Announcement

Collapse
No announcement yet.

AMD Gallium3D Performance Is Much Better Than Two Years Ago

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

  • AMD Gallium3D Performance Is Much Better Than Two Years Ago

    Phoronix: AMD Gallium3D Performance Is Much Better Than Two Years Ago

    With the imminent release of Mesa 9.2, out this morning are Radeon "R600" Gallium3D driver benchmarks comparing Mesa Git master (9.3-devel) to Mesa 9.2 Git, Mesa 9.1.6, Mesa 9.0.3, Mesa 8.0.5, and Mesa 7.11.2. These are the past two years worth of Mesa releases and testing occurred on an ATI Radeon HD 4890 (RV790XT) graphics card.

    http://www.phoronix.com/vr.php?view=19068

  • #2
    Nice, and actual improvements are MUCH, much better. All of these tests were run with the latest kernel and different versions of userspace (Mesa). If each test for a given Mesa version would be run with the current kernel version at the time it was released, results would be both more realistic, and much more impressive.

    Comment


    • #3
      Would have been nice to see how the power usage changed as well, if the last ones have DPM enabled.

      Comment


      • #4
        Originally posted by GreatEmerald View Post
        Would have been nice to see how the power usage changed as well, if the last ones have DPM enabled.
        I doubt it, Michael doesnt change from defaults and even if he did, he used the same kernel for all tests. It isnt at all realistic because of that. When people were using 7.11 they definitely werent using that kernel.

        Comment


        • #5
          half the tests have massive regressions

          Comment


          • #6
            Originally posted by peppercats View Post
            half the tests have massive regressions
            As noted in the article the changes are likely due to newer features (like MSAA) being available on newer versions of the driver. Older versions of the driver only supported GL 2.1 while newer versions support GL 3.1.

            Comment


            • #7
              Originally posted by agd5f View Post
              As noted in the article the changes are likely due to newer features (like MSAA) being available on newer versions of the driver. Older versions of the driver only supported GL 2.1 while newer versions support GL 3.1.
              IMHO, then tests with these games are useless if you want to show performance differences between driver versions and shouldn't be used. If you want to benchmark different driver versions you change nothing but the driver versions. Using different game settings (even if they are applied automatically) for the different driver versions distorts the benchmark and makes them useless.

              Comment


              • #8
                Originally posted by Vim_User View Post
                IMHO, then tests with these games are useless if you want to show performance differences between driver versions and shouldn't be used. If you want to benchmark different driver versions you change nothing but the driver versions. Using different game settings (even if they are applied automatically) for the different driver versions distorts the benchmark and makes them useless.
                Which i agree with. But no one has stepped up to the plate and created a PROPER (eg NOT driconf) configuration utility for the open source drivers.

                Comment


                • #9
                  Maybe test Second Life?

                  If there's one game I've played off and on is SL and its performance is poor on my HD 3650. I don't buy these performance improvements in Gallium 3D for this specific GPU. I'd like to see more intensive games that really stress the OpenGL stack show improvements.

                  Michael maybe try Second Life in your test suite? It's quite demanding for shaders and things.

                  Comment


                  • #10
                    Excellent article!!

                    The ONLY thing keeping me on Catalyst Legacy (instead of r600) in Ubuntu 13.04 is that I had trouble running some of my Humble Bundle games with the OSS drivers (ie. Dungeon Defenders, Shank, Shank2, etc), while the catalyst legacy was able to handle them w/o problems. Also, the Catalyst seems to run Wine with higher compatibility than the OSS drivers.

                    Comment


                    • #11
                      Originally posted by gururise View Post
                      Excellent article!!

                      The ONLY thing keeping me on Catalyst Legacy (instead of r600) in Ubuntu 13.04 is that I had trouble running some of my Humble Bundle games with the OSS drivers (ie. Dungeon Defenders, Shank, Shank2, etc), while the catalyst legacy was able to handle them w/o problems. Also, the Catalyst seems to run Wine with higher compatibility than the OSS drivers.
                      Not higher compatibility, but different. Some games work better on one, some on the other.

                      Comment


                      • #12
                        Originally posted by spstarr View Post
                        If there's one game I've played off and on is SL and its performance is poor on my HD 3650. I don't buy these performance improvements in Gallium 3D for this specific GPU. I'd like to see more intensive games that really stress the OpenGL stack show improvements.
                        spstarr, are you running with the sb optimizer (R600_DEBUG=sb <program>) ? If not, give it a try.

                        Comment


                        • #13
                          Originally posted by Vim_User View Post
                          IMHO, then tests with these games are useless if you want to show performance differences between driver versions and shouldn't be used. If you want to benchmark different driver versions you change nothing but the driver versions. Using different game settings (even if they are applied automatically) for the different driver versions distorts the benchmark and makes them useless.
                          Mesa has the env var overrides to disable extensions, so Michael could easily do a no-MSAA bench.

                          Comment


                          • #14
                            New Shader optimizer

                            Originally posted by bridgman View Post
                            spstarr, are you running with the sb optimizer (R600_DEBUG=sb <program>) ? If not, give it a try.
                            I should try that, you are right...

                            Shawn

                            Comment


                            • #15
                              Originally posted by curaga View Post
                              Mesa has the env var overrides to disable extensions, so Michael could easily do a no-MSAA bench.
                              Unfortunately, MSAA cannot be disabled by environment variables completely. You can only disable the GL extensions GL_EXT_framebuffer_multisample and GL_ARB_texture_multisample, but I think you can't disable MSAA through GLX, which is what Nexuiz and other apps use.

                              Comment

                              Working...
                              X