installing AMDGPU-PRO 16.40. I've encountered the AMDGPU-PRO 16.40 build problems previously when trying out too new of a kernel on some systems, but during this recent testing I don't recall encountering any problems, obviously otherwise would have reverted to an older kernel, and the 16.40 install process had succeeded.
Bridgman also explained that while the hybrid OpenGL and Vulkan driver components make use of ioctls not found in the mainline AMDGPU DRM code, they can still work albeit slower. One of the ioctls not present in the mainline AMDGPU code is for fencing used by the hybrid driver. So rather than having fences, the driver waits for timeouts. "We try to make them degrade rather than fail when running over the wrong kernel driver but that is not a release requirement."
No error messages are printed when this happens, but hopefully in a future AMDGPU-PRO update they will make it more clear when the unsupported combination is in use.
Last week I did run some AMDGPU-PRO vs. NVIDIA OpenCL benchmarks prior to finding out about the -PRO + 4.8 issue. So take those results worth a grain of salt, but actually the numbers aren't all that bad for AMDGPU-PRO and the issue seems to have greater impact on Vulkan and OpenGL. Anyhow, I'll have some fresh AMDGPU-PRO 16.40 benchmarks up shortly when falling back to Linux 4.4.
| 76
| 1,760,719,776.043793
|
https://www.phoronix.com/news/AMDGPU-F16-Lands-VI
|
FP16 Half-Precision Support Added To AMDGPU LLVM Back-End
|
Michael Larabel
|
The AMDGPU backend within LLVM that's used by the open-source Radeon graphics driver stack has landed half-precision / FP16 support.
With this commit overnight is nearly five thousand lines of new code for supporting F16 half-precision support in this GPU back-end for AMD hardware. This new LLVM code was under review the past few weeks but is now ready for LLVM 4.0.
This half-precision support is capable on AMD Volcanic Islands (3rd gen GCN) hardware and newer. Now if only the open-source AMD OpenCL/compute state was much more mature and widespread...
| 15
| 1,760,719,776.491335
|
https://www.phoronix.com/news/AMDGPU-PRO-16.40-Released
|
AMDGPU-PRO 16.40 Released For Ubuntu & Red Hat Linux Systems
|
Michael Larabel
|
There is finally a new release of the AMDGPU-PRO hybrid Radeon graphics driver stack for Linux.
With this new release, AMDGPU-PRO 16.40, there is finally official support for Red Hat based systems rather than only Ubuntu LTS releases. However, no change-log has yet been published to outline the other changes with AMDGPU-PRO 16.40. The previous public release was based on AMDGPU-PRO 16.30.
The Ubuntu AMDGPU-PRO 16.40 download link is available here. It seems to only indicate Ubuntu 16.04 LTS support and no mention of Ubuntu 16.10.
The Red Hat download link can be found here with support for RHEL 7.2. No word on whether it will work with Fedora, but likely not if it's targeting only the RHEL7 stack.
More details on the AMDGPU-PRO 16.40 Linux changes as they become available. Benchmarks to follow.
| 110
| 1,760,719,777.528769
|
https://www.phoronix.com/news/AMD-Zen-Memory-Encrypt-State
|
The Linux State Of AMD's Zen x86 Memory Encryption
|
Michael Larabel
|
With AMD's forthcoming Zen processors is support for some new memory encryption technologies that are of particular benefit for virtualized environments.
I wrote about Linux patches for AMD memory encryption earlier this year while since then more information has come to light. At last month's Linux Security Summit, David Kaplan presented on these technologies coming with Zen; only today I had come across the slide deck for this presentation.
The technologies come down to Secure Memory Encryption (SME) and Secure Encrypted Virtualization (SEV). SME provides memory encryption on a per-page-table basis using AMD's ARM-based security co-processor. AMD SME + SEV are designed against both user-access attacks and physical access attacks with a particular focus on VM / hypervisor security.
While there are the open-source kernel patches for supporting these memory encryption technologies, the slides confirm that AMD's Secure Processor firmware is not going to be open-source but rather a binary blob distributed with AGESA.
Those interested in this forthcoming AMD memory security technology from the Linux perspective can see this PDF slide deck.
| 19
| 1,760,719,778.657255
|
https://www.phoronix.com/news/AMD-DAL-XDC2016
|
AMD's DAL Was Just Presented At XDC2016, Still Not Clear When It Will Be Mainlined
|
Michael Larabel
|
Harry Wentland of AMD just presented at the XDC2016 conference about DAL, the big Display Abstraction Layer code-base, which many AMD Linux users have been waiting to see merged in order to have Polaris audio support and this is one of the stepping stones for seeing FreeSync/Adaptive-Sync and other modern display capabilities.
We have been covering DAL for months since AMD open-sourced it and since then they've been trying to clean it up, remove some redundancies compared to what core DRM offers, etc. DAL is a big piece of the puzzle that's left for getting mainlined so the AMDGPU open-source kernel driver can be closer to feature parity with the closed-source driver and what's provided on Windows.
This "DAL 3" code is focused on supporting HDMI 2.0, DisplayPort 1.4, 4K@120Hz, 5K display support, 8K display support plus for supporting DRM atomic mode-setting, dynamic refresh rates, compression, wide gamut displauy, HDR, and more.
DAL is currently made use of in the AMDGPU-PRO driver, but there is no indication of when it may be ready for merging into the mainline AMDGPU DRM driver or even if it's a high a priority now for AMD developers to get it merged.
Hearing about mainlining plans was a big hope going into this presentation, but unfortunately nothing to share at this time. If you are interested in learning more about DAL, you can see the PDF slides or watch the stream embedded below. As mentioned yesterday, it's certainly not coming for Linux 4.9 so now Linux 4.10 would be the next possible target.
| 18
| 1,760,719,779.30785
|
https://www.phoronix.com/news/More-PowerPlay-Linux-4.9
|
More AMD PowerPlay Fixes Queued For Linux 4.9
|
Michael Larabel
|
The AMDGPU DRM code for Linux 4.9 is already queued in DRM-Next with virtual display support and other features as well as experimental GCN Southern Islands support while today another feature pull request was submitted to DRM-Next with more Radeon/AMDGPU changes.
This final feature pull for the Radeon/AMDGPU code planned for Linux 4.9 brings mostly PowerPlay changes. Alex Deucher of AMD noted in today's pull request, "Last set of radeon and amdgpu changes for 4.9. This is mostly just the powerplay cleanup for dGPUs. Beyond that, just misc code cleanups and bug fixes."
The pull request with the numerous PowerPlay changes can be found via this pull request. With this being the final PR for AMDGPU Linux 4.9 features, it also confirms there will be no DAL (Display Abstraction Layer) code for AMDGPU in Linux 4.9... DAL is needed for supporting HDMI audio with Polaris, FreeSync, and other display-related features to bring it to parity with the closed-source AMD driver and other modern display features.
| 41
| 1,760,719,780.020726
|
https://www.phoronix.com/news/AMDGPU-SI-Next-4.9
|
Southern Islands Support Will Come To AMDGPU On Linux 4.9
|
Michael Larabel
|
One month after the first AMDGPU feature pull of new functionality for DRM-Next to in turn land in Linux 4.9, the second feature pull request has now been sent out and it presents experimental Southern Islands (GCN 1.0) support for AMDGPU.
As has been expected for a while, the Southern Islands hardware (a.k.a. GCN 1.0 / HD 7000 series) will be experimentally supported by this AMDGPU DRM driver. This Southern Islands support will be disabled by default and must be configured at the kernel's build-time via a Kconfig switch. This is similar to the still-experimental GCN 1.1 support in AMDGPU.
With this S.I. support, the AMDGPU DRM driver is capable of supporting all GCN GPUs. This is good news for those ultimately wanting Vulkan support, AMDGPU-PRO hybrid driver usage for OpenCL/OpenGL, and the other benefits of the improved design of AMDGPU DRM over the mature Radeon DRM. But keep in mind this needs to be enabled via a Kconfig switch and most Linux distributions will continue using GCN 1.0/1.1 on Radeon DRM for the near future at least. This AMDGPU SI support also requires using the latest Mesa and xf86-video-amdgpu DDX Git code too.
Besides this new pull request having AMDGPU SI support, it also has various TTM memory management clean-ups, run-time power management fixes, S3/S4 suspend fixes, power improvements, and other code clean-ups and optimizations.
The complete list of changes with this 4.9 AMDGPU secondary pull request can be found here. Notably absent from this pull request and not being readied for Linux 4.9 is AMD's big DAL rework of its display abstraction code.
| 125
| 1,760,719,780.730928
|
https://www.phoronix.com/news/R9-290-Still-Not-Fixed
|
Linux+Mesa Git Remains Problematic For Some Regressed R9 290 GPUs
|
Michael Larabel
|
Last week a Mesa fix landed to target the Radeon R9 290/390 performance regression that's been covered a few time on Phoronix since the issue was originally noticed. While the Mesa fix is working for some users, it didn't fix all problems, including with my Hawaii test card.
Some users reported that the latest Mesa Git has fixed their issue with select problematic Radeon GPUs regressing with very slow performance while others have said their issues still persist. On a Radeon R9 290, I tried out the latest Mesa Git (with the newest Padoka PPA, updated on Monday) as well as Linux 4.8 Git. Unfortunately, for my Radeon R9 290, the performance is still heavily regressed...
No real change with this newest Linux and Mesa code compared to a known faulty state for this R9 290 system from a few weeks ago... For just how bad the regression is:
Ouch.
| 43
| 1,760,719,781.593796
|
https://www.phoronix.com/news/AMD-GPUOpen-HIP-Summer
|
AMD's GPUOpen HIP Project Made Progress Over The Summer
|
Michael Larabel
|
The HIP project has made good progress over the summer. HIP from AMD's GPUOpen project is part of the puzzle for converting CUDA to portable C++ code. That source code can then run on AMD GPUs while having little to no performance impact, at least according to AMD.
HIP developers have been working on stability improvements, packaging work, testing with the AMD testing suite, updates to match each ROCm release, and more. Some of the recent changes include FP16 software support, support for Hawaii discrete GPUs, support for hipArray, improved profiling, better clang-hipify utility, and support for dynamic shared memory allocations.
Developers have also made public a HIP developer-preview branch on GitHub. That developer-preview code has the latest fixes and features, including support for the CUDA Driver API as the latest addition.
You can learn more about this HIP work via GPUOpen.com or going straight to the GitHub repository.
| 25
| 1,760,719,782.299206
|
https://www.phoronix.com/news/AMD-ARM-Dev-Boards-End-2016
|
It Doesn't Look Like We'll See AMD ARM Development Boards This Year
|
Michael Larabel
|
Things don't appear to be looking up for AMD's ARM efforts. It's looking like we probably won't be seeing AMD ARM development boards publicly available this year, if not the end of 2016, and there won't be many of them going around.
Last month I wrote about There's Still No Sign Of AMD's Low-Cost ARM Development Boards. While I've been quite excited to get my hands on some AMD ARM hardware, I haven't been able to yet. This is while the AMD-powered 96Boards HuskyBoard was supposed to ship at the end of 2015 and the LeMaker Cello AMD A1120 board announced earlier this year was supposed to ship by the end of Q2. The Cello is a quarter late and it's looking like it will be at least another quarter before we possibly see any AMD ARM hardware.
Since writing that article last month, I heard from a trusted comrade that a technical obstacle has delayed AMD's ARM SoCs. Board partners are waiting on AMD but it's looking like it won't be until the end of the calendar year when they are back up to speed... And it's looking like they won't end up being produced in large volumes.
That's a pity to hear and will now be curious to see if SoftIron's Overdrive system with the A1100 SoC ends up shipping in September or this technical problem has pushed back all AMD ARM products. Of course, the SoftIron system is rather expensive ($600+) but soon as I can get my hands on an affordable development board (e.g. Cello or HuskyBoard) will be glad to order one for Linux benchmarking.
| 9
| 1,760,719,783.261241
|
https://www.phoronix.com/news/AMD-Advanced-Media-Framework
|
AMD Open-Sources Advanced Media Framework, But No Linux Support Yet
|
Michael Larabel
|
AMD this week open-sourced the Advanced Media Framework (AMF) as their replacement to the earlier AMD Media SDK. But before getting too excited about this latest AMD open-source project, there isn't yet any Linux support.
AMD's AMF is self-described as, "a light-weight, portable multimedia framework that abstracts away most of the platform and API-specific details and allows for easy implementation of multimedia applications using a variety of technologies, such as DirectX 11, OpenGL, and OpenCL and facilitates an efficient interop between them." Another description puts it as "The AMF SDK allows optimization of application performance by utilizing CPU, GPU compute shaders and hardware accelerators for media processing. These optimizations are applicable to a wide range of applications such as gaming or content creation. Programming of AMD Video Engines (UVD and VCE blocks) is also an important part of the functionality that AMF provides to developers."
AMD AMF is designed for AMD GCN GPUs and newer. This media framework is developed under the MIT license. With OpenGL and OpenCL support being targeted and AMF also making use of FFmpeg, hopefully we'll see it ported to Linux -- if not by AMD, the open-source community. Would be nice to see more cross-platform work in the video acceleration department between Windows and Linux.
The project's samples directory shows off hardware accelerated video playback, simple video decoders/encoders/converters, and more with AMF. The code is hosted on GitHub and there's also the GPUOpen page.
| 6
| 1,760,719,783.641682
|
https://www.phoronix.com/news/CodeXL-2.2-AMDGPU-PRO
|
AMD GPUOpen's CodeXL 2.2 Now Supports Linux With AMDGPU-PRO
|
Michael Larabel
|
Earlier this year AMD made CodeXL 2.0 open-source as a developer tool with GUI centered around profiling/optimizing D3D, OpenGL, and Vulkan (since CodeXL 2.1) under Windows and Linux. Today marks the release of CodeXL 2.2.
