Announcement

Collapse
No announcement yet.

AMD Radeon Performance Preview On Linux 3.8

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

  • AMD Radeon Performance Preview On Linux 3.8

    Phoronix: AMD Radeon Performance Preview On Linux 3.8

    With word this week that there's some performance improvements for the open-source Radeon Linux graphics driver to be found with the Linux 3.8 kernel as a result of the a-synchronous DMA engine support, some very early benchmarks of the "drm-next" code were done from five different AMD Radeon graphics cards. In extreme cases, the open-source graphics driver can deliver 10x higher OpenGL frame-rates with the experimental kernel.

    http://www.phoronix.com/vr.php?view=18266

  • #2
    Dear santa...

    performance isn't that important to me - the games i have run perfectly well

    what i'd like to see is my radeonhd 4650 do under/overscan from grub as my tv/monitor sucks balls when it comes to edid although catalyst can compensate from the gui

    Comment


    • #3
      Random note: Xonotic was run with Low settings here, and produced no change. Xonotic on my machine with Ultimate settings tripled its framerate.

      Also, what's up with that 6570? Seems like someone broke something, the question is where...
      Last edited by Veerappan; 12-14-2012, 12:56 PM.

      Comment


      • #4
        Originally posted by Article
        While the a-synchronous DMA engine support is there on the kernel side, for exploiting its full capabilities is also user-space support that has yet to be integrated into the mainline repositories.
        Can this explain some of the problems here, or is that mostly due to the code being new and needing more time to get the bugs worked out?

        Comment


        • #5
          I think I have had a similar bug, but with a different card series... https://bugs.freedesktop.org/show_bug.cgi?id=50655

          Anyway, with that bug I had some tests continue even though they had frozen for a time. PTS was able to detect this because the FPS changed, but then I would need to intervene. In most cases it would just run the test many many times and then call it done.

          Look in the logs for GPU lockup / soft lockup / lookup for 1000ms etc. This might be a good thing for PTS to autodetect... Because it may be in some of your finished results..

          Thanks!
          Bryan

          Comment


          • #6
            Well thankfully, most of the performance regressions seem to be very minor (except with Nexius), and some performances gains are very prominent (when they're there). It does look like something got broke in the HD 5xxx series code though.

            Comment


            • #7
              Originally posted by D0pamine View Post
              Dear santa...

              performance isn't that important to me - the games i have run perfectly well

              what i'd like to see is my radeonhd 4650 do under/overscan from grub as my tv/monitor sucks balls when it comes to edid although catalyst can compensate from the gui
              You can adjust the underscan with xrandr using the underscan, underscan vborder, and underscan hborder parameters.

              Comment


              • #8
                Hmm, I wonder what makes 4770 so special (it has basically only regressed here), and if it is the same when using higher-end R700 cards...

                Comment


                • #9
                  Originally posted by GreatEmerald View Post
                  Hmm, I wonder what makes 4770 so special (it has basically only regressed here), and if it is the same when using higher-end R700 cards...
                  once on some kernel's DRM change for radeons my Radeon's 4730 (RV770CE) had all its video output garbled up, because some small thing for it had to be handled differently.
                  i damn hope, that i will not get 10x slow down now with 3.8 similarly

                  Comment


                  • #10
                    IMHO, these tests should start not only to be tested with discrete video cards but also with APUs like the A8-5500, A8-5600K, A10-5700 and A10-5800K...

                    Comment


                    • #11
                      Originally posted by AJSB View Post
                      IMHO, these tests should start not only to be tested with discrete video cards but also with APUs like the A8-5500, A8-5600K, A10-5700 and A10-5800K...
                      I second that request. btw trying out this new code [drm next tree + mesa git + hyperz patches for drm/mesa] resulted in several system hangs for me on E-350 [HD 6310] playing Ufo-AI. reverting to regular 3.7 kernal brought situation back to normal.

                      Comment


                      • #12
                        On my AMD E350 (Radeon HD 6310) even youtube in 720p doesn't play well (aka unusable), so I would this area to be improved.

                        Comment


                        • #13
                          Originally posted by arekm View Post
                          On my AMD E350 (Radeon HD 6310) even youtube in 720p doesn't play well (aka unusable), so I would this area to be improved.
                          The problem here would be that it has nothing to do with your graphic card. Flash/html5 are both CPU bound with radeon at the moment AFAIK.

                          But great news seeing all this coming in the 3.8 kernel! really looking forward to it.

                          Serafean

                          Comment


                          • #14
                            Smokin' Guns' ten-fold increase is something to go after.

                            Why did Smokin' Guns benefit from the new drivers?

                            Very intriguing.

                            Comment


                            • #15
                              Youtube

                              Originally posted by arekm View Post
                              On my AMD E350 (Radeon HD 6310) even youtube in 720p doesn't play well (aka unusable), so I would this area to be improved.
                              If I watch a youtube video as flash in fullscreen (1920x1080), my laptop goes nuts with the open radeon driver. If I use html5, the CPU uses somewhat 20%.

                              For the open driver, the best is to prefer html5 videos:
                              http://www.youtube.com/html5
                              click 'join the html5 trial' and you get the html5 player if available for the video. I watched nearly all episodes of 'Sledge Hammer!' yesterday in fullscreen on an AMD C-50 netbook with the html5 player (ok, only 1366768).

                              Comment

                              Working...
                              X