Announcement

Collapse
No announcement yet.

Mesa 21.0-devel RADV vs. AMDVLK 2021.Q1.1 Vulkan Driver Performance

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

  • Mesa 21.0-devel RADV vs. AMDVLK 2021.Q1.1 Vulkan Driver Performance

    Phoronix: Mesa 21.0-devel RADV vs. AMDVLK 2021.Q1.1 Vulkan Driver Performance

    For those wondering how the open-source Radeon Vulkan drivers of Mesa's RADV and AMD's official AMDVLK are competing as we start the new year, here are some fresh benchmarks looking at the performance for various Linux games (native and via Steam Play with DXVK) as well as Vulkan compute tests.

    http://www.phoronix.com/vr.php?view=29855

  • #2
    I suspect there is something wrong with the test environment here. Tests that were previously neck and neck now have +50% perf differences. You'd almost think one driver changed a lot but it is not consistently in favor of the same driver (and then check total war three kingdoms where there is a huge swing between 1440p and 4k which shouldn't really happen).

    Comment


    • #3
      What memory exactly you used in this benchmark? The table just says "16GB".

      The results are a bit suspicious overall. Need some extra investigations probably.

      Also, you need to make sure all the options in the settings are the same. Some of these games like to randomly change quality settings, even if you load same preset.

      Comment


      • #4
        Originally posted by BNieuwenhuizen View Post
        I suspect there is something wrong with the test environment here. Tests that were previously neck and neck now have +50% perf differences. You'd almost think one driver changed a lot but it is not consistently in favor of the same driver (and then check total war three kingdoms where there is a huge swing between 1440p and 4k which shouldn't really happen).
        The only things that come to mind short of a recent Mesa Git regression would be in terms of recent changes to that system would be something quirky happening with Resizable BAR issue (it was enabled from BIOS throughout) or Linux 5.11.
        Michael Larabel
        http://www.michaellarabel.com/

        Comment


        • #5
          Originally posted by baryluk View Post
          What memory exactly you used in this benchmark? The table just says "16GB".

          The results are a bit suspicious overall. Need some extra investigations probably.

          Also, you need to make sure all the options in the settings are the same. Some of these games like to randomly change quality settings, even if you load same preset.
          2 x 8GB DDR4-3600. Can pretty much assume the config is always at least rated memory speed and optimal number of memory channels. Sadly no way short of sudo/root for querying the DIMM information under Linux...
          Michael Larabel
          http://www.michaellarabel.com/

          Comment


          • #6
            Originally posted by Michael View Post

            2 x 8GB DDR4-3600. Can pretty much assume the config is always at least rated memory speed and optimal number of memory channels. Sadly no way short of sudo/root for querying the DIMM information under Linux...
            Install "dmidecode" and run dmidecode --type memory This yields all kinds of usefull information and the 'configured memory speed' not sure where that value is pulled but everything seems accurate on my machine

            Comment


            • #7
              Originally posted by Yttrium View Post

              Install "dmidecode" and run dmidecode --type memory This yields all kinds of usefull information and the 'configured memory speed' not sure where that value is pulled but everything seems accurate on my machine
              PTS does do this already, but for dmidecode to run on most distributions you need to be sudo/root to access /dev/mem, thus falls back to just reporting memory size.
              Michael Larabel
              http://www.michaellarabel.com/

              Comment


              • #8
                Originally posted by Michael View Post
                PTS does do this already, but for dmidecode to run on most distributions you need to be sudo/root to access /dev/mem, thus falls back to just reporting memory size.
                gotcha

                Comment


                • #9
                  Originally posted by Michael View Post
                  The only things that come to mind short of a recent Mesa Git regression would be in terms of recent changes to that system would be something quirky happening with Resizable BAR issue (it was enabled from BIOS throughout) or Linux 5.11.
                  Above 4G Decoding can also cause severe performance issues without SAM. I'm going to give SotTR a run on my system.

                  Comment


                  • #10
                    SotTR integrated benchmark, 1440p high TAA, RX 5700 XT (1,92GHz), 6700k (though completely GPU bound), no Above 4G Decoding available, linux 5.10-tkg, most recent drivers/git-master:
                    amdvlk-pro: 96fps
                    amdvlk-open: 94fps
                    radv: 94fps

                    Comment

                    Working...
                    X