CodeXL 2.2 is out as the newest open-source release and on the Linux side has support for interfacing with the AMDGPU-PRO hybrid driver stack. Previously the Linux support for CodeXL depended upon the now-deprecated Catalyst driver. There's also support for Ubuntu 16.10. On the WIndows side is support for the Windows 10 Anniversary Update.
CodeXL 2.2 also brings official support for AMD Radeon RX 460/470/480 Polaris graphics cards. Other additions with CodeXL 2.2 include support for capturing CPU-only and GPU-only traces, cross-platform capturing between Windows and Linux, support for new 7th Generation APUs, support for OpenGL-OpenCL interoperability in the debug mode, and a wide range of other changes.
Developers interested in the open-source CodeXL 2.2 profiling software can find out more details about this big update via GPUOpen.com.
| 0
| 1,760,719,784.495143
|
https://www.phoronix.com/news/AMD-Radeon-RX-460
|
Radeon RX 460 Released, Linux Review Later This Week
|
Michael Larabel
|
Just days after the Radeon RX 470 began shipping, the Radeon RX 460 is shipping this morning and the embargo concerning the RX 460 has expired.
This Polaris 11 graphics card has 14 compute units, 896 stream process, 1090MHz boost clock speed with 1200MHz boost clock speed, and is rated for up to 2.2 TFLOPS of compute power. The video memory is GDDR5 on a 128-bit bus. The TDP for this graphics card is less than 75 Watts.
Pricing on the Radeon RX 460 will begin at $99 USD. This morning most of the graphics cards I am seeing from AIB partners at NewEgg and Amazon are $109~119+ USD.
Similar to the forthcoming RX 470 Linux tests, AMD didn't send over any review samples of the RX 460 either... However, thanks to their low price tag, I already placed an order for a Radeon RX 460 and it looks like it will be arriving tomorrow or Wednesday. Tomorrow my RX 470 Linux review will be posted while by Thursday or Friday then I should have my RX 460 Linux review out there with OpenGL, OpenCL, and Vulkan benchmarks.
For those curious, the Radeon RX 460 I ordered just minutes ago from NewEgg was the cheapest model I found so far: PowerColor RED DRAGON Radeon RX 460 DirectX 12 AXRX 460 2GBD5-DH/OC 2GB. The card cost $109.99, or $125.17 when factoring in tax and shipping.
Stay tuned for many red Linux benchmarks this week! If any Phoronix Premium members have any special test requests, feel free to let me know ASAP so I can try to incorporate them into the Linux reviews.
| 63
| 1,760,719,785.028789
|
https://www.phoronix.com/news/AMD-FreeSync-DDX-Change
|
AMD Publishes More Open-Source FreeSync Code
|
Michael Larabel
|
AMD published a patch this week for prepping its X.Org driver for supporting FreeSync.
AMD FreeSync aims to address game stuttering and tearing by allowing the monitor's refresh rate to dynamically change with Low Framerate Compensation when used in conjunction with a newer AMD graphics card and a supported display.
The patch published this week won't magically bring FreeSync to the open-source AMD Linux driver stack, but it's working in that direction. The new patch is just about allowing the xf86-video-amdgpu X.Org driver to notify the OpenGL game when there is a FreeSync-capable application/client to enable/disable FreeSync when entering/exiting the full-screen game/application. The xf86-video-amdgpu driver patch is on the amd-gfx mailing list.
This patch doesn't do the heavy lifting for FreeSync support with the big piece of the puzzle being needed inside the AMDGPU DRM driver. The FreeSync support in AMDGPU DRM is contingent upon the DAL display abstraction layer code that has yet to be mainlined. It won't be until at least Linux 4.9 that the DAL code is merged in AMDGPU as it's a massive addition to the driver. However, the DAL support is already being prepped within the AMDGPU-PRO hybrid driver with its DKMS DRM driver.
Hopefully by the end of the year we'll see working AMD FreeSync/Adaptive-Sync support on the Linux desktop.
| 29
| 1,760,719,785.939493
|
https://www.phoronix.com/news/AMD-FirePro-July-2016
|
AMD Releases Updated Catalyst Pro / FirePro Unified Driver For Linux
|
Michael Larabel
|
It looks like that AMD is going to continue maintaining their Catalyst-based Linux driver until the new AMDGPU-PRO driver architecture is up to scratch. That is, support for FirePro workstation users at least caring about stable support on Enterprise Linux distribution. A new Catalyst Pro / FirePro Unified driver for Linux was released this week.
I wasn't aware that AMD was still maintaining this non-AMDGPU-PRO based alternative for FirePro users, but it makes sense given these heavy workstation users are AMD's main cash cow on the Linux side and that there would be support continued for this older driver stack until AMDGPU-PRO becomes a suitable, stable replacement for mission critical work.
This new driver update under the AMD FirePro Software Suite (also called the "Catalyst Pro" driver on the page) is marked 15.302.2301 and was released this past week.
All of my FirePro hardware is many years old (pre-GCN) with not getting any workstation review samples since, so unfortunately I will not be able to do any driver comparisons of this driver vs. AMDGPU-PRO, etc. The release notes mention that there's a fix for a OpenCL-OpenGL texture sharing deadlock has been resolved, a performance fix dealing with slow performance after sleeping, a Xinerama fix, and some Maya bugs have also been resolved.
FirePro Linux users wanting to try out this latest driver stack can do so via AMD.com.
| 16
| 1,760,719,786.673434
|
https://www.phoronix.com/news/AMD-Radeon-Pro-SSG
|
Radeon Pro SSG Packs 1TB Of SSD Storage On The Graphics Card
|
Michael Larabel
|
Last night after writing about the Radeon Pro WX series I immediately fell asleep, but it turns out AMD continued with their SIGGRAPH 2016 announcements by rolling out the Radeon Pro SSG, a graphics card with onboard solid-state storage.
The Radeon Pro SSG is their first graphics card that has an onboard M.2 SSD. The 1TB of onboard storage is intended to supplement the high-performance video memory traditionally found on graphics cards. Having this immense amount of storage that can be quickly accessed by the GPU can help in areas of video encoding/decoding, VR, rendering of large scenes, and a lot in the scientific fields or anything else dealing with huge data-sets.
The GPU for the Radeon Pro SSG is, of course, Polaris based. The SSD storage is connected to the GPU via a dedicated PCI Express bus; it won't be as fast as dedicated vRAM, but still should be much better than transferring to/from a local disk off the graphics card. AMD will be selling developer kits of the Radeon Pro Solid State Graphics soon while it won't go into full production until next year.
Those wanting to learn more about the Radeon Pro SSG can visit Radeon.com.
| 24
| 1,760,719,787.312867
|
https://www.phoronix.com/news/AMDGPU-Mesa-RX480-July-16
|
Fresh AMDGPU-PRO vs. Mesa 12.1-dev Performance FIgures For The Radeon RX 480
|
Michael Larabel
|
As it's been a few weeks since the Radeon RX 480 "Polaris" launch and there's been more RadeonSI Polaris commits since launch day, I figured I'd run some fresh benchmarks this weekend of the open vs. hybrid AMD Linux driver stacks.
While there have been RadeonSI Gallium3D related work published since, there's still just the same launch-day AMDGPU-PRO 16.30.3 driver release available, even while the Windows driver stack has seen multiple updates in this time for the RX 480. But that launch-day driver is slightly newer than the AMDGPU-PRO tests I did a few days in advance of its formal launch, so may be somewhat more interesting.
On the open-source driver side I was using Linux 4.7 Git and Mesa 12.1-dev Git from the Padoka PPA as of this Saturday. I used Linux 4.7 Git as the 4.8 DRM-Next tests didn't end up being too interesting that I did a few days ago. So here's the latest AMDGPU-PRO vs. RadeonSI mainline performance as of this weekend.
View more of these OpenGL driver benchmarking results via this OpenBenchmarking.org result file.
| 23
| 1,760,719,788.427024
|
https://www.phoronix.com/news/Radeon-New-Beta-Sign-Up
|
AMD Posts Sign-Up For Radeon Software Beta Testing
|
Michael Larabel
|
As a follow-up to this morning's article about a new driver beta program, the sign-up application is now available.
As outlined in my posting this morning when the embargo expired, the driver beta program hopefully won't be too relevant for Linux enthusiasts/gamers once their Vulkan driver is open-sourced (I still haven't heard anything new from AMD when that might happen...) and when the last kinks with the RadeonSI Gallium3D stack are worked out for the latest OGL4 Linux games. Nevertheless, this new program is said to be for their AMDGPU-PRO Linux driver too.
There isn't any Linux mentions on the specific page, but those interested in applying to the beta testing program can do so via this Radeon.com page.
| 6
| 1,760,719,789.516459
|
https://www.phoronix.com/news/AMD-New-Beta-Program
|
AMD Is Launching A New Beta Program For Their Drivers
|
Michael Larabel
|
For years AMD had a very active, community-driven semi-private beta program for their (Catalyst, at the time) drivers but in recent years while going through tough financial times they cut back the program. However, they will now be working to restore this program and they will be looking for Linux participants too.
During the earlier beta program there was Linux coverage too with select community members being part of the NDA'ed process. It's been quiet in recent years but now they are working to ramp things up and also hired one of the former community organizers for their beta program.
Terry Makedon was talking about this renewed beta effort on a conference call recently when also talking about WattMan and how "Radeon Settings" might be open-sourced.
It's good to hear they are back to engaging on this beta program, but it will be interesting to see how useful it is on the Linux side. After all, AMD continues open-sourcing more of their code and the AMDGPU kernel DRM driver is now leveraged by their hybrid driver too. It may though be useful for those wanting to use AMD's closed-source OpenGL library for Linux gaming to help test what new Linux games do/don't work with it. Additionally, until the AMD Vulkan Linux driver is open-sourced, it may be useful there too. So for those few closed areas it may be worthwhile but it's great that more code is being opened, AMD is being more public on mailing lists with their patches and early feature work, etc.
It will be interesting to see what happens with this program in the months ahead on the Linux side.Update: The beta sign-up is now available for applicants.
| 4
| 1,760,719,789.781781
|
https://www.phoronix.com/news/RX480-Firmware-Blobs-Updated
|
The Updated AMD Polaris Firmware Blobs Needed For RX 480 Support Land
|
Michael Larabel
|
One day ahead of the Radeon RX 480 "Polaris" launch, the necessary firmware updates for the production graphics card support have landed in linux-firmware.git.
A few days back I outlined the requirements for the open-source RX 480 Linux support. Besides needing a Linux kernel branch, Mesa Git, and more, newer AMDGPU firmware files are needed beyond what's been distributed via linux-firmware.git. These updated firmware files were available via the radeon_ucode HTTP/FTP directory while as of this afternoon they have landed in the official Git tree.
So if you plan on picking up a Radeon RX 480 for Linux this week or in the near future, be sure to get the latest Linux-Firmware Git if you are wanting to use the fully open driver stack.
Benchmarks coming up tomorrow at 9.00 EST!
| 54
| 1,760,719,790.856312
|
https://www.phoronix.com/news/AMD-Zen-Intel-KBL-Delay
|
AMD Zen Reportedly Delayed Until Early Next Year
|
Michael Larabel
|
According to reports, AMD's Zen processors have been delayed until the start of the next year and it's also affecting Intel's Kabylake launch.
While we were looking for AMD's much anticipated Zen processors were expected to launch by the end of Q3 according to information from AMD just recently, now it's believed that Zen might not be shipping until January 2017.
Via information today from Digitimes, the launch schedule for Zen has been pushed back until January 2017 at the annual CES show while Kabylake is also being pushed back until that point too.
The reason for the few month delay for both Zen and Kabylake is said to be due to supply chain issues and slow market demand. It's really sad to hear Zen being delayed while Kabylake is just Intel's successor to Skylake before Cannonlake and is less to get excited about, but now the Intel Linux developers will have a few extra months to ensure the Kabylake support is in good standing for launch, particularly on the graphics side.
| 62
| 1,760,719,791.401315
|
https://www.phoronix.com/news/Computex-2016-AMD-Go
|
AMD Confirms RX 480 At $199 USD, Other APU & Polaris Announcements
|
Michael Larabel
|
If you weren't able to watch the AMD Computex 2016 live-stream happening now, here are my key notes from the event.
The live-stream talked about AMD Polaris GPUs, 7th Gen Mobile APUs, and future tech.
- Yes, the Radeon RX 480 will sell for just $199 USD! This is part of their goal to "bring real VR to the first 100 million people" with "gamers and consumers being left behind" due to cost factors.
- The Radeon RX 480 is being promoted as being competitive with $500 current generation cards like the GTX 980 and R9 390X. There's 4 or 8GB of GDDR5 memory, 150 Watt TDP, AMD FreeSync, premium VR support, 36 compute units, and 256GB/s of memory bandwidth. The RX 480 is said to have 1.7x performance/Watt improvement.
- AMD now talking up Vulkan, particularly about Doom on Vulkan.
- 7th Gen AMD A-Series APU announced.- HEVC / H.265 support heavily being talked up.- 7th Gen AMD FX 9800P @ 15 Watts is 28% faster for LibreOffice (yes, the LO icon was shown), 44% faster for compression, 43% faster for Photoshop.- AMD A9 series APUs are designed for sub-$400 notebook PCs.- Blah... Lots of Windows stuff happening.
- Finally, Zen is being talked about! (53 minutes in for those watching the live stream later on.)
