Announcement

Collapse
No announcement yet.

Radeon Gallium3D Has Made Much Progress In Two Years

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

  • Radeon Gallium3D Has Made Much Progress In Two Years

    Phoronix: Radeon Gallium3D Has Made Much Progress In Two Years

    With the upcoming release of Mesa 8.1, here's a look at how the AMD Radeon "R600" Gallium3D driver performance has changed over the past two years. This article has benchmarks of each major Mesa release going back to Mesa 7.9 of 2010 back when the R600 classic DRI driver was still in early development but the only viable choice for using accelerated open-source graphics on AMD Radeon HD 2000 series graphics cards and newer. For the most part, the open-source Radeon Linux graphics performance has advanced greatly in terms of OpenGL performance over the past two years, but it's not without some outstanding regressions.

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    The development version seems to have some nice improvements there. By the time it gets mainstream, FGLRX will have been phased out already, so it's good timing.

    Comment


    • #3
      I see only World of Padman and Xonotic improved. The rest stayed same over the years or even degraded in performance terms.

      Comment


      • #4
        Originally posted by crazycheese View Post
        I see only World of Padman and Xonotic improved. The rest stayed same over the years or even degraded in performance terms.
        Michael mentioned that there was something fishy going on in both cases, where performance degraded. In OpenArena there might have been VSYNC issues, in Warsow there was incorrect rendering in the older release.
        In general (except for Open Arena) Q3 Engine performance has increased quite a lot. Lightsmark has doubled in Framerate. Vdrift is up by a factor of two. I think it's a tremendous gain and a reason to celebrate.

        Most interesting for me is however the Doom 3 performance, which didn't change at all. D3 really runs into a brick wall with completely similar benchmark results. Finding this bottleneck somehow might be worthwhile.

        Comment


        • #5
          Originally posted by crazycheese View Post
          I see only World of Padman and Xonotic improved. The rest stayed same over the years or even degraded in performance terms.
          From my own experience, I can only play ET:QW (which is a derivative of Doom 3) with the current git version. The previous version are too slow to be playable. The Doom 3 chart is quite strange.

          Comment


          • #6
            There is so much more to r600g than quake3 frames per second.

            For example:

            OpenGL 2.1 -> OpenGL 3.0
            OpenCL is starting to work
            Shader-based VDPAU works with MPEG2

            Comment


            • #7
              The Warsow 0.61 frame-rate is lower on more recent versions of Mesa than the 7.9.2 release. The important matter here though is that the OpenGL rendering was incorrect on 7.9.2 R600c, so it's faster but wasn't rendering all textures correctly.
              If the results are invalid, then don't post them!
              The FPS is meaningless if it doesn't render correctly!
              If somebody takes that photo and posts it up out of context, or finds it in a google image search
              without reading the article, then it's just going to mislead people!

              Do us all a favor and make the FPS = 0 or put N/A in the chart.

              Also, does anybody know how to run an OpenArena 0.8.5 benchmark with phoronix test suite? Every time I run it, it tries to run OpenArena 0.8.8. I'm trying to do some of my own comparisons.

              Comment


              • #8
                Originally posted by tstrunk View Post
                Most interesting for me is however the Doom 3 performance, which didn't change at all. D3 really runs into a brick wall with completely similar benchmark results. Finding this bottleneck somehow might be worthwhile.
                Considering that Doom 3 is so heavy in the shaders, I'd guess it has to do with the lack of a decent GLSL shader optimizer / compiler.. A lot of the games that are heavy in shaders, do a shader compile on-the-fly generating shaders that are optimized specifically for your hardware.. These shaders run a lot better than generic pre-compiled shaders.


                The new version of OpenArena 0.8.8 also makes heavy use of shaders / special effects such as "Bloom". Seeing such a big drop in FPS I believe means that there is not enough being done in shader optimizations for the open source drivers. The Catalyst drivers don't drop in FPS anywhere near as much as the open source drivers do when it comes to turning on shaders. So there's still a lot of work to be done...
                Last edited by Sidicas; 08 August 2012, 09:33 PM.

                Comment


                • #9
                  Originally posted by Sidicas View Post
                  Also, does anybody know how to run an OpenArena 0.8.5 benchmark with phoronix test suite? Every time I run it, it tries to run OpenArena 0.8.8. I'm trying to do some of my own comparisons.
                  Run: phoronix-test-suite benchmark openarena-1.4.0 (basically just append the test profile version to end of string as can be seen from the versions at http://openbenchmarking.org/test/pts/openarena openarena-1.5.0 is the latest / new version. If no version is appended, it defaults to latest version).
                  Michael Larabel
                  https://www.michaellarabel.com/

                  Comment


                  • #10
                    Originally posted by pingufunkybeat View Post
                    Shader-based VDPAU works with MPEG2
                    HW-accelerated video decode is nice but I'm not sure why you need it for MPEG2. Even atoms are fast enough for HD MPEG2 :-).

                    Originally posted by Sidicas
                    Considering that Doom 3 is so heavy in the shaders, I'd guess it has to do with the lack of a decent GLSL shader optimizer / compiler...
                    Huh doom3 has a very simple main shader (yes a single one), and a couple of special ones which aren't that complicated neither. Given how hw has improved (alu:tex ratio) you could probably double the arithmetic instructions and it would still run at the same speed...
                    Maybe it's limited by stencil or texture throughput or something, I'm too lazy to dig up other benchmarks to see how it fares in other resolutions (or compared to fglrx). I would expect the recent back-end map fixes to make a difference but those didn't apply to the tested card (HD4850).

                    Comment

                    Working...
                    X