Announcement

Collapse
No announcement yet.

The Impact Of Radeon DRM On Linux 3.3, DRM-Next

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

  • phoronix
    started a topic The Impact Of Radeon DRM On Linux 3.3, DRM-Next

    The Impact Of Radeon DRM On Linux 3.3, DRM-Next

    Phoronix: The Impact Of Radeon DRM On Linux 3.3, DRM-Next

    Recently I showed off some Intel DRM benchmarks from the Linux 3.3 kernel along with what will be in the Linux 3.4 kernel via drm-next. These results indicated some performance improvements on the side of Intel Sandy Bridge hardware, but how do the new kernel and Linux 3.4 impact the discrete Radeon graphics? Here's some benchmarks covering that side of the graphics table.

    http://www.phoronix.com/vr.php?view=17170

  • agd5f
    replied
    If anyone is seeing regressions, please file bugs:
    https://bugs.freedesktop.org
    and attach your xorg log and dmesg output. Also, if you could narrow down the problematic component (kernel, ddx, mesa) and bisect it, that would be really helpful.

    Leave a comment:


  • DaemonFC
    replied
    GL_RENDERER = Gallium 0.4 on AMD REDWOOD
    GL_VERSION = 2.1 Mesa 8.0.1
    GL_VENDOR = X.Org

    814 frames in 5.0 seconds = 162.771 FPS
    783 frames in 5.0 seconds = 156.577 FPS
    772 frames in 5.0 seconds = 154.280 FPS
    717 frames in 5.0 seconds = 143.378 FPS
    673 frames in 5.0 seconds = 134.488 FPS

    ----


    From glxgears -info on the Fedora 17 KDE Beta TC2 disc.

    Leave a comment:


  • Fenrin
    replied
    Originally posted by DaemonFC View Post
    [...]
    I use Fedora 17 and a Radeon 3200 IGP. On the current kernel version 3.3.0-0.rc7.git2.3.fc17.x86_64 I don't have any issues with a performance regression. On a rc6 kernel I had much worse fps than ussually in a game, but I didn't test it much, because I avoided this rc6 kernel then. I had 1-2 fps when I used this kernel, and ussually I get around 15 fps with the free Radeon driver (game: regnum online).
    It helps also to use a kernel without debugging options enabled. With this option enabled my system freezes sometimes. I have no idea if Beta TC2 already used the kernel version above or a older version with debugging enabled.

    list of recent fedora kernels
    Last edited by Fenrin; 03-19-2012, 08:09 PM.

    Leave a comment:


  • bridgman
    replied
    Whoops, sorry... I was responding to DaemonFC, should have quoted for clarity. Sounded like he was seeing a much greater regression than you did.

    Leave a comment:


  • Michael
    replied
    Originally posted by bridgman View Post
    First thought is that sounds like more than a performance regression. Maybe check the renderer string ?
    It was all correct. The data is available from OB but if the OpenGL renderer was wrong it would have been reported to the system table, etc.

    Leave a comment:


  • bridgman
    replied
    First thought is that sounds like more than a performance regression. Maybe check the renderer string ?

    Leave a comment:


  • DaemonFC
    replied
    Originally posted by glisse View Post
    Definitly not able to reproduce HD6950 regression you are seeing. F16 drm-next + mesa git from today (also tried with git ddx)
    Regarding Fedora 17, there seems to be a recent and rather severe regression for my Radeon HD 5670. even glxgears can only manage about 100 fps. I know that's not a benchmark per se, but damn...

    Games are unplayable, kwin is barely even usable.

    Do you know what the story is with that and when it might be resolved?

    I checked it out using the Beta TC2 ISO.

    Leave a comment:


  • allquixotic
    replied
    My HD7970 is currently getting 0 fps with the open source graphics stack, so I'd be happy to see performance numbers like Cayman's World of Padman -- that would be infinity times better than what I have now.

    Also, it's very interesting that you think that "The results for Nexuiz are more interesting than Nexuiz". I would tend to agree; Nexuiz is not a very interesting game, but the performance results that you produced, using it, are very interesting indeed. I think you might have typoed this and meant to say something like "The results for Nexuiz are more interesting than Doom 3", but that would be eliding the very deep philosophical point you validly made about the level of intrigue associated with playing the game Nexuiz versus benchmarking it.

    Leave a comment:


  • glisse
    replied
    Definitly not able to reproduce HD6950 regression you are seeing. F16 drm-next + mesa git from today (also tried with git ddx)

    Leave a comment:

Working...
X