- The Zen Summit Ridge processor was shown off. Still in "early stages of bring-up" but "the product looks good, really good." 40% IPC improvement being talked about, 8 cores / 16 threads, new AM4 desktop platform.
- Zen will sample to priority customers in "few weeks" and roll out in Q3.
| 99
| 1,760,719,792.228737
|
https://www.phoronix.com/news/Radeon-RX-480-199
|
AMD May Sell Its Radeon RX 480 "Polaris" For Just $199 USD
|
Michael Larabel
|
Just hours to go until AMD's Computex live-stream, details are being leaked out about what's expected. From what we're hearing so far, AMD is going to undercut their prices of Polaris 10 hugely: the Radeon RX 480 is said to be priced retail at $199 USD and will compete with the likes of a GeForce GTX 970~980.
If AMD can ship a card in June that costs under $200 and compete with a GTX 970~980, simply wow! Plus packed with the already out there open-source Linux driver support. This Polaris 10 card won't be able to compete with NVIDIA's new GTX 1070/1080 "Pascal" cards (AMD's higher-end offerings will come out later this year with "Vega"), but for delivering performance-per-dollar this could be quite an attractive offering if these initial reports are true. Plus there's the open-source AMDGPU Polaris support compared to no open-source driver support yet for Pascal. NVIDIA did supply a GTX 1080 card to Nouveau's lead developer, Ben Skeggs, but it still could be months before NVIDIA releases the necessary signed firmware images for allowing accelerated hardware support.
Other leaked details say the Radeon RX 480 will have 8GB of GDDR5 memory, up to 5.5 TFLOPS of compute power, 256-bit memory interface, full HEVC support, DisplayPort 1.4 support, async compute supported, and other Polaris 10 details that have already been speculated. The $199 alleged pricing was revealed by the WSJ a short time ago.
Stay tuned and I'll certainly be watching the AMD Computex live-stream.UPDATE: The $199 price confirmed, more details here.
| 22
| 1,760,719,792.967229
|
https://www.phoronix.com/news/AMD-Polaris-Live-Tomorrow
|
Get Ready For AMD's Livestream Tomorrow Night
|
Michael Larabel
|
Mark your calendars for tomorrow to watch AMD's Computex livestream for Polaris.
There's only one day to go until more AMD Polaris details will be revealed in a videostream live from Computex in Taipei. If you haven't already done so, you can register here to be reminded of the stream on 31 May. More details via the streaming page.
The stream is scheduled to begin on 31 May at 10PM EST -- an inconvenient time at least for many in the US, due to the event taking place in Taiwan.
I'll certainly be watching the stream. As some of you may have heard already, this past weekend AMD held their press event for Polaris in Macau. This time I was invited (last time AMD sent me to one of these press launch events was for RV770...) but unfortunately had to respectfully decline given my constantly busy schedule with not taking a day off in years between Phoronix and PTS, etc. While AMD was going to pay for the flights/hotel, I also wasn't too motivated to fly ~20+ hours to China to listen to them talk about Windows gaming and other non-Linux material for three days. But I mention this as hopefully that means I'm on the list and will be able to provide launch-day Polaris 10 Linux coverage this time around. While that appears to be the case right now, even if that's not the case, I'll definitely be buying the available Polaris 10 GPUs for Linux testing and to compare to my mass assortment of other hardware. Between Polaris and the GTX 1070/1080 Linux testing (plus the Phoronix 12th birthday next week!), June should be a very interesting month.
| 18
| 1,760,719,793.737439
|
https://www.phoronix.com/news/AMD-Compress-Open-Source
|
AMD Open-Sources Compressonator Project
|
Michael Larabel
|
The latest open-source fruits of AMD's GPUOpen project is the Compressonator.
AMD Compressonator is also known formerly as just "AMD Compress" and was a closed-source compression tool available to developers. Compressonator is about compressing textures and creating mip-map levels. The tool is designed around DXTn/S3TC.
More details on Compressonator going open-source were announced today via GPUOpen.com. The code is hosted on GitHub. The project up to now has been focused upon Microsoft Windows but with now being open-source we will hopefully see it ported to Linux for helping game developers/porters.
| 11
| 1,760,719,794.989968
|
https://www.phoronix.com/news/AMD-FireRays-2.0-Released
|
AMD Open-Sources FireRays 2.0 Ray-Tracing Library
|
Michael Larabel
|
The latest project AMD is open-sourcing is FireRays.
FireRays is a "ray intersection acceleration library provided by AMD which makes the most of AMD hardware and allows for efficient ray queries independently of the generation of underlying AMD hardware."
The FireRays 2.0 release works with a variety of low-level APIs including OpenCL, Embree, and others. Vulkan support for FireRays is currently a work-in-progress. This library works with Windows, OS X, and Linux. Not only are AMD GPUs supported but also NVIDIA/Intel GPUs and various CPUs too.
More details on the AMD FireRays 2.0 release can be found via GPUOpen.com. The code for FireRays is on GitHub.
| 7
| 1,760,719,795.420089
|
https://www.phoronix.com/news/ATI-DRI3-Default-On
|
DRI3 Is Now Flipped On For xf86-video-ati X.Org Driver
|
Michael Larabel
|
There's a follow-up to yesterday's story about AMD To Enable DRI3 By Default On Latest X.Org Servers.
AMD's Michel Dänzer didn't get any push-back to his patch for enabling DRI3 by default in the xf86-video-ati driver, so he's gone ahead and landed the patch in the Radeon DDX driver code-base. This now means that if you're building this X.Org driver against X.Org Server 1.18.3 or newer, you'll have DRI3 enabled by default rather than DRI2.
DRI3 should reduce/eliminate tearing and boost the performance for select graphics workloads, but regressions are very possible. Hopefully other drivers will work on enabling DRI3 by default and help ironing out any lingering DRI3 bugs within the xorg-server. You can fetch the latest Radeon DDX driver Git here.
| 4
| 1,760,719,796.283399
|
https://www.phoronix.com/news/AMD-DRI3-Default-Latest-X
|
AMD To Enable DRI3 By Default On Latest X.Org Servers
|
Michael Larabel
|
Michel Dänzer of AMD has published a patch today that would enable DRI3 by default for the xf86-video-ati X.Org driver.
Direct Rendering Infrastructure 3 has proven to be able to deliver faster performance than DRI2 (the default) while reducing tearing, but hasn't been enabled by default due to various bugs in the X.Org Server's implementation. However, things have been in rather good shape if using the latest X.Org Server 1.18 point release, thus driver maintainers are finally back to re-evaluating their DRI defaults.
The patch today by Dänzer would enable DRI3 by default if it's being built against xorg-server 1.18.3 or newer. The patch for now is floating on the xorg-driver-ati mailing list.
If you want to manually enable DRI3 for your driver to get a better experience or reduce the OpenGL performance in select workloads, simply add Option "DRI" "3" to your xorg.conf/x.org.conf.d -- assuming you're on a newer X.Org Server. Hopefully the other DDX maintainers will also resort to re-enabling DRI3 by default too.
| 15
| 1,760,719,796.855429
|
https://www.phoronix.com/news/AMD-DAL-Working-For-Linux-4.7
|
Developers Still Hoping For AMD DAL Support In Linux 4.7
|
Michael Larabel
|
Open-source developers working on the Radeon Linux graphics driver stack remain hopeful that their massive "DAL" code-base will be ready for merging with Linux 4.7.
DAL is the massive proposed addition to the AMDGPU DRM driver and is a lot of code opened up from the Catalyst proprietary driver. This display abstraction code is AMD's approach for implementing atomic mode-setting, DP MST, HDMI 2.0, better PowerPlay, better multi-display support, etc. It will also allow them to hopefully implement FreeSync support in their Linux driver.
When the DAL code was published, it received some criticism by DRM developers outside of AMD for its massive size, adding more complexities and abstractions to the code, not following all of the mainline kernel development practices, etc. But they hope to have this DAL code readied for the Linux 4.7 kernel.
They are in a bit of a rush to get DAL merged since the open-source Polaris GPU driver support for the next-generation hardware depends upon it. So far they continue making progress.
The very latest DAL code for testing can be found via the drm-next-4.7-wip-dal branch. Hitting the mailing list this weekend are also more DAL patches for review. Hopefully everything will get tidied up in time for Linux 4.7: stay tuned.
| 23
| 1,760,719,797.890243
|
https://www.phoronix.com/news/AMD-Secure-Memory-Encryption
|
AMD Posts Secure Memory Encryption For The Linux Kernel (SME)
|
Michael Larabel
|
Well, today seems to be the day for x86 CPU vendors to push out memory security related features for the Linux kernel... After Intel posted the Secure Guard Extensions driver for Linux, AMD has come out with a patch-set for "Secure Memory Encryption" (SME) that looks like it will be a hardware feature of Zen.
I hadn't heard much about AMD's Secure Memory Encryption (SME) feature up to now and some quick searching isn't turning up too much, but presumably is a feature for Zen. Tom Lendacky of AMD describes SME as:
SME can be used to mark individual pages of memory as encrypted through the page tables. A page of memory that is marked encrypted will be automatically decrypted when read from DRAM and will be automatically encrypted when written to DRAM. Details on SME can found in the links below.
The SME feature is identified through a CPUID function and enabled through the SYSCFG MSR. Once enabled, page table entries will determine how the memory is accessed. If a page table entry has the memory encryption mask set, then that memory will be accessed as encrypted memory. The memory encryption mask (as well as other related information) is determined from settings returned through the same CPUID function that identifies the presence of the feature.
The approach that this patch series takes is to encrypt everything possible starting early in the boot where the kernel is encrypted. Using the page table macros the encryption mask can be incorporated into all page table entries and page allocations. By updating the protection map, userspace allocations are also marked encrypted. Certain data must be accounted for as having been placed in memory before SME was enabled (EFI, initrd, etc.) and accessed accordingly.
This patch series is a pre-cursor to another AMD processor feature called Secure Encrypted Virtualization (SEV). The support for SEV will build upon the SME support and will be submitted later. More details on Secure Memory Encryption and Secure Encrypted Virtualization from AMD can be found via this 12-page PDF whitepaper that was just published a few days ago. This SME support for the Linux kernel affects just over one thousand lines of x86 kernel code and until being merged into a future version of the Linux kernel can be found via this patch series.
| 28
| 1,760,719,798.252265
|
https://www.phoronix.com/news/AMD-Updates-CZ-UVD-Blob
|
AMD Updates Carrizo Firmware To Support More UVD Sessions
|
Michael Larabel
|
For those running a Linux system powered by an AMD "Carrizo" APU, there's an updated firmware blob out today to benefit your UVD video decoding experience.
Landing today in the linux-firmware Git tree is an updated binary-only Carrizo UVD firmware file.
The lone reported change is that this updated Unified Video Decoder firmware now supports up to 40 sessions. This should be useful if you're doing any large-scale GPU-accelerated video decoding on a Carrizo system. The AMD open-source UVD support is in turn exposed via the VDPAU Gallium3D state tracker.
| 10
| 1,760,719,799.2499
|
https://www.phoronix.com/news/AMD-CodeXL-2.0-Open-Source
|
AMD CodeXL 2.0 Released, Now Is Open-Source
|
Michael Larabel
|
AMD this afternoon announced CodeXL 2.0 as the newest version of their GPU debugger, CPU/GPU profiler, and static kernel analyzer. CodeXL 2.0 is a big leap forward and is now open-source!
AMD has made the CodeXL program open-source except for some parts they didn't open as they were "IP confidential" according to the company. In making it open-source, AMD explained, "CodeXL, previously a tool developed as closed-source by Advanced Micro Devices, Inc., is now released as Open Source. AMD believes that adopting the open-source model and sharing the CodeXL source base with the world can help developers make better use of CodeXL and make CodeXL a better tool. To encourage 3rd party contribution and adoption, CodeXL is no longer branded as an AMD product. AMD will still continue development of this tool and upload new versions and features to GPUOpen."
CodeXL 2.0 adds a frame analysis mode, host code debugging with the Linux version, support for Vulkan GLSL shaders with the analyze mode, improvements to the power profiling mode, the HSA support in CodeXL now supports the Boltzmann driver, and various other improvements.
You can learn more about CodeXL 2.0 via the announcement at GPUOpen.com. The code to the Windows/Linux program is available via GitHub.
| 13
| 1,760,719,799.703766
|
https://www.phoronix.com/news/AMD-Polaris-No-ETC2
|
AMD Polaris Doesn't Support ETC2 Texture Compression
|
Michael Larabel
|
Well this is surprising and unfortunate: it appears that AMD's next-generation Polaris GPUs don't support the royalty-free ETC2 texture compression.
Recent Intel hardware supports ETC2 and NVIDIA's latest hardware also supports ETC2. On the AMD side, the forthcoming "Stoney" APUs brought ETC2 support to the open-source driver stack.
While the forthcoming AMD Stoney APUs have ETC2 support, the upcoming desktop "Polaris" GPUs do not. This patch today by AMD's Marek Olšák for the RadeonSI Gallium3D driver no longer exposes ETC2 on Polaris and his commit message reads "not supported."
It's sad that AMD desktop GPUs aren't yet offering ETC2 where as other vendors are, ETC2 is supported with the latest OpenGL and OpenGL ES specifications, and ETC2 along with ASTC are the most prominent hopes for widespread, royalty-free texture compression to replace the notorious S3TC texture compression.Update: Marek at AMD would like to clarify that while the Polaris hardware doesn't natively support ETC2, Gallium3D drivers have a software fallback for converting ETC2 compressed data to other formats supported by the GPU.
| 34
| 1,760,719,800.875355
|
https://www.phoronix.com/news/AMD-DAL-Hopes-For-Linux-4.7
|
AMD Is Hoping To Upstream DAL For Linux 4.7, But A Lot Of Work Is Needed
|
Michael Larabel
|
Along with open-sourcing the next-gen Polaris GPU driver code yesterday, Alex Deucher of AMD laid out their plans for aiming to get the DAL display code into the Linux 4.7 kernel.
