Announcement

Collapse
No announcement yet.

Basemark GPU 1.2 Brings Linux Support - Wins For NVIDIA, Woes For Mesa

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

  • #21
    Originally posted by GruenSein View Post
    BTW: Any idea how well the AMDGPU Pro driver works with this benchmark?
    I was wondering as well.

    Comment


    • #22
      Originally posted by shmerl View Post
      Expected. Nvidia is likely optimizing for the benchmark on Windows, so it's essentially not telling much. Since their driver shares the code with Windows one, it affects the Linux result.

      Code optimized for the benchmark isn't necessarily a better code.
      This.

      Comment


      • #23
        Vega56 on 20.0.1 mesa and 10.0.0_rc3 llvm gives 40/55/129 FPS for 4k OpenGL and 52/65/92 FPS for 4k Vulkan(ACO). No glitches or corruption.




        P.S. Stupid benchmark, want's root. And uploads null results..
        /sys/firmware/dmi/tables/smbios_entry_point: Permission denied
        /dev/mem: Permission denied

        Comment


        • #24
          This test is batshit broken, it kinda works on OpenGL, and is pretty broken on RADV, regardless of ACO or not. If one tries setting custom test resolutions in windowed mode, it always set the same window size despite of the resolution selected, RADV looks totally broken in windowed mode, yet score seems to be in expected range (much better than same resolution full-screen score). Same shit on Wayland or X, this benchmark needs some serious fixing.

          Comment


          • #25
            Originally posted by Volta View Post
            Who cares about some meaningless benchmark?
            Given the number and frequency of benchmark results reported on here, and that they have at times found regressions in the kernel and other components, probably quite a lot of people care about them.

            Comment


            • #26
              Originally posted by Ribs View Post
              I've said it before and I'll say it again, unpopular as it may be around these parts...

              The end-user experience with nVidia's drivers is quite often superior to the AMD drivers (I concede not always superior), open source or not.

              Things usually "just work", and if you "don't mind" the closed source nature of the beast, you'd be forgiven to just not caring about the details and excuses as to why things aren't working if you just wanna fire up a game and kill some brain cells for an hour. I absolutely wish the nVidia drivers were as open as AMD's, but hey, if it works, it works.
              It is not unpopular what you say, it simply rarely corresponds to reality. If you tell me that Nvidia is still superior to AMD in terms of game performance, I can also agree, but if you say that with Nvidia things work for the user in GNU / Linux, take a tour of the support forums by Ubuntu & CO. By the way, in this period Nvidia has become an ordeal also for Ubuntu users, who after normal updates have seen their Nvidia driver break miserably. Let's not talk then if you decide to install a rolling release distribution ... it's pain in the ass!

              Comment


              • #27
                Well if Feral ported the benchmark to Linux I bet it wouldn't perform like this. Hey, has anyone tried running the Windows version under DXVK?! :-D

                But then again, this has Windows Vulkan support... I wonder how AMD Windows Vulkan performance is.

                Comment


                • #28
                  I don't trust the benchmark results because of my personal experience.

                  Bought an RTX 2070 and had some strange problems.
                  nVidia distro compatibility is not so great: openSUSE hangs on system reboot/poweroff at a grey screen with no apparent fix. MXLinux is worse- X freeze and force to alt-fx to VT and back to X and then desktop works.

                  It could once be said that nVidia Linux drivers were at least as good or perform better than Windows counterparts, but that was mostly in games where missing extensions or non-compliant behavior (think moss in water and living walls for UT2003/2004 on Linux). The extensions thing has caught up a bit to where those old problems (like the one with UT2003) are gone.

                  HOWEVER I compared Quake2RTX on both Linux and Window, and Windows was approximately %200 faster (89 fps) vs Linux (45 fps).

                  There are games (even opensource ones) that perform better on nVidia hardware just because the developers use those GPU. For example ioquake3, which is bottled at around 370 fps on AMD hardware - while quake3e port (on AMD) is around 1300 fps (opengl) and 1900 fps (vulkan).

                  All those problems I have with nVidia, so I use AMD now.

                  Comment


                  • #29
                    Guys, what's the fuss about? This isn't different from any other bug that we've seen. I'm sure it will get fixed eventually.

                    (I guess it would've been nice if we got a bug report in mesa about it, but it's no biggie.)

                    Comment


                    • #30
                      works fine on mesa 20.0 (Fedora Silverblue 32) RX 580

                      Comment

                      Working...
                      X