Announcement

Collapse
No announcement yet.

My Three Hopes For AMD's Open-Source Stack The Rest Of 2017

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • My Three Hopes For AMD's Open-Source Stack The Rest Of 2017

    Phoronix: My Three Hopes For AMD's Open-Source Stack The Rest Of 2017

    Being half-way through 2017 now, there are three wishes I hope will still be fulfilled this calendar year by the AMD open-source Linux graphics driver stack...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    openCL would be on the top of my wishlist

    Comment


    • #3
      Originally posted by Space Beer View Post
      openCL would be on the top of my wishlist
      I thought they open sourced it? Or ist ROCm in an unusable state?

      Comment


      • #4
        My sole wish for AMD drivers are to double their manpower. Day 1 conformance drivers are science-fiction on AMD hardware. Recent OpenCL 2.2 ratification is nice, but do you remember when OpenCL 1.1 and 1.2 SDKs were practically Day 1 releases? I do, and boy, those were good times to be GPU developers. Now again there are 3 C++ single-source, open GPGPU SDKs (C++AMP, HCC, SYCL) and none of them work. We have an unofficial CUDA to SPIR-V compiler, but serious projects won't pick it up unless someone stands behind it... To be honest it very much sucks that C++ is moving forward faster than ever, but GPGPU standardization lacks vendor effort. SYCL 1.2 is still not ready, neither it is cross-platform (yet)... If I had to start a multi-year project today, my sole option were OpenCL 1.2, which is how old? 5 years?

        Comment


        • #5
          Originally posted by Space Beer View Post
          openCL would be on the top of my wishlist
          Indeed, clover based openCL that benefits all drivers
          AMD specific :
          - get all possible features into r600g.
          - Atomic modesetting
          - Official RADV support

          Comment


          • #6
            Just for a fair comparison: It's not like Nvidia's control panel on Linux would be any good.
            -3D features are mostly non-existent (even the old anisotropic filtering and anti aliasing options are without function/completely broken nowadays)
            -display's features are limited crap, say hello to messing around with xorg.conf (stupid driver btw. also sabotages xrandr...)
            -using profiles is really painful to set up (and I suppose half of the functionality is broken anyway)

            The only things that aren't pure rubbish in the UI are OC and fan control. On the other hand, you can't use custom fancurves with nvidia-settings CLI since polling/setting fan speed comes with an abnormous overhead which makes games stutter terribly.

            Comment


            • #7
              For me the main priority is to see AMDGPU drm enabled by default on my R9 290.

              Comment


              • #8
                I'd be very happy chappy if these 3 things happened this year. But I'd settle for DC/DAL first followed by a simple control panel to start off with.

                Comment


                • #9
                  Originally posted by karolherbst View Post

                  I thought they open sourced it? Or ist ROCm in an unusable state?
                  Fiji (R9 fury) and Polaris (R9 400-500) only, at the moment. The code is sloppy and not readily runable on any given linux build without custom work done.
                  it's not yet in a desirable state for wide adoption on the Linux desktop or servers. ROCm still depends upon unmerged kernel changes, so the ROCm Ubuntu packages in AMD's repository ship with its own 4.9-based kernel. Additionally, it has its own fork of LLVM/Clang/LLD.
                  http://www.phoronix.com/scan.php?pag...5-opencl&num=1

                  The main reason why I want good OpenCL support is for Folding@Home, as biology is still one of the least understood things in existence. Folding@Home has a benchmark mode tool thing I would like to see in the Phoronix testing lineup too. I think the minimum requirement is OpenCL1.1 or 1.2, but I think the benchmark thing will go up to 2.0? (not too sure about that)

                  At the moment, I'm too much in debt to be spending money on anything, but hopefully around Christmas will be able to grab a phoronix premimum membership and more strongly request the Folding@Home benchmark be added in :3
                  fahbench - Folding@home GPU benchmark
                  Last edited by tiwake; 04 June 2017, 10:09 AM. Reason: removed double link

                  Comment


                  • #10
                    Originally posted by tiwake View Post

                    Fiji (R9 fury) and Polaris (R9 400-500) only, at the moment. The code is sloppy and not readily runable on any given linux build without custom work done.

                    http://www.phoronix.com/scan.php?pag...5-opencl&num=1

                    The main reason why I want good OpenCL support is for Folding@Home, as biology is still one of the least understood things in existence. Folding@Home has a benchmark mode tool thing I would like to see in the Phoronix testing lineup too. I think the minimum requirement is OpenCL1.1 or 1.2, but I think the benchmark thing will go up to 2.0? (not too sure about that)

                    At the moment, I'm too much in debt to be spending money on anything, but hopefully around Christmas will be able to grab a phoronix premimum membership and more strongly request the Folding@Home benchmark be added in :3
                    Hey Michael - new benchmark to try out https://folding.stanford.edu/2013/03...-for-your-gpu/

                    Comment

                    Working...
                    X