This Polaris driver code is dependent upon the massive AMDGPU DAL patches for display handling that amount to over 93 thousand lines of code (and the Polaris code is another 67 thousand lines of code). When that DAL code was sent out last month, it was quick to be critiqued by other upstream DRM driver developers.
With that Polaris support dependent upon DAL, AMD needs to get this massive patch series into the Linux 4.7 kernel. The Upstreaming DAL mailing list message covers their approach. Alex explained, "When we initially released the patches, there were some rough edges and quite a few things were pointed out that need to be fixed. Some are relatively easy to fix, others will take time. Our goal is to make those changes. We'd like to target 4.7 for upstreaming DAL. To that end, I think it would be easier to review and track our progress if I maintained a public DAL branch and send out patches against that branch rather than respinning giant squashed patches every time we fix something. It's much harder to track what progress has been made. DAL is currently at ~400 patches."
Among the changes aimed for Linux 4.7 to DAL include removing various malloc/free/sleep/delay wrappers, switching to Linux i2c, using the Linux DRM AUX interface, converting cursors to the planes API, and native DRM EDID fetching. Later on they will take care of other work.
However, as of yet, DRM subsystem maintainer David Airlie is still uneasy about this code. David responded to Alex with, "I think people are focusing on the minor comments and concerns and possibly deliberately ignoring the bigger concern that this code is base is pretty much unmergeable as-is. Cleaning this up won't be a matter of just removing some memset's, mallocs and moving on. This code needs redesign by someone with Linux kernel development experience. Alex if you have any other tasks in AMD, I suggest you apply to have them all scrapped and you take this on board."
Airlie went on to add, "So much of this code seems to be 'architected' design. I mean someone senior draws a bunch of powerpoint slides with nice blocks, with what are abstraction layers between them, then a bunch of other coders take that powerpoint slide and work on a box each until it all comes together. However the abstractions don't survive so well and you end up with a bunch of leaky boxes all co-dependent and joined together but still abstracted. So let's take a massive step back from the edge here and figure out what we expect a modesetting codebase for a driver in the Linux kernel to look like and write that...I'm actually nearly getting to the point of realising nobody actually
understands my concerns here and just trying to write a driver on my own. I'm still slightly open to some sort of STAGING for this, but I think I'm nearly at the level, that I'd only accept that on the understanding we'd try and evolve the driver to this level, rather than think we can clean DAL to the kernel standards."
We'll see what happens in the weeks ahead for this code that's needed for Polaris support. DAL supports AMDGPU hardware and provides reliable DisplayPort support, HDMI 2.0 support, DVI support, full integration with power management, Atomic KMS, proper 4K @ 60Hz support, and other display-related features for this code derived from AMD's internal code-base.
| 100
| 1,760,719,801.491847
|
https://www.phoronix.com/news/AMD-Open-Sourced-Polaris
|
AMD Publishes Initial Open-Source Driver Code For Next-Gen Polaris
|
Michael Larabel
|
One week after the surprise of delivering a beta of their new hybrid "PRO" driver stack, here's another big surprise: AMD has just published the initial open-source code for driver support with their upcoming "Polaris" graphics processors!
Hitting the web today are the AMDGPU kernel patches for the Polaris hardware. There are 52 kernel patches for their AMDGPU DRM driver to support graphics, UVD video decoding, VCE video encoding, power management, and display support for these next-generation GPUs. The patches were published by AMD's Alex Deucher.
The initial Baffin PCI IDs include 0x67E0, 0x67E1, 0x67E8, 0x67E9, 0x7EB, and 0x7FF. The Ellesmere IDs are 0x67C0 and 0x67DF.
This support for the yet-to-be-released Polaris GPUs is dependent upon the DAL display code that has yet to be mainlined. Polaris includes the ELLESMERE and BAFFIN chip families. It's great to see power management included in the initial support code and not being held off for later as was unfortunately the case with the earlier GCN hardware when bringing up the original AMDGPU driver.
It's too late for this Polaris support to be merged into Linux 4.6, so Linux 4.7 will be the next opportunity for the code to land. Hopefully it will be all reviewed and ready to merge by that time -- and that the DAL will be ready too since the early DAL open-source code was criticized by upstream developers.
The early AMDGPU Polaris patches can be found here. Polaris support adds more than 67,000 lines of code to the kernel.
Polaris is likely to be marketed as the Radeon Rx 400 series and will be manufactured on a 14nm FinFET process, the higher-end hardware has HBM2, and there will be HDMI 2.0a and DisplayPort 1.3 display support. Polaris hardware should be a big advancement when it comes to power efficiency over earlier Radeon GPUs. Polaris hardware should begin shipping by this summer.
| 69
| 1,760,719,802.182281
|
https://www.phoronix.com/news/AMD-XConnect-Thunderbolt-Gfx
|
AMD Rolls Out XConnect For External Graphics Via Thunderbolt 3
|
Michael Larabel
|
AMD this morning unveiled their XConnect technology to connect an external, high-performance Radeon graphics card to a notebook or 2-in-1 device.
XConnect relies upon Intel's Thunderbolt 3 specification for allowing notebooks and other portable devices to benefit from an external graphics card if wishing to ramp up the device's performance for gaming. The enclosure for the graphics card is said to be "as simple as using a USB flash drive."
AMD has been developing XConnect in conjunction with Razer and the Thunderbolt working group. Razer's Blade Stealth is the first notebook supporting the XConnect technology in conjunction with R9 300 series graphics cards.
Those wishing to learn more about XConnect can visit AMD.com/xconnect. For now though XConnect appears limited to supporting Windows 10 with no signs of near-term Linux support.
| 24
| 1,760,719,802.9751
|
https://www.phoronix.com/news/AMD-LeMaker-Cello
|
An AMD ARM 64-bit Dev Board Is Launching For $299 USD
|
Michael Larabel
|
Since last year we have been waiting for AMD to launch their "HuskyBoard" ARM development board built around their Opteron A1100 ARM 64-bit SoC. That board was originally supposed to ship in Q4'15 while now available for pre-order is a new A1100 development board that looks like it may be taking its place.
AMD had been teasing their ARM development board for nearly one year and talked of it being a low-cost ARM development board that would be in compliance with the 96Boards' Enterprise Edition specification.
There still is sadly nothing to report on the availability of HuskyBoard and am concerned that AMD may have changed their plans. Today we're hearing of the LeMaker Cello as a A1100 development board similar to the HuskyBoard. The LeMaker Cello has an AMD Opteron A1120 quad-core Cortex-A57 processor running at 1.7GHz, 2x DDR3 SO-DIMM slots, two Serial ATA 3.0 ports, Gigabit Ethernet, two USB 3.0 ports, one PCI Express x16 3.0 slot, etc. The AMD A1120 SoC has a 25 Watt TDP so it will also need active cooling. The LeMaker Cello should be able to mount in a mini-ITX / micro-ATX chassis.
This first lower-cost development board built around an AMD ARM SoC is set to retail for $299 USD and will begin shipping in Q2. $299 USD is a decent target for this board and is cheaper than the Jetson TX1 albeit this SoC doesn't have any integrated graphics like the beefy Maxwell on the Tegra X1.
Hopefully I'll be able to get a review sample of the LeMaker Cello for benchmarking on Phoronix or some Phoronix readers will be able to help out with their support (or view the site without ad-blockers if you are unable to join Premium or tip) so will be able to make a budget for one as it should be an interesting ARM Linux test target. You can pre-order from Lenovator.com.
| 28
| 1,760,719,803.836024
|
https://www.phoronix.com/news/AMD-Vulkan-Linux-Soon
|
AMD Representative Says Their Vulkan Linux Driver Will Be Here Soon
|
Michael Larabel
|
A technical marketing representative of AMD / Radeon Technologies Group is participating in a Reddit "Ask Me Anything" today where he's been answering a few Linux questions.
While he's answered some Linux questions, nothing has come out as a shocker. Some highlights:
- In response to one comment about the Vulkan Linux driver, he says it's coming "quite soon". But in response to another question about the Vulkan Linux driver, he reported, "The Vulkan Linux driver will be released soon." Then again, 'soon' and 'quite soon' is not really surprising. AMD has a working Vulkan Linux driver, they just haven't made it public yet.
- As I reported months ago, there still are no plans for AMD/RTG to support their Vulkan driver outside of the AMDGPU kernel realm. Their Vulkan driver is designed for AMDGPU. "Our Linux Vulkan drivers will only run on the amdgpu kernel driver. We have no plans, on Linux, to support Vulkan on any other driver stack."
- There's still no word when in the future the AMD Vulkan Linux driver will be open-sourced.
You can read the full AMA and answers here.
| 167
| 1,760,719,804.700169
|
https://www.phoronix.com/news/AMD-G3D-HSA-CL-Interop
|
AMD Sends Out Big Patch Series For HSA/OpenCL Interop Support
|
Michael Larabel
|
AMD's Marek Olšák sent out a set of 26 patches this morning for preparing the RadeonSI Gallium3D driver to have interoperability support between OpenGL and HSA/OpenCL.
With the patches, the groundwork is laid for sharing more assets between OpenGL and OpenCL/HSA. Currently, this interoperability covers buffer and texture sharing.
This is continuing from the work talked about a few weeks ago with the broader issue of AMD Is Looking At A Interoperability Interface For OpenCL Outside Of Mesa. That public interface would be present in the DRI state tracker, libGL, and libEGL.
More details on this RadeonSI interop prep work for OpenGL/OpenCL sharing of buffers/textures can be found via this patch series. Hopefully all of the code will be prepped in time for the Mesa 11.3 release in about three months.
| 17
| 1,760,719,805.902685
|
https://www.phoronix.com/news/AMD-OpenVX-Beta
|
AMD Publishes OpenVX AMDOVX Open-Source Beta
|
Michael Larabel
|
The latest fruits of AMD's GPUOpen initiative is the open-sourcing of a beta of AMDOVX.
AMDOVX is short for AMD OpenVX as an open-source implementation of the Khronos Group's OpenVX computer vision specification. AMDOVX works on both x86 CPUs and GPUs via OpenCL. AMDOVX is supported on both Windows and Linux while providing interoperability with OpenCV. This open-source code does depend upon the AMD APP SDK 3.0 release.
Details on the new AMD OpenVX implementation can be found via GPUOpen.com while the newly-opened code (marked as v0.9 beta) was open-sourced this week to this GitHub repository.
| 2
| 1,760,719,806.507235
|
https://www.phoronix.com/news/AMD-FreeSync-DAL-Plans
|
It Looks Like AMD Will Support FreeSync With Their New Linux Display Stack
|
Michael Larabel
|
While NVIDIA has long supported G-SYNC on Linux as their adaptive sync technology for eliminating screen tearing, AMD hasn't supported their FreeSync tech via their open or closed-source Linux drivers. Fortunately, it's looking like that will change.
FreeSync is designed to reduce tearing, help battery life, and more via dynamically adapting the display refresh rate as needed. There are more and more FreeSync-supported displays hitting the market and FreeSync is backed by VESA as an option in the DisplayPort 1.2a specification. AMD Linux customers have been missing out on this feature, but it appears as part of the new AMDGPU DAL driver where they are working towards feature display parity with the proprietary driver, FreeSync is on the TODO list.
In looking to address the concerns by upstream developers about the massive size and complexity of DAL, AMD's Harry Wentland described the architecture a bit more. Here's the main part of his message:
The goal with DAL is to provide a unified, full featured display stack to service all of our Linux offerings. This driver will have to support our full feature set beyond what's supported by amdgpu, e.g.
- synchronzied timings across different displays
- freesync
- solid support of 6 displays in any configuration (HDMI, DVI, DP, DP MST, etc)
- solid support of 4k at 60 timings on APUs
- power features, such as
- clock-accurate bandwidth formulas
- improved interaction with powerplay to maximize power savings
- Improved audio and other infoframe related features
- Improved stability with powerplay since display hw is involved in the SMC hw interactions and improper programming sequences can lead to GPU hangs, etc.
The current amdgpu display stack grew somewhat organically and as such is not well suited to handling all of the hardware dependencies involved especially in areas like audio. The drm abstractions used by the old code map less and less well to new hw pipelines. Atomic helps, but if we are going to convert, it seemed like a good time to start fresh.FreeSync is mentioned along with other exciting work like synchronized timings across displays, better multi-monitor support, improved stability, and more power management related features! Let's hope this all pans out in a timely manner.
He also explained more with their reasoning for some of the code abstraction, how their display core doesn't always map well to the DRM interfaces, the DAL internal abstractions match what is used by the AMD drivers on other operating systems, and how they can further trim this ~93k lines of code (there's a lot of blank lines and code comments).
| 37
| 1,760,719,807.573845
|
https://www.phoronix.com/news/Massive-DAL-Driver-Woes
|
AMD's Massive New DAL Display Driver Is Facing An Uphill Battle
|
Michael Larabel
|
We weren't the only ones surprised by the massive size of the new AMDGPU DAL display driver. Weighing in at 93k lines of code, upstream Linux graphics developers outside of AMD are also stunned by the size of this code-base for handling display-related functionality with the AMDGPU driver for Tonga / Carrizo / Fiji.
This DAL driver code for the AMDGPU DRM driver that was sent out yesterday is for bringing the open-source driver's display support up to near parity with what was in the Catalyst driver. In fact, given the copyrights and briefly looking at the code, it looks like a fair amount of this code was ported over from the AMD Catalyst Linux driver. With just this display-related code weighing in at nearly 94 thousand lines of code is huge: from back in October, you can see a look at the different DRM driver sizes. This AMD display code is larger than the smaller DRM drivers and almost equal to the size of the entire Intel DRM driver.
