Announcement

Collapse
No announcement yet.

July 2011 Open-Source Graphics Driver Comparison

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

  • July 2011 Open-Source Graphics Driver Comparison

    Phoronix: July 2011 Open-Source Graphics Driver Comparison

    Following last week's benchmarks of Intel's New Sandy Bridge Acceleration architecture with the very latest open-source driver code, it was decided to throw a few NVIDIA and ATI/AMD graphics cards into the mix to see where the open-source driver performance is comparatively at for some other hardware. This article presents these Linux graphics results for eight configurations.

    http://www.phoronix.com/vr.php?view=16248

  • #2
    The GeForce 9500GT had a surprisingly strong lead over the competition when it came to running OpenArena, which may be due in part to the Nouveau driver not dealing with waiting for swap buffers and the other "features" of the Radeon driver.
    So swap buffers were on again for radeon cards?

    And why were all radeons were capped to vsync, and other cards were not? Because that's the default? "Surprising lead" indeed....

    I'm looking forward to the proper set of tests.
    Last edited by pingufunkybeat; 07-22-2011, 11:24 AM.

    Comment


    • #3
      Curious how much the 5450 loses to 4550, they are near identical cards except for DDR2 vs DDR3. One would think at the low-end bandwidth wouldn't play that big a part.

      Comment


      • #4
        It would be interesting to also have a graph showing results as a fraction of the blob results, so the hardware differences were eliminated (as much as is possible). Then we could see how the open source drivers compare to each other, and how well they drive different hardware generations.

        Comment


        • #5
          Originally posted by curaga View Post
          Curious how much the 5450 loses to 4550, they are near identical cards except for DDR2 vs DDR3. One would think at the low-end bandwidth wouldn't play that big a part.
          Bandwidth is a pretty huge issue everywhere, honestly. Even low-end GPUs have processing power that can do an awful lot of computations in the time it takes to do even a single simple memory fetch.

          It's actually pretty hard to saturate a modern GPU's processing power (that is, not Intel) compared to how easy it is to saturate the GPU's bandwidth.

          Comment


          • #6
            Did you apply xorg patches? Otherwise it's useless to test sna...
            You got 19.66 fps in nexuiz, I got 25,46 fps.
            You got 47,68 fps in openarena, I got 98,50 fps.
            You got 43.4 fps in world of padman, I got 114,13 fps. 3x!
            http://openbenchmarking.org/result/1...DARK-THEBIGO11
            ## VGA ##
            AMD: X1950XTX, HD3870, HD5870
            Intel: GMA45, HD3000 (Core i5 2500K)

            Comment


            • #7
              Originally posted by darkbasic View Post
              Did you apply xorg patches? Otherwise it's useless to test sna...
              You got 19.66 fps in nexuiz, I got 25,46 fps.
              You got 47,68 fps in openarena, I got 98,50 fps.
              You got 43.4 fps in world of padman, I got 114,13 fps. 3x!
              http://openbenchmarking.org/result/1...DARK-THEBIGO11
              What patches? I use Gentoo Linux. Are they in the portage tree?

              Comment


              • #8
                No, they aren't. You will have to apply 8 patches to the xorg-server-9999 ebuild.
                If you have an old chip like GM45 the boost will be even greater with sna!
                https://bugs.freedesktop.org/show_bug.cgi?id=39261
                Be aware that some bugs may bite
                ## VGA ##
                AMD: X1950XTX, HD3870, HD5870
                Intel: GMA45, HD3000 (Core i5 2500K)

                Comment


                • #9
                  These are very important benchmarks. This way we can see whether the OSS radeon drivers, with the support and documentation they get from AMD, are indeed better than the Nouveau one, which gets no support from NVidia.

                  Comment


                  • #10
                    Originally posted by RealNC View Post
                    These are very important benchmarks. This way we can see whether the OSS radeon drivers, with the support and documentation they get from AMD, are indeed better than the Nouveau one, which gets no support from NVidia.
                    In theory, yes.

                    In practice, they only show whether nouveau is faster than 60Hz, since all radeon tests are capped at 60Hz.

                    Comment


                    • #11
                      Originally posted by pingufunkybeat View Post
                      In theory, yes.

                      In practice, they only show whether nouveau is faster than 60Hz, since all radeon tests are capped at 60Hz.
                      Thanks, I was wondering
                      If you check here:
                      http://www.phoronix.com/scan.php?pag...mparison&num=2
                      and then here:
                      http://www.phoronix.com/scan.php?pag...000_open&num=2
                      same hardware (HD6570) and older driver but ~20 fps difference

                      For OpenArena the difference is ~80 fps ^^
                      Last edited by disi; 07-24-2011, 04:02 PM.

                      Comment


                      • #12
                        Originally posted by curaga View Post
                        Curious how much the 5450 loses to 4550, they are near identical cards except for DDR2 vs DDR3. One would think at the low-end bandwidth wouldn't play that big a part.
                        Bandwidth matters allot more on the low end then anywhere else, since the bitrate is often 64 bit, you get all kinds of werid combos of bitrate and ram types, basically if youu want something more then a glorified frame buffer never go below 128-bit GRRD3, GDDR5 would be even better since it's quad pumped you have near the effective bandwidth of 256 bit GDDR3.

                        But yeah, almost universally low end cards gain more from a bandwidth increase then they do from a GPU core clock increase, the reason is something has to feed the GPU. On higher end chips theres often so much bandwith that OCing the ram gets you nothing, just as is seen in traditional desktops. The same is not true for AMD's Fusion APUs, where the GPU shares the memory bus with the CPU, but has so much extra grunt that the memory bandwidth holds it back in a big way if you pair it up with slow bargain basement ram.

                        Comment

                        Working...
                        X