Announcement

Collapse
No announcement yet.

AMDGPU-PRO vs. Open-Source Gallium3D OpenGL Performance On Polaris Is A Very Tight Race

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

  • #71
    Well, if you have time and i if am in your place i would first try to find one worse one case... Then start bisect anything, either code but bisecting in game settings first as it might happen only on some setting scenario, etc...

    That way, we might conclude on the end what we actually guessing currently here

    Comment


    • #72
      Originally posted by Azpegath View Post

      I was thinking the same regarding Windows comparisons. And add the 290 (hopefully the regression is taken care of soon).
      The 290 regression is a bug in DRI3 exposed by async page flipping. The fix is actually a mesa patch:

      Comment


      • #73
        Great, so hopefully that should take care of the 260X regression as well...
        Test signature

        Comment


        • #74
          Originally posted by Nille_kungen View Post
          That makes me sure that more and more of the close source team is contributing to the open source driver and this is also what AMD said will happen.
          yes, because they switched to open kernel, so all blob kernel devs are now working on amdgpu. and i think some other components too

          Comment


          • #75
            Originally posted by haagch View Post
            Well, it could still be accidentally expensive checks when freeing some memory or something like that, but that is for people who know the code to decide.
            Nothing matching your description comes to mind offhand, but I guess it's possible.

            Since the CPU seems to stay pegged during the stalls, a system profile taken around one of them might be enlightening, but that might be difficult without a way to reproduce it reliably.

            Comment


            • #76
              Originally posted by agd5f View Post
              The 290 regression is a bug in DRI3 exposed by async page flipping. The fix is actually a mesa patch:
              https://lists.freedesktop.org/archiv...st/126338.html
              Note that this patch only helps with sync-to-vblank enabled. Since PTS always sets vblank_mode=0 AFAIK, I'm afraid it can't help for the regressions Michael has reported. If anyone else can reproduce the regressions with PTS (or otherwise with sync-to-vblank disabled), bisecting would still be interesting.

              Comment


              • #77
                RadeonSI is still slower than GpuPro at lower resolutions on counter strike go
                Those 20fps more or less can be the difference between staying always above the 144hertz or having small breaks during gameplay,
                that can hamper the game experience if playing in a more competitive manner.

                Comment

                Working...
                X