The first response to yesterday's DAL patches was by DRM subsystem maintainer David Airlie. If Airlie won't accept the code into the DRM tree, it's not getting into the Linux kernel. He commented:
So the first minor criticism is this patch doesn't explain WHY.
Why does the Linux kernel need 93k lines of code to run the displays when whole drivers don't even come close.
We've spent a lot of time ripping abstraction layers out of drivers (exynos being the major one), what benefits does this major change bring to the Linux kernel and the AMDGPU driver over and above a leaner, more focused work.
If were even to consider merging this it would be at a guess considered staging level material which would require a TODO list of major cleanups.
I do realise you've put a lot of work into this, but I think you are going to get a lot of review pushback in the next few days and without knowing the reasons this path was chosen it is going to be hard to take.
Dave. Intel DRM maintainer Daniel Vetter was also taken away by the massive patch set:
Yeah agreed, we need to figure out the why/how first. Assembling a de-staging TODO is imo a second step. And the problem with that is that before we can do a full TODO we need to remove all the os and drm abstractions. I found delayed_work, timer, memory handling, pixel formats (in multiple copies), the i2c stuff Rob noticed and there's more I'm sure. With all that I just can't even see how the main DAL structures connect and how that would sensibly map to drm concepts. Which is most likely needed to make DAL properly atomic.
So de-staging DAL (if we decided this is the right approach) would be multi-stage, with removal of the abstractions not needed first, then taking a 2nd look and figuring out how to untangle the actual concepts.
Aside: If all this abstraction is to make dal run in userspace for testing or whatever - nouveau does this, we (Intel) want to do this too for unit-testing, so there's definitely room for sharing the tools. But the right approach imo is to just implement kernel services (like timers) in userspace.
Another thing is that some of the features in here (hdmi 2.0, improved dongle support) really should be in shared helpers. If we have that hidden behind the dal abstraction it'll be pretty much impossible (with major work, which is unreasonable to ask of other people trying to get their own driver in) to extract&share it. And for sink handling having multiple copies of the same code just doesn't make sense.
Anyway that's my quick thoughts from 2h of reading this. One wishlist: Some design overview or diagram how dal structures connect would be awesome as a reading aid.
-Daniel Daniel also followed up by saying he found other DAL code that is redundant compared to shared functionality already available to DRM drivers like some EDID handling, its own infoframe encoding, and home-grown logging. Rob Clark also pointed out duplications around the i2c code. Presumably much of this duplication (and extra abstractions) occurred due to porting the code from the Catalyst driver as opposed to building the code up from scratch around the DRM interfaces.
It will be interesting to see how quickly the AMD developers will be able to efficiently whip the DAL code into shape, but I'd hazard to guess that it won't be ready for Linux 4.6.
| 43
| 1,760,719,808.21497
|
https://www.phoronix.com/news/Catalyst-XCOM-2-Linux
|
AMD Catalyst Appears To Work With XCOM 2 On Linux
|
Michael Larabel
|
Not only does RadeonSI Gallium3D work with XCOM 2 on Linux for AMD graphics processors, but it looks like the Catalyst (or now known as Radeon Software, officially) too works with this brand new, highly anticipated strategy game seeing a same-day release across OS X / Linux / Windows.
While we all know the NVIDIA Linux driver works fine with this game, after finding out that it's possible to get the Mesa/Gallium3D drivers working, my next focus was on seeing whether the Catalyst binary blob would work for this game. Surprisingly, it appears to!
The game fired up fine on Ubuntu 15.10 with the packaged fglrx/Catalyst driver. In the few minutes of initiating a new game, I didn't notice any apparent rendering issues.
So at least the game appears to be playable, but I can't confirm the entire game works fine on Catalyst. Unfortunately with XCOM 2 lacking a benchmark mode, that's to the extent of this round of testing and about as far as I'll get with this game.
Eric Griffith will be sharing his thoughts on XCOM 2 gameplay under Linux in the days ahead.
| 10
| 1,760,719,809.105671
|
https://www.phoronix.com/news/AMD-Iceland-Now-Stable
|
Open-Source AMD Iceland/Topaz Support No Longer Considered Experimental
|
Michael Larabel
|
The AMDGPU DRM driver support for Iceland (Topaz) graphics processors is now considered stable with the experimental flag set to be removed.
Last year AMD marked the Iceland support as experimental due to not having tested the new GPUs much and some bugs were coming in regarding the AMDGPU DRM driver's support for the mobile GPUs.
Fast forward to today, Alex Deucher posted a series of eight AMDGPU patches for Iceland/Topaz and at the end they remove the experimental support flag. Thus the hardware-accelerated open-source support will now be available without the amdgpu.exp_hw_support=1 kernel module parameter. The fixes done prior to re-enabling the out-of-the-box graphics support appear to be mainly related to the firmware handling for this family of graphics processors.
Unless this patch series gets sent in today as "fixes" for Linux 4.5, this experimental support flag won't be removed until Linux 4.6.
Iceland/Topaz is found in the Radeon R5 M255 and R7 M260 mobile graphics processors. The PCI IDs listed by this family of GPUs are 0x6900, 0x6901, 0x6902, 0x6903, and 0x6907.
| 24
| 1,760,719,809.584933
|
https://www.phoronix.com/news/AMDGPU-sisched-Test-1
|
Testing The LLVM SI Machine Instruction Scheduler
|
Michael Larabel
|
Landing last month in the LLVM SVN/Git code-base was the SI machine scheduler for the AMDGPU LLVM back-end. This scheduler has the potential to improve the performance for some hardware/workloads, but not by the wide margins originally reported by some early testers.
While the SI machine scheduler has been in the LLVM back-end, landing in Mesa Git a few days ago was an option for easily enabling it.
Assuming you are building against the latest LLVM 3.8 or LLVM 3.9 SVN code, the R600_DEBUG=sisched option on Mesa Git can be used for turning on the LLVM SI machine scheduler.
While it's on my TODO list for running some of these SI machine instruction scheduler tests, a Phoronix reader has already run a few tests on his R9 Fury in conjunction with the Linux 4.5 development kernel.
These SI machine scheduler tests have already been discussed within our forums, but if you missed these preliminary independent results, you can find them via this OpenBenchmarking.org result file. However, at least for this user's system, the sisched option doesn't make a huge difference. I'll run some of my own tests on multiple GPUs when time allows.
| 6
| 1,760,719,810.516878
|
https://www.phoronix.com/news/Two-Kaveris-And-An-Excavator
|
AMD Releases Two New Kaveri APUs & An Excavator-Based CPU
|
Michael Larabel
|
AMD this morning updated its Kaveri APU line-up and also released a new Athlon X4 processor.
The new APUs are the A10-7860K quad-core and A6-7470K dual-core parts. These are part of Kaveri/Godavari with the A10-7860K coming in with a 3.6GHz base clock speed and 4.0GHz boost clock speed while having a 65 Watt TDP. The A6-7470K has a 3.7GHz clock speed with 4.0GHz boost while also being a 65 Watt TDP.
Those APUs are with Steamroller cores, but AMD also released the Athlon X4 845. The Athlon X4 845 is a quad-core Excavator design, a.k.a. the same cores found in Carrizo. The four Excavator cores run at 3.5GHz with a 3.8GHz boost clock speed and also is a 65 Watt TDP part. There's just 2MB of L2 cache for this new Athlon X4.
The Athlon X4 845 retails for just $70 USD. I'm actually quite tempted to buy one (AMD hasn't offered any review samples to us yet...) as it should be fun for some Linux benchmarking, some compiler testing for the Excavator cores, etc. If you are interested in seeing Athlon X4 845 Linux benchmarks on Phoronix, voice your support via Twitter or the forums as I will decide shortly as soon as I see this processor in stock at Internet retailers.
With the new hardware also comes AMD's new Wraith Cooler, which is a much more efficient stock heatsink than AMD's earlier designs.
| 12
| 1,760,719,811.082068
|
https://www.phoronix.com/news/NIISI-Open-AMD-Results
|
Russian Super-Computing Users Get Tired Of Catalyst, Start Looking At Open-Source AMD
|
Michael Larabel
|
Super-computing researchers part of the Institute of System Research for the Russian Academy of Sciences recently presented on using the open-source Radeon driver for OpenCL.
While inquiring this morning about some fresh Radeon OpenCL benchmarks on Phoronix, one of the researchers shared their slide-deck from a presentation they did in November that included some initial results. Their slide covers the "AMD OpenCL and Catalyst Linux driver curse" but they consider "a new hope with Mesa OpenCL."
They feel AMD is weak in the high-performance computing (HPC) space over "lack of high quality working software development model and system software" and "lack of application development libraries." But there's newfound hope with AMD's more open-source nature and moving toward their unified AMDGPU driver approach.
This presentation at Russia's 4th National SuperComputing Forum listed among the "Linux AMD OpenCL curse" issues as the inability to easily install their drivers, runtime segmentation faults, freezing, weak OpenCL implementation, OpenCL C compiler errors, and regressions / changes in behavior between driver versions. It's got to the point they feel the Catalyst driver has become "impossible to operate."
They are looking "as soon as possible" to move off Catalyst and switch to AMDGPU while desiring Hawaii support. They are also hoping that Mesa OpenCL will improve with this open-source implementation having all of the features of AMD's proprietary OpenCL runtime.
OpenCL benchmarks carried out by the organization show that the Catalyst Linux driver with the "clpeak" test is 4% to 58% faster than the current Mesa stack. With an OpenCL FFT Radix7 benchmark, the Catalyst driver is 40% to 237% faster than Mesa.
More details can be found via their PDF slide deck.
| 31
| 1,760,719,811.942321
|
https://www.phoronix.com/news/AMD-Boltzmann-Guide
|
AMD's Guide To Using Boltzmann ROCK/ROCR & HCC On Linux
|
Michael Larabel
|
Last week AMD launched GPUOpen and began shipping their new and open code. Today the company has published a guide for taking advantage of the Boltzmann stack with their Radeon Open eCosystem Kernel and Runtime.
The post published today by AMD's Adrian Edwards covers the process of installing and configuring the ROCK kernel driver, setting up the ROCR runtime, and then setting up the Heterogeneous Compute Compiler (HCC). From the HCC you can then compile code using the HSA API and program HSAIL kernels, support for C++ AMP 1.2, or utilize AMD's HIP API for porting CUDA code-bases over to Boltzmann for AMD hardware.
The guide is catered to Ubuntu and Fedora users. For learning how to setup the Boltzmann stack, visit this GPUOpen.com page.
| 17
| 1,760,719,812.642969
|
https://www.phoronix.com/news/R9-Fury-Linux-4.5-rc2-Test
|
R9 Fury Performance Gains With Linux 4.5-rc2? I'm Not So Lucky
|
Michael Larabel
|
Since yesterday's release of Linux 4.5-rc2 there was at least one report of AMDGPU performance improvements with Linux 4.5-rc2 for an R9 Fury "Fiji" graphics card.
When reading this morning of "double or quadruple the framerates that I got with RC1" for an R9 Fury owned by a Phoronix reader, I immediately set out to run some R9 Fury benchmarks on Linux 4.5-rc2 compared to my 4.5-rc1 results last week and compared to Catalyst. I also did the same for an R9 285 Tonga on AMDGPU as well for reference purposes.
Sadly, the performance didn't improve at all in my conditions for Linux 4.5-rc2 with the R9 Fury nor with the R9 285 (but at least with the R9 285, it's been performing well already compared to Catalyst speeds). Yes, PowerPlay was enabled. Yes, the latest firmware files were used.
The R9 Fury also still failed with Unigine Valley and Xonotic with image quality settings.
You can dig through all of the R9 Fury AMDGPU results via this new OpenBenchmarking.org result file.
| 22
| 1,760,719,813.637722
|
https://www.phoronix.com/news/AMD-Cypress-OpenGL-4.1
|
Testing OpenGL 4.1 With An AMD Cypress GPU On The Latest Open-Source Driver
|
Michael Larabel
|
When it comes to OpenGL 4 support on the AMD R600 Gallium3D driver for pre-GCN graphics cards, currently the only R600g-supported cards advertising OpenGL 4.1 right now are the Radeon HD 5800 "Cypress" and Radeon HD 6900 "Cayman" series. Here are some tests done with OpenGL 4.1 on a Radeon HD 5830 compared to Cayman and various GPUs with the RadeonSI Gallium3D driver.
When OpenGL 4.1 was exposed in R600g for the select cards, I ran several benchmarks from my Radeon HD 6950 Cayman. I didn't have the chance to run any GL 4.1 tests with the Radeon HD 5830 Cypress since it was lodged away in a 4U system racked up for daily benchmarking. However, in cleaning and organizing some of the system racks this weekend, I pulled out the Radeon HD 5830 for running some tests.
The HD 5830 and other open-source AMD OpenGL 4.1-supporting cards were tested with Linux 4.5 and Mesa 11.2-devel from the Padoka PPA on Ubuntu 15.10.
OpenGL 4.1 was indeed working for this AMD Cypress GPU!
The HD 5830 was on par with the other graphics processors for DiRT Showdown in being largely CPU bound.
The HD 5830 comes in at just under 30 FPS when running Metro Last Light Redux at 1080p.
