Announcement

Collapse
No announcement yet.

More Radeon Power Management Improvements

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

  • I managed to patch the 2.6.35 kernel so that I can undervolt the CPU and compare in equal conditons the new power saving features. It seems to be working, or at least the fan doesn't start very often. However, I'm confused for I don't see any performance difference between the 'high' and 'low' profiles. I tested OpenArena and a molecular viewer. The latter can also display the frame rate and nothing seems to change between both profiles. Isn't this odd?

    Comment


    • By default with 2.6.35 and xf86-video-ati 6.13.1+ 3D rendering is synced to vertical refresh. you can disable it by setting vblank_mode=0 with mesa from git.

      Comment


      • Not using driconf?

        Comment


        • Yes, you mentioned it before somewhere in this thread. Unfortunately I haven't been able to load the 3D drivers compiled from the git repository. I must be doing something wrong here or there, so I'm just testing with the packages from Debian sid (*).

          In any case, I tested both profiles in scenarios where I get well less than 60 FPS (complex maps, big molecules). Would the vsync issue still influence the results in these conditions? If this is the case that explains everything.

          (*) A quick question. Do the 3D drivers from freedesktop need the latest ddx too? I haven't built the latter for I thought it wasn't needed. Perhaps that was what I was lacking.

          Comment


          • Originally posted by Kano View Post
            Not using driconf?
            driconf doesn't seem to pick it up here. But I'm not sure cause I'm using the packages from the repos. At least, explicitly typing the vblank option throws some information like "environment variable overrided and ignored".

            Comment


            • Originally posted by yotambien View Post
              In any case, I tested both profiles in scenarios where I get well less than 60 FPS (complex maps, big molecules). Would the vsync issue still influence the results in these conditions? If this is the case that explains everything.
              Yes, it does.
              ## VGA ##
              AMD: X1950XTX, HD3870, HD5870
              Intel: GMA45, HD3000 (Core i5 2500K)

              Comment


              • You need some slight changes to your .drirc for dri2. See these bugs for more details:

                Comment


                • Got the bits from git working. I guess my previous mistake was being lazy and not having compiled the 2D driver. However, I can't disable vsync. Setting vblank_mode=0 as an environment variable results in glxgears shouting

                  "ATTENTION: default value of option vblank_mode overridden by environment."

                  twice. If I write a drirc file with a dri2 block as per the comments of that bug I get yet another warning:

                  "ATTENTION: option value of option vblank_mode ignored."

                  Comment


                  • Maybe I wasn't paying much attention before. Building everything again it seems vsync is disabled, although only for applications running on a window. When they go fullscreen I get 60 FPS again. Is this expected?

                    Comment


                    • I believe the default for DRI2 is "tearing avoidance" (trying to ensure that the scanout pointer does not cross over an area of the screen which is being updated during the update) rather than full "sync to vblank".

                      With a small window that can run much faster than a full sync-to-vblank, but for a full-screen app which is not throttled by something like video frame rate the behaviour will be very much like sync-to-vblank.
                      Test signature

                      Comment

                      Working...
                      X