Announcement

Collapse
No announcement yet.

Mesa 11.1-dev Tests With The Reverted RadeonSI Performance

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

  • Mesa 11.1-dev Tests With The Reverted RadeonSI Performance

    Phoronix: Mesa 11.1-dev Tests With The Reverted RadeonSI Performance

    This week I posted some AMD RadeonSI/R600g tests on Mesa 11.0 with DRM-Next along with a Intel Skylake vs. Radeon comparison using this new version of Mesa that will be officially released next month. Of course, following those tests, the requests turned to testing Mesa 11.1-devel rather than the Mesa 11.0 Git code.

    http://www.phoronix.com/vr.php?view=22097

  • #2
    Thanks Michael for the updated testing.
    Those who would give up Essential Liberty to purchase a little Temporary Safety,deserve neither Liberty nor Safety.
    Ben Franklin 1755

    Comment


    • #3
      Michael, can you test AMDGPU SI Machine Schedule For LLVM? Thx.

      Comment


      • #4
        Originally posted by ShFil View Post
        Michael, can you test AMDGPU SI Machine Schedule For LLVM? Thx.
        Darkbasic already does revisited test, so forget about big llvm improvments... it is all about that just this mesa patch regressed things

        http://openbenchmarking.org/result/1...DARK-150829771

        Comment


        • #5
          Can I assume that this performance patch affects the r600 driver too? since there were big regressions on that driver as well on the last benchmark.
          But its good to see than 11.1 helps a lot, thanks for the benchmark

          Comment


          • #6
            Still some performance regressions since 10.5: http://openbenchmarking.org/result/1...HA-GPUCOMPAR85

            Only Unigine is really improved.

            Comment


            • #7
              Originally posted by haagch View Post
              Still some performance regressions since 10.5: http://openbenchmarking.org/result/1...HA-GPUCOMPAR85

              Only Unigine is really improved.
              Those are probably not regressions, but apps are now rendered more correctly ... for example Tesseract does not render GI with that old mesa and old llvm:

              https://bugs.freedesktop.org/show_bug.cgi?id=79155

              Which means GPU render more, fps rate goes down obviously - that is not performance regression... and so on.
              Last edited by dungeon; 08-29-2015, 02:44 PM.

              Comment


              • #8
                Yeah, besides those r600 odd results in gputest - someone with r600 should checked out that, those are dunno maybe 50 times faster then Titan X

                Comment


                • #9
                  Originally posted by haagch View Post
                  Still some performance regressions since 10.5: http://openbenchmarking.org/result/1...HA-GPUCOMPAR85

                  Only Unigine is really improved.
                  Looks to me like there were a lot of perf improvements to the GCN 1.0 hardware, and some small perf regressions on the GCN 1.1 hardware.

                  Comment


                  • #10
                    Originally posted by smitty3268 View Post

                    Looks to me like there were a lot of perf improvements to the GCN 1.0 hardware, and some small perf regressions on the GCN 1.1 hardware.
                    Likely related to http://cgit.freedesktop.org/mesa/mes...fc5332128f9652

                    That was a big fix for SI stability... previously things are probably being a bit faster on CI but horribly unstable for SI.

                    Comment

                    Working...
                    X