While not using OpenGL 4, here's the Counter-Strike: Global Offensive results just for showing a greater comparison of these GPUs. You can further analyze the results or run your own comparison via this OpenBenchmarking.org result file. There are also more Linux OpenGL benchmark results via today's More AMDGPU & Radeon Benchmarks Of Linux 4.5 With Mesa 11.2 Git.
| 26
| 1,760,719,814.22886
|
https://www.phoronix.com/news/GPUOpen-Launches
|
AMD Launches GPUOpen Today
|
Michael Larabel
|
Today's the day that AMD opens up GPUOpen.
Last month AMD announced GPUOpen. With GPUOpen they are opening up more software from libraries to tools and SDK under the MIT license. This just isn't a Linux-focused project but is equally focused on Windows with DirectX code samples and more. It's basically AMD's competition to NVIDIA's GameWorks but with going for the open-source approach.
There is an initial blog post today on GPUOpen.com and it explains the principles:
The first is to provide code and documentation allowing PC developers to exert more control on the GPU. Current and upcoming GCN architectures (such as Polaris) include many features not exposed today in PC graphics APIs, and GPUOpen aims to empower developers with ways to leverage some of those features. In addition to generating quality or performance advantages such access will also enable easier porting from current-generation consoles (XBox One and PlayStation 4) to the PC platform.
The second is a commitment to open source software. The game and graphics development community is an active hub of enthusiastic individuals who believe in the value of sharing knowledge. Full and flexible access to the source of tools, libraries and effects is a key pillar of the GPUOpen philosophy. Only through open source access are developers able to modify, optimize, fix, port and learn from software. The goal? Encouraging innovation and the development of amazing graphics techniques and optimizations in PC games.
The third is a collaborative engagement with the developer community. GPUOpen software is hosted on public source code repositories such as GitHub as a way to enable sharing and collaboration. Engineers from different functions will also regularly write blog posts about various GPU-related topics, game technologies or industry news.There are also some blog posts about using GPU PerfStudio, using the CodeXL Analyzer CLI, optimizations, and other initial content. However, as of writing I have yet to see any new code released on GitHub. Explore more at GPUOpen.com.
| 42
| 1,760,719,815.227221
|
https://www.phoronix.com/news/AMD-Stoney-Has-ETC2
|
AMD Stoney APUs Bring ETC2 Open-Source Driver Support
|
Michael Larabel
|
AMD's upcoming "Stoney" APUs has support for ETC2 texture compression.
A commit today enables ETC2 hardware support for Stoney. The commit by Marek reads, "radeonsi: add ETC2 support for Stoney. Tested and working."
There's been ETC2 in Gallium3D for Freedreno and Nouveau but up to now there's been no hardware support for ETC2 and within Mesa there is support in the Intel driver.
ETC2 is very promising as an advanced patent-free texture compression algorithm for OpenGL unlike S3TC. Separately work also continues to go into ASTC texture compression as another alternative backed by Khronos. ETC2 codecs are mandatory in OpenGL ES 3.0 and OpenGL 4.3.
| 6
| 1,760,719,815.66686
|
https://www.phoronix.com/news/Compute-Shader-G3D-Q
|
The State of Compute Shaders For Gallium3D Drivers
|
Michael Larabel
|
Since last month Intel has offered compute shader support via their open-source Linux graphics driver. The ARB_compute_shader support is needed for OpenGL 4.3 but so far Intel is the only Mesa/Gallium3D driver having support for this important extension.
The matter of compute shaders for other Gallium3D drivers like Radeon and Nouveau were brought up this week in this forum thread.
AMD's John Bridgman commented about the support, "If you look at GL 4.3 you'll see that compute shaders are one of a handful of remaining features that need to be supported in order to claim GL 4.3 support. In addition I believe there is at least one recently released Linux game which requires them (I think there are two but not sure) so it will help there as well. Support for compute shaders was recently added to the Intel driver (you can see the 2015-12-10 date stamps), and IIRC other folks have been working on the radeon/nouveau drivers... generally once the first driver has a feature working the other drivers pick it up fairly quickly afterwards."
Additionally, AMD's Alex Deucher provided a bit more as to the current state, "The shader compiler and [hardware] setup is already implemented for clover [the OpenCL Gallium3D state tracker]. It just needs to be plumbed up to the GL compute extensions. That said, it's not particularly useful until shader buffer and image support is implemented. Most of that has to be done in gallium and can be shared between gallium drivers."
On the Nouveau side, there's this active GitHub branch where ARB_compute_shader support is being worked on for the Nouveau NVC0 driver. That Nouveau compute effort is being led by prolific contributor Ilia Mirkin. Correction: Most of the Nouveau compute shader work is being done by Samuel Pitoiset and there is more of his work on this OpenGL extension via this Git repository.
If you are unfamiliar with compute shaders, you can learn all the fun details via the OpenGL.org Wiki.
| 10
| 1,760,719,816.50441
|
https://www.phoronix.com/news/AMD-HSA-GCC-Merged
|
AMD HSA Support Merged Into GCC Trunk
|
Michael Larabel
|
The AMD Heterogeneous System Architecture (HSA) code has been mainlined within the GCC compiler!
A few days ago the latest patches were published and today the work, which was done by SUSE under contract with AMD, is now in the mainline GCC code-base.
As mentioned in that earlier article, The patches provide an HSA libgomp plug-in for run-time with hooking into GCC's OpenMP library, an HSA register allocator, and the necessary changes for providing HSA offload support. New GCC options relating to HSA are: --with-hsa-runtime, --with-hsa-runtime-include, --with-hsa-runtime-lib and --with-hsa-kmt-lib. Besides a supported compiler, you also need to be using a recent kernel that's equipped with the AMDKFD driver and on your system have the new HSAKMT library. The work landed this morning and now makes it another nice feature for the upcoming GCC 6 release.
| 2
| 1,760,719,817.38726
|
https://www.phoronix.com/news/AMD-Looking-At-GLVND
|
AMD Is Planning To Support GLVND For Easier Linux Driver Setup & Maintenance
|
Michael Larabel
|
The other follow-up question I received an answer to on Friday from AMD's media liaison was whether the company is looking at supporting the OpenGL Vendor Neutral Dispatch Library (GLVND) to make it easier to install and maintain their user-space GL driver on Linux systems.
GLVND has been in development for a few years by NVIDIA as part of a new Linux OpenGL ABI. NVIDIA's 361 beta driver released earlier this month finally shipped with GLVND support while NVIDIA developers this GL Vendor Neutral Dispatch Library as an open-source project.
As explained in that earlier article, "Using libglvnd will ultimately simplify the Linux OpenGL driver installation and make it easier for multiple different OpenGL drivers to co-exist on the same system. With this new format, the different proprietary drivers won't overwrite each other nor cause havoc to the stock Mesa libraries. That is as long as the other prominent Linux OpenGL drivers also opt for the GLVND design."
I followed up with AMD to see whether they intend to support GLVND to make it easier for installing/uninstalling and maintaining their drivers. Here's their response:
We are interested in implementing support for LIBGLVND now that it has met a reasonable level of completeness and stability. We are in the process of evaluating when we can add support to our roadmap.Yes! So in the future, it will be easier and more sane for dealing with Linux OpenGL video driver installations and not fearing your system is in a borked state upon removing one of these drivers that would otherwise have overwritten the libGL.so.
| 30
| 1,760,719,817.920167
|
https://www.phoronix.com/news/Help-Bring-Older-GCN-To-AMDGPU
|
You Can Help Bring Vulkan Support To Older AMD GCN GPUs
|
Michael Larabel
|
Earlier today I wrote about how AMD will only be supporting Vulkan with the AMDGPU DRM kernel driver and not the more common Radeon DRM kernel driver. Here's a few more points to clarify the situation.
That article was about receiving confirmation from AMD that "our Linux Vulkan drivers will only run on the amdgpu kernel driver. We have no plans, on Linux, to support Vulkan on any other driver stack." Right now that means there is just support for the Tonga / Fiji / Carizo / Stoney hardware and all future graphics processors like Polaris. Where there's been some confusion has been whether AMD will support older GCN GPUs prior to Volcanic Islands with the AMDGPU kernel driver where already these older GPUs are supported by the Radeon DRM. The AMDGPU driver does have the experimental Sea Islands support, but that's disabled by default and there is currently no coverage of Radeon HD 7000 series hardware by this newer driver. AMD is investigating possibilities around supporting older GCN GPUs, but that's undecided as mentioned in the AMDGPU update I posted last week after a briefing by AMD.
This situation was made a bit more confusing after my article earlier today as Graham Sellers of AMD had tweeted, "To be perfectly clear, #AMD is planning to support #Vulkan on ALL GCN parts and to use the same #user mode driver on Windows + #Linux." Depending upon how you interpret it, that sounds like AMD will support Vulkan on all GCN (HD 7000 series and later) hardware. But how can that happen if those GPUs don't (currently) have AMDGPU kernel support? He's only talking about the user-space side of the equation
After bouncing a few emails this evening with Graham Sellers and Alex Deucher, it's been clarified that in the user-space AMD Vulkan driver there is support for all GCN GPUs. But, yes, the Vulkan driver is only targeting the AMDGPU kernel driver and AMD hasn't come to any formal conclusion about bringing support for older GCN GPUs to AMDGPU. However, as they were quick to remind, technically the community can bring the older GCN support to AMDGPU.
According to Graham, the Vulkan user-space driver -- even with initially being a closed-source blob -- would be able to support the older GCN GPUs with it just a matter of enhancing the open-source AMDGPU kernel driver to support the hardware going back to the HD 7000 series. Graham commented, "If the Open Source amdgpu kernel driver could boot up and talk to a pre-VI GPU, our Linux user-mode driver would run on it just fine. Honestly, that is the part that’s blocking us from supporting pre-VI on Linux immediately. You have great ties to the community. I see at the end of your post, you wrote, 'Later on once it's open-sourced perhaps by the community we'll be able to see the support extended to other GCN GPUs.' As I said, it’s not the part that will be initially closed source that lacks support, it’s the Open Source part. The community has access to that now."
Graham wrote a similar follow-up message on Twitter, "We have an #open source kernel mode #driver on #Linux which supports #VI and would love #community involvement. http://cgit.freedesktop.org/mesa/drm/log/amdgpu" (Not to be confused, the link Graham used for the open-source kernel driver wasn't the actual kernel driver but rather the AMDGPU bits that are part of Libdrm, the DRM library that interfaces between the user-space components and the AMDGPU DRM driver. To see the actual kernel driver code, that's over here).
So while AMD may end up porting the older GCN GPU support to the AMDGPU kernel driver, there are no plans to bring the Vulkan user-space driver to interface with the Radeon DRM driver. However, with both of these kernel drivers being open-source, those able to in the community could lend AMD a hand and work to bring older GCN support to AMDGPU.
Graham in an email to Phoronix had also reaffirmed that the Vulkan user-space driver is the same code-base between Windows and Linux, "Our user mode Vulkan driver supports all GCN parts starting from 1.0 (Southern Islands). We use an identical user-mode component on Windows and Linux (compiled for the appropriate target of course). The differences are minimal – window system stuff, and some abstractions of system level things like mutexes and shared memory."
| 87
| 1,760,719,819.071944
|
https://www.phoronix.com/news/AMD-HSA-GCC-Merge-Ready
|
AMD HSA Support Finally Appears Ready To Be Merged In GCC
|
Michael Larabel
|
It looks like the merging of AMD HSA (Heterogeneous System Architecture) support into the GCC compiler is imminent.
For months we have been covering the HSA patches for GCC and their hopes of getting the code merged for GCC 6. Feature development on GCC 6 is over, but there still is the possibility of release exceptions and this HSA support would be new functionality that can be optionally enabled.
SUSE has been working in cooperation with AMD on this HSA support inside GCC, thanks to the German company's compiler experts. Martin Jambor of SUSE posted a new GCC patch thread today entitled Merge of HSA branch. He commented:
This is hopefully the last big re-post of the HSA patches. We have incorporated all the feedback and found and fixed a couple more bugs. The complete patch-set bootstraps and tests fine on an x86_64-linux, when you do not enable HSA, there are a few expected warnings when HSA is enabled which I will address as a followup together with more testsuite changes. The patches were specifically designed so that the impact on pople not enabling HSA should be minimal. A last round of complete testing on an actual HSA-capable APU is still underway and I won't have the results until tomorrow but preliminary results were good and I did not want to hold up these patches for any longer.
The libgomp, omp and configuration bits have been reviewed by Jakub, a few other bits by Richi, but still Honza should review the IPA parts and I suppose someone other than me should ack the hsa-* files, even though I probably now have the authority to do it myself.
Thanks everybody for patience and feedback. While we are of course opened for mor more of it, let's also hope the approval process will finish soon as it should now.The patches provide an HSA libgomp plug-in for run-time with hooking into GCC's OpenMP library, an HSA register allocator, and the necessary changes for providing HSA offload support. New GCC options relating to HSA are: --with-hsa-runtime, --with-hsa-runtime-include, --with-hsa-runtime-lib and --with-hsa-kmt-lib. Besides a supported compiler, you also need to be using a recent kernel that's equipped with the AMDKFD driver and on your system have the new HSAKMT library. AMD's John Bridgman should be able to comment more in our forums on other requirements and details about this HSA support for the GNU Compiler Collection.
| 12
| 1,760,719,819.315559
|
https://www.phoronix.com/news/AMD-A1100-Tomorrow
|
AMD Finally Set For Opteron A1100 "Seattle" ARM Launch
|
Michael Larabel
|
AMD appears poised to announce their long-awaited A1100 "Seattle" ARMv8 processor tomorrow.
Per this new thread it looks like other media was briefed on the A1100 being set to launch tomorrow and some had accidentally jumped the gun. The already known information on the Opteron A1100 Seattle SoC have included up to eight ARM Cortex-A57 cores, 2 x 64-bit DDR3/DDR4 channels up to 1866MHz, 2 x 10GB Ethernet, 8-lane PCI Express 3.0, and 14x Serial ATA 3.0 ports.
These Opteron ARM processors are obviously designed for the data-center and AMD says they're now in mass production.
Hopefully AMD will soon be announcing their 96Boards HuskyBoard, which was last reported on as being set for release in Q4'15. Their development board looks great and should very exciting as the NVIDIA Jetson TX1. I've been going crazy to get my hands on an AMD ARM board but haven't heard anything at all from AMD yet nor any briefing. Hopefully they end up sending over a sample (particularly concerning all the ARM Linux benchmarks routinely found on Phoronix and not the mainstream Windows-focused review sites; I still haven't seen any other site do as much testing as I did of the JTX1 performance). Otherwise hopefully reader tips and new premium subscribers will be able to cover getting a HuskyBoard as I imagine many of you are also interested in this AMD ARM board for Linux use.
| 29
| 1,760,719,820.565533
|
https://www.phoronix.com/news/Linux-4.5-Fam15h-Power
|
New AMD CPUs To Support Power Monitoring With Linux 4.5
|
Michael Larabel
|
With the Linux 4.5 merge window's hwmon subsystem pull request is an update for new AMD Family 15h processors to support power monitoring.
With the hwmon update, the Family 15h Model 70h-7fh processors now support monitoring their power usage (in Watts) via the fam15h_power driver. I haven't been able to find any concrete information for what APU/CPUs technically cover this Family 15h 70h series, but anticipate it being either the rumored desktop Carrizo APUs or Zen. If anyone knows for sure, please let us know in the forums.
The fam15h_power driver has been around for a few years. I haven't tried the driver in a few kernel releases, but last time I did on some Bulldozer era CPUs the results were wildly inaccurate.
In some other good news, it seems AMD is finally officially claiming responsibility for this fam15h_power driver. Back in 2012 when AMD let go of some of their Linux kernel developers that included the maintainers of this AMD power driver. Over three years later, AMD engineer Huang Rui is officially taking over maintenance of this driver as of the Linux 4.5 kernel.
Huang Rui has also been working on separate patches for AMD accumulated power reporting support on Linux, which isn't part of today's pull request.
The hwmon pull request besides the fam15h_power driver update also has basic support for the NCT6683 on Mitac boards, a new pmbus client driver for the LTC3815, and removes the htu21 driver as it's now supported by the IIO subsystem.
| 2
| 1,760,719,820.984807
|
https://www.phoronix.com/news/Reminiscing-OSS-AMD-2016
|
Reminiscing Over The Early Days Of The Open-Source AMD Linux Strategy
|
Michael Larabel
|
With this morning's article about New AMDGPU Details and Looking Forward To Major Radeon Linux Improvements In 2016, there is much to be excited about with the new driver model finally set to roll-out and other open-source projects coming about. However, if you are a relatively new Linux users, you may not know how this whole process began.
If you've been a Linux user for a few years or less, or rather have only been reading Phoronix for a short time, be sure to read this weekend: This Is What Started AMD's Open-Source Strategy from 2011 that looked back on the original events from 2007.
Celebrating A Year Of Open-Source ATI If this new AMD Catalyst-over-open-kernel-driver model is new information to you, also read AMD Is Exploring A Very Interesting, More-Open Linux Driver Strategy and AMD Moves Forward With Unified Linux Driver Strategy, New Kernel Driver.
If you still have a desire to learn more and get up to speed, there are the 700+ display driver articles I've written and nearly 200 Linux graphics card reviews along with thousands of items in the news archive over the past decade about the open and closed-source AMD Linux driver stack and the other GPU drivers too.
| 0
| 1,760,719,822.004301
|
https://www.phoronix.com/news/AMD-New-Linux-Power-Reporting
|
AMD Posts Accumulated Power Reporting Support For Linux
|
Michael Larabel
|
New patches have been posted by an AMD engineer for exposing accumulated power reporting.
This set of patches would calculate the average power consumption for the processor and expose it via the perf subsystem as a new performance counter as well as hwmon. Compared to the Family 15h power driver published long ago that just tries to tell how much power the CPU is actively using (though in all my tests, it was never close to accurate on the hardware I've tried), this accumulated power reporting would calculate the average power consumed during a specified measurement interval.
This accumulated power mechanism is supported in the hardware on Family 15h Model 60 CPUs (Carrizo, I believe) and newer.
More details on these tentative patches for AMD accumulated power reporting for the Linux kernel can be found via the patch series.
| 0
| 1,760,719,822.329673
|
https://www.phoronix.com/news/AMD-Cayman-GL4.1-DRI3
|
Running OpenGL 4.1, DRI3 With Mesa Git On An AMD Cayman GPU
|
Michael Larabel
|
As of earlier this month in Mesa Git is finally OpenGL 4.0 and 4.1 support for the Radeon R600g driver for pre-GCN hardware, albeit the subset capable of advertising GL4 compliance is right now just Cypress and Cayman. I took this opportunity to run some fresh Mesa Git benchmarks on an AMD Cayman GPU and a third run when enabling DRI3.
The Cayman GPU used for this round of testing was the Radeon HD 6950 but rather a FirePro V7900, the old workstation card with a "Cayman Pro GL" GPU and 2GB of GDDR5 video memory that's still available to this day and commanding a $700+ USD price for a brand new card.
The days when AMD used to send out (a lot of) review samples to Phoronix for Linux testing... This Cayman graphics card was indeed able to advertise OpenGL 4.1 when upgrading to Mesa Git. For an easy experience, from an Ubuntu 15.10 install I just enabled the Padoka PPA that ships Mesa Git alongside LLVM SVN.
For this brief comparison, I compared the performance with stock Mesa (v11.0.2) to Mesa 11.2-devel Git and then finally Mesa 11.2-devel Git while manually enabling DRI3. In an article last month I was showing off the significant DRI2 vs. DRI3 performance.
The kernel version was maintained on Linux 4.4 Git throughout testing. All benchmarks were run in a reproducible manner via the Phoronix Test Suite.
The tests run for this one-page article aren't using OpenGL 4 but this comparison is mainly for some new Mesa Git + DRI3 data.
You can find more of this data via this OpenBenchmarking.org result file.
| 12
| 1,760,719,823.397178
|
https://www.phoronix.com/news/AMD-GPUOpen
|
AMD/RTG Announces GPUOpen For Offering More Open-Source Code
|
Michael Larabel
|
AMD this morning is lifting an embargo on GPUOpen, a new effort for embracing open-source with their Linux drivers.
With GPUOpen it sounds like they're going to offer more open-source software -- tools, libraries, SDKs, etc -- under the MIT license on GitHub. However, it doesn't look like all of the details are clear and public right now. AMD had invited me out to California earlier this month for their Radeon Technologies Group press briefing, but their PR representative had only been informed of it a few days in advance when I already had other commitments. While I asked for information remotely or after the fact, that didn't happen and now I'm piecing together details of "GPUOpen" via the Windows sites writing about it this morning.
GPUOpen will launch in January with various effects, tools, and other libraries and SDKs. This isn't just for Linux with OpenGL/Vulkan but will also include DirectX 11/12 samples and other Windows SDKs.
Radeon Technologies Group is also continuing to talk up the AMDGPU driver stack we've been writing about on Phoronix for almost two years now... With Catalyst becoming isolated to a user-space blob and sharing the same AMDGPU DRM kernel driver. Unfortunately, no details about when the reformed Catalyst Linux driver is expected.
| 54
| 1,760,719,823.67733
|
https://www.phoronix.com/news/A10-8700P-Intel-Cores
|
How AMD's Carrizo A10-8700P Compares To Intel's Core i3/i5
|
Michael Larabel
|
While I ended up returning my AMD A10-8700P "Carrizo" laptop due to its faulty fan, I did run a few benchmarks of it prior to sending it back. Here's roughly what you can expect in terms of its performance against Intel Core i3 and i5 laptops.
There was a broken fan blade that led to the noise, but the heatsink fan was still working... As the thermal data shows from that aforelinked article, it wasn't getting too warm during those prior benchmarks so it shouldn't have been thermal throttling. Nevertheless, take these results for what you want. Additionally, the Carrizo tests were done from a live USB due to returning the laptop. This is just a quick, one-page article with the limited data I have available so figured I would share with not knowing if/when I'll have my hands on another Carrizo device under Linux. I compared that Toshiba A10-8700P performance to the new Broadwell-based Toshiba C55-C5241 and ASUS F555LA-AB31 laptops I've been talking about recently. The Toshiba has a Core i5 5200U while the ASUS laptop had a Core i3 5010U. All three devices were running Ubuntu 15.10 x86_64 with Mesa 11.0. The Carrizo laptop was using the open-source RadeonSI Gallium3D driver.
For the graphics tests, the A10-8700P laptop was coming up short of the HD Graphics 5500 of the i5-5200U laptop, but there's just this limited data I have available. The i3-5010U wasn't tested for graphics since it sports the same HD Graphics 5500.
When it comes to the CPU performance, the A10-8700P Carrizo APU was generally running between these Core i3 and i5 Broadwell CPUs.
Well, that's all of the data from this rough comparison. Again, take the data as you wish and once I get my hands on any new AMD APUs you can expect to see more thorough results.
| 31
| 1,760,719,824.937475
|
https://www.phoronix.com/news/Smach-Zero-Moves-Ahead
|
The Handheld Steam Machine With Linux & AMD SoC Moves Ahead
|
Michael Larabel
|
Smach Z, the portable/handheld Steam Machine powered by Linux, is moving ahead and hopes to launch to market this time next year.
The Smach Z is powered by Linux and designed to run the compatible Steam Linux games. This handheld has a 5-inch 1080p touch-screen, 64GB of internal memory, haptic controls, 4GB of RAM, and is powered by a new AMD G-Series embedded APU. The AMD SoC provides Jaguar CPU cores and Radeon GCN graphics with 2GB of video memory. The reported battery life of the device is five hours.
The Smach Z just launched on Kickstarter and has already raised $112,195 of their $976k goal. There are 34 days left. Pricing on the Smach Z with early bird sales starts out at 249 EUR and goes up from there with different offerings.
The Smach Z pre-orders are expected to begin shipping next October while they hope to go retail next December. More details can also be found on SmachZ.com. I haven't seen them say what Linux graphics driver they plan to use, but time should tell. With the hardware not shipping for another year, by then hopefully the new Catalyst stack that rides atop the AMDGPU DRM driver will be into good shape.
| 31
| 1,760,719,825.052916
|
https://www.phoronix.com/news/R600g-Tessellation-Lands
|
AMD R600g Tessellation Support Lands In Mesa Git
|
Michael Larabel
|
Just one week after talking about progress being made with AMD tessellation support for pre-GCN GPUs, the support has been finally added to Mesa Git!
David Airlie's work has come to fruition and landed in Mesa Git for enabling tessellation on Evergreen and Cayman GPUs. Due to hardware factors, not all R600g graphics processors are able to support tessellation easily, but just the HD 5000 "Evergreen" and HD 6900 "Cayman" graphics cards. Testing is needed for other HD 5000/6000 series hardware.
However, beyond tessellation, the R600g driver will only support FP64 for Cypress and Cayman GPUs meaning just the HD 5800 and HD 6900 series are going to move a bit further into the OpenGL 4 world.
This now technically marks the R600 Gallium3D driver as supporting OpenGL 4.1 for select GPUs, which matches the state of RadeonSI. Blocking the two drivers from OpenGL 4.2 are ARB_shader_atomic_counters and ARB_shader_image_load_store.
| 68
| 1,760,719,826.658818
|
https://www.phoronix.com/news/R600g-Gets-Copy-Image
|
R600 Gallium3D Lands Another OpenGL 4.3 Extension (Copy Image)
|
Michael Larabel
|
AMD's R600 Gallium3D driver just made it up to OpenGL 4.1 support (for the few select GPUs that provide the necessary capabilities), but things aren't stopping there with another OpenGL 4.3 extension now landing.
ARB_copy_image support of OpenGL 4.3 has been implemented for R600g. The commit landed this morning. ARB_copy_image is needed by BioShock Infinite and other modern Linux games.
Landing alongside that new extension support were a number of other R600 driver fixes and two RadeonSI fixes. R600g now supports the same OpenGL 4 extensions as RadeonSI, but again it's not relevant for all of the hardware supported by this older driver for Radeon HD 6000 series and prior.
| 18
| 1,760,719,826.731014
|
https://www.phoronix.com/news/Few-Carrizo-Linux-Benchmarks
|
A Few Benchmarks Of The AMD A10-8700P "Carrizo" On Linux
|
Michael Larabel
|
For those excited to see some AMD A10-8700P "Carrizo" notebook benchmarks on Linux, here are a few numbers.
In continuation of yesterday's article of The Toshiba Carrizo-Powered Laptop Is Screaming, Literally, it seems universally that my initial concern was correct that it's a faulty fan on this brand new Toshiba Satellite notebook.
So I called up Toshiba USA to try to RMA it... And even though this is a brand new, latest-generation Toshiba laptop, they wouldn't honor doing an RMA for a faulty fan and sounded like I would end up having to pay for it to be repaired. (Or I could take it apart myself and void whatever this worthless warranty is in the process...) Since it was new, the outsourced telephone representative was telling me to take it back to the retailer. So that's what I plan to do. Assuming it gets refunded fine, I think I've had enough of this short-lived, cheap Toshiba laptop experience.
However, before doing so, I decided to at least boot up an Ubuntu 15.10 Live distribution from USB and run a few benchmarks, for what it's worth. While running the benchmarks I was trying to run it where the noise wasn't horribly loud and would still allow sufficient ventilation for the laptop.
Anyhow, at least the peak reported system temperature was 72C and an average of 60C.
If you want to see some C-Ray, FLAC audio, Stockfish, and OpenSSL benchmark results while the system monitoring module was activated, see this OpenBenchmarking.org result file.
For those curious about the Radeon graphics performance out of this A10-8700P Carrizo APU, there are some Xonotic results via this OpenBenchmarking.org result file.
Unfortunately, that's it for this brief Toshiba Carrizo 15.6-inch laptop testing under Ubuntu Linux... Being bound by the Live OS and 8GB of RAM plus the concern of heat with the faulty heatsink fan, I stopped there and packaged it back up to return it. Unless I find some other low-cost A10-8700P laptop that fits the tight budget, and the return goes fine, this is sadly the last of the benchmarks for this Toshiba Satellite L55D-C5269 and will have to wait for some other interesting AMD APUs to come around.
| 24
| 1,760,719,828.242907
|
https://www.phoronix.com/news/ASoC-V5-AMD-APUs
|
ASoC Support Still Being Worked On For AMD APUs
|
Michael Larabel
|
Alex Deucher on Friday sent out the latest patches for implementing ASoC support for AMD APUs. These patches provide i2s audio support via a new driver and integrates with the AMDGPU DRM.
AMD open-source developers have been working on this support for a while and sent out Friday was the fifth revision to these patches.
These latest patches can be found via the DRI devel list. About the new ACP driver, "This adds the ACP (Audio CoProcessor) IP driver and wires it up to the amdgpu driver. The ACP block provides the DMA engine for i2s based ALSA driver. This is required for audio on APUs that utilize an i2s codec...ACP IP block consists of dedicated DMA and I2S blocks. The PCM driver provides the platform DMA component to ALSA core."
Hopefully this i2s audio support for AMD APUs will be prepped in time for the Linux 4.5 kernel.
| 0
| 1,760,719,828.411684
|
https://www.phoronix.com/news/AMD-Carrizo-Laptop-Coming
|
AMD A10-8700P "Carrizo" Linux Laptop Testing
|
Michael Larabel
|
As a follow up to What Sub-$500 Laptops Are You Most Interested In For Linux?, one of them has been chosen so far and will satisfy the many from the dozens of comments wanting to see an AMD Carrizo laptop tested under Linux.
The first of two laptops ordered so far is the Toshiba Satellite L55D-C5269. For $449 USD, this laptop provides a 15.6-inch display, AMD A10-8700P APU, 8GB of DDR3L memory, and 1TB SATA HDD. The most interesting part of that for the forthcoming tests is the A10-8700P -- a Carrizo APU. The A10-8700P is a Carrizo APU with two Excavator CPU modules and Radeon R7 graphics with six compute cores.
I've been dying to test a Carrizo APU under Linux and next week once this laptop arrives from Amazon I'll be able to do so. It should be interesting to see the A10-8700P for others interested in a budget AMD Linux laptop.
The second sub-$500 laptop to be decided for this December article is likely going to be one with an Intel Core i5 5200U "Broadwell" processor. In the same price range as this Carrizo laptop, the i5-5200U seems to be popular. In fact, for just $25 less is this Toshiba laptop that's almost identical (15.6-inch, 1TB HDD, 8GB RAM), but with this Core i5 Broadwell CPU. The i5-5200U is dual-core with Hyper Threading, 2.2GHz with 2.7GHz Turbo, 15 Watt TDP, and HD Graphics 5500.
Comparing the AMD A10-8700P Carrizo and Intel i5-5200U Broadwell laptops with them being quite similar and both coming in at under $450 USD could be quite interesting... Given that the laptops both ship with Windows 10, I'll also do some Windows vs. Linux benchmarks on both laptops for the compatible cross-platform test profiles.
I'll likely firm up my decision on the 5200U laptop tomorrow once looking out for any other Cyber Monday deals. If anyone has any feedback or thoughts to share, please comment on this article in the forums.
| 37
| 1,760,719,829.704073
|
https://www.phoronix.com/news/Radeon-Crimson-RelNotes
|
AMD Crimson Linux Release Notes: Glxgears Stuttering Fixed
|
Michael Larabel
|
As the latest on today's interesting AMD Crimson driver release for both Linux and Windows, the release notes for the new Linux driver build are now available.
You can find the release notes now live via this link. However, as was common with Catalyst Linux updates, the Radeon Software release notes for Linux aren't too useful. The mentioned issues that were resolved are:
[SWDEV-55204] Stuttering when running glxgears with VSync enabled
[SWDEV-7339] Intermittent mouse cursor corruptionThat's all that is officially mentioned as the first proprietary Linux driver update in two months. The release notes mention stuttering for glxgears but doesn't seem to mention the big performance boosts were expecting -- tests are still ongoing.
The release notes indicate HD 5000/6000 series support still while AMD has dropped support for pre-GCN GPUs. The release notes also mention that only Linux 3.19 is supported as the latest kernel version when in fact there is initial Linux 4.x support. X.Org Server 1.17 is listed as the latest xorg-server supported so unless the notes are wrong there too, the modern X.Org Server 1.18 isn't yet handled with its new ABI.
| 23
| 1,760,719,831.043989
|
https://www.phoronix.com/news/AMD-GCN-OpenGL-Hack
|
A Developer Hacked AMD's GCN GPUs To Run Custom Code Via OpenGL
|
Michael Larabel
|
A Phoronix reader pointed out an interesting slide deck that's gone relatively unnoticed up until now about a game developer "Hacking GCN via OpenGL" for allowing some interesting possibilities.
Tomasz Stachowiak, a developer at Frostbite Games in Sweden, posted some slides recently about this hacking of GCN (Graphics Core Next) via OpenGL code. Tomasz was looking to take full control over shader code and be able to feed hardware-specific ISA to the GPU without being bound by the bytecode of DirectX or Vulkan, etc. He's exploring possibilities like self-modifying shaders, shader sub-routines, and more.
His GCN "hack" involves using OpenGL's ARB_get_program_binary extension, cheating a CRC function of the driver, and then passing custom ISA to the GPU. The ISA is put into a buffer and the hacked shader causes the AMD GCN GPU to jump to that buffer and execute.
Among the future items he or others may explore is for generating code on the GPU, custom shader compilers, and a GPU-based OS. This is all the short story, go see his 50+ slides for all of the technical details.
| 27
| 1,760,719,831.482778
|
https://www.phoronix.com/news/AMD-Crimson-Pre-GCN
|
Latest On AMD Crimson For Linux: Supports 4.x Kernels, Drops Pre-GCN GPUs
|
Michael Larabel
|
Here's the latest in my hurried but exciting testing of the AMD Radeon Software Crimson Edition for Linux atop Ubuntu.
While AMD Crimson for Linux doesn't bring a new GUI like it does on Windows, the good news is that this release is finally building against the latest Linux 4.x kernels. I've confirmed it for at least Linux 4.1 it's building fine and others are having success to with Linux 4.x. However, no word if this release supports the latest Linux 4.3 kernel nor if it can handle the fresh X.Org Server 1.18.
As a bit of unfortunate news for those with older AMD graphics cards: Crimson on both Linux and Windows has gone away with pre-GCN GPU support. The Radeon HD 5000/6000 series are no longer supported by the mainline Radeon Software (Catalyst) driver, leaving you with Catalyst 15.9 driver as the last official update.
The R600g driver and rest of the open-source Radeon driver stack is mature for those with pre-GCN graphics processors, but the only sad part about that is that this driver hasn't yet reached complete OpenGL 4.0 support. So if you previously relied on some OpenGL 4 Linux games with a higher-end Northern Islands GPU, now you'll be out of luck or just have to stick to Catalyst 15.9 without any kernel or xorg-server updates. Likewise, if you were using CrossFire or other features only offered by Catalyst that don't yet have open-source support. Granted, Graphics Core Next has been out since 2011 so you may want to think about upgrading your hardware this holiday season.
Stay tuned for more details as I continue exploring this new driver release on Linux and the OpenGL performance tests are running.
| 54
| 1,760,719,832.40131
|
https://www.phoronix.com/news/More-AMD-HCC-Details
|
AMD Working On CUDA Source Translation Support To Execute On FirePro GPUs
|
Michael Larabel
|
Early this morning I wrote a brief article about AMD working on an LLVM-based Heterogeneous Compute Compiler and since then more details have come to light.
That Heterogeneous Compute Compiler article was ran after coming across some slides from the LLVM Developers' Meeting that took place a few weeks back. SuperComputing '15 kicked off today in Austin and AMD has coincidentally revealed more details on their compiler work.
As part of HCC, AMD is also working on HIP (Heterogeneous-compute Interface for Portability) for porting CUDA source code to a common C++ programming model. This translated CUDA could could then run back through NVIDIA's NVCC CUDA compiler or be sent through AMD's new HCC compiler. It's basically translated CUDA support at the source level, but with no plans for allowing CUDA binaries to execute atop AMD GPUs.
With AMD "HIPify" tools, it should be a "simple" process for porting CUDA to HIP. The HCC compiler will support C++ 11/14 and C11 along with OpenMP 4.0 offloading. C++17 support will come. AMD is also working on Linux driver improvements as part of this latest GPU computing initative and they're working to deliver proper headless Linux support for AMD GPU compute servers.
More details can be found via AMD's press release for what they're calling the Boltzmann Initiative. Meanwhile, it seems they're leaving the Open64 compiler support left to die.
| 17
| 1,760,719,833.169341
|
https://www.phoronix.com/news/AMD-HCC-LLVM-Plans
|
AMD Plans To Contribute Heterogeneous Compute Compiler
|
Michael Larabel
|
AMD has been open-sourcing several components of their Linux HSA (Heterogeneous System Architecture) stack for the past several months including the AMDKFD kernel driver and HSAKMT run-time. In cooperation with SUSE, they also hope to have HSA accelerator support in GCC 6. Besides the GCC support, AMD is apparently planning to publish a Heterogeneous Compute Compiler.
While AMD has seemingly focused on the GNU Compiler Collection for early HSA support, they're reportedly planning to publish a Heterogeneous Compute Compiler (HCC) that utilizes LLVM. In still going through a few new slide uploads from the LLVM Developers' Meeting, one of the slides makes mention of this Heterogeneous Compute Compiler:
I don't think I've heard AMD or any of the open-source developers mention this Heterogeneous Compute Compiler for LLVM previously, but should be interesting. Hopefully Bridgman or others will be able to shed some light on it in our forums.Update: More details are now available.
| 4
| 1,760,719,833.768949
|
https://www.phoronix.com/news/R600g-Catch-Texture-View
|
AMD R600g Catches Up With Another GL 4.3 Extension
|
Michael Larabel
|
Thanks to Glenn Kennard, the R600 Gallium3D driver now supports another OpenGL 4.3 extension that was already available for RadeonSI, Intel i965, Nouveau NV50/NVC0, and even the Softpipe and LLVMpipe drivers.
The extension that R600g has now caught up with is support for ARB_texture_view as needed by the OpenGL 4.3 specification.
The ARB_texture_view specification is explained at the OpenGL.org registry. "This extension allows a texture's data store to be "viewed" in multiple ways, either reinterpreting the data format/type as a different format/type with the same element size, or by clamping the mipmap level range or array slice range. The goals of this extension are to avoid having these alternate views become shared mutable containers of shared mutable objects, and to add the views to the API in a minimally invasive way."
In the latest Mesa Git code, R600g still lacks OpenGL 4.0 support due to ARB_tessellation_shader and then after that is nearly caught up with the RadeonSI driver for OpenGL 4 extensions. However, not all graphics cards supported by the R600g driver will be able to handle all of the GL4 extensions and for some are currently limited in the Git code to just a few select GPU generations.
| 18
| 1,760,719,834.637168
|
https://www.phoronix.com/news/New-GCN-Assembler
|
An AMD GCN Assembler For Linux That Supports The Open & Closed Drivers
|
Michael Larabel
|
A Phoronix reader pointed out that a developer has released an assembler for AMD GCN GPUs that supports both the open and closed-source Catalyst drivers on Linux.
This CLRadeonExtender project has complete GCN assembler/disassembler support for all GCN GPUs from GCN 1.0 through GCN 1.2, including full Fiji support. The assembler supports the binary formats of the AMD Catalyst driver with OpenCL 1.2 as well as Gallium3D compute for using the RadeonSI open-source driver.
More information on the CLRadeonExtender assembler can be found via this AMD community thread. The project is hosted at clrx.nativeboinc.org and the code for it is open-source.
| 19
| 1,760,719,835.137395
|
https://www.phoronix.com/news/AMD-Stoney-Linux-4.4
|
AMD Stoney APU Support Is Going Into The Linux 4.4 Kernel
|
Michael Larabel
|
Alex Deucher sent in another pull request of new AMDGPU/Radeon DRM material for landing in DRM-Next to in turn make it into Linux 4.4.
Earlier this month AMD sent in their main DRM pull request that included the new AMDGPU GPU scheduler by default, new AtomBIOS opcodes, and other improvements.
Meanwhile, the primary addition with this second feature pull for DRM-Next is the new AMD Stoney APU open-source support. Stoney is the codename for a forthcoming APU due out in 2016. The code to enable Stoney support isn't too big since it's based on a Volcanic Islands GPU. You can find out more about Stoney via the aforelinked article as well as the forum comments over there.
Another change with this latest pull besides the Stoney support is increasing the VM default size to 64GB from 8GB. More details on this latest code that just landed can be found via the DRM-Next repository.
Sadly, not making it for Linux 4.4 is power management / re-clocking for the discrete GPUs with the AMDGPU driver thus the open-source performance with these latest graphics cards is very slow.
| 14
| 1,760,719,836.107287
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.