Announcement

Collapse
No announcement yet.

More Radeon Power Management Improvements

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

  • I see. Is there a way to avoid that 'tearing avoidance' then? I guess people tend to play in full-screen mode...

    Comment


    • --- a/src/radeon_dri2.c
      +++ b/src/radeon_dri2.c
      @@ -315,7 +315,7 @@ radeon_dri2_copy_region(DrawablePtr drawable,
      }

      vsync = info->accel_state->vsync;
      - info->accel_state->vsync = TRUE;
      + //info->accel_state->vsync = TRUE;

      (*gc->ops->CopyArea)(src_drawable, dst_drawable, gc,
      0, 0, drawable->width, drawable->height, 0, 0);

      Comment


      • I have "edid invalid" messages in my dmesg log. Maybe they can give a hint on my problem?
        I am now on gentoo-sources kernel 2.6.35 and still now power_method file. Are there Radeon X1950 Cards without Power Management?

        Originally posted by derdestiller View Post
        I can't seem to find power_method in my sysfs directory.
        Any idea what could be the reason?

        Comment


        • Originally posted by agd5f View Post
          --- a/src/radeon_dri2.c
          +++ b/src/radeon_dri2.c
          @@ -315,7 +315,7 @@ radeon_dri2_copy_region(DrawablePtr drawable,
          }

          vsync = info->accel_state->vsync;
          - info->accel_state->vsync = TRUE;
          + //info->accel_state->vsync = TRUE;

          (*gc->ops->CopyArea)(src_drawable, dst_drawable, gc,
          0, 0, drawable->width, drawable->height, 0, 0);
          Okay...this works.

          I guess it has some drawbacks...I found one application that behaves like glxgears, it tries to display as many frames per second as possible, even when it's meant to be idle waiting for user input. Not ideal. I'm not sure what Celestia is doing, it's also quite high on CPU, but that may be for all the stuff going on behind the scenes. I'll try tomorrow without the patch. GoogleEarth seems like it's behaving OK.

          One thing is for sure, OpenArena is much better with this : D

          Couldn't applications just choose what to do, instead of, well, users patching the driver? In any case, thank you.

          PS. Oh, and with this I could finally see the difference between 'high' and 'low' profiles.

          Comment


          • Celestia's CPU usage is a known issue...

            Originally posted by yotambien View Post
            I'm not sure what Celestia is doing, it's also quite high on CPU, but that may be for all the stuff going on behind the scenes.
            I find that Celestia's CPU usage while I am viewing the Earth drops sharply after I have zoomed out to a distance of approx 48,600km. This is regardless of the rendering path I am using too, from "Basic" to "OpenGL 2.0".

            Other planets presumably have their own "optimal viewing distances".

            Comment


            • Celestia handles vertexes poorly. See this thread for more discussion:
              http://phoronix.com/forums/showthrea...t=24206&page=4

              Comment


              • Yes, I remember that thread.

                Originally posted by agd5f View Post
                Celestia handles vertexes poorly. See this thread for more discussion:
                http://phoronix.com/forums/showthrea...t=24206&page=4
                But however poorly Celestia may handle vertices, some drivers cope better than others. If the blame were entirely Celestia's then surely we'd expect all drivers to suffer equally on the same hardware, wouldn't we?

                Comment


                • Originally posted by derdestiller View Post
                  I have "edid invalid" messages in my dmesg log. Maybe they can give a hint on my problem?
                  I am now on gentoo-sources kernel 2.6.35 and still now power_method file. Are there Radeon X1950 Cards without Power Management?
                  Ah drm.debug=1 helped me to find out: Number of Power States 1. So I guess i need a new Card or Power Supply. And Low Power Mode non kms-Option never worked . But it will be ATI because of the good oss-driver support.

                  Comment


                  • I'm using Ubuntu 10.04 with the 2.6.35 mainline kernel. Pm works great except two issues:
                    1. When i'm watching a movie in fullscreen and with low-profile it stutters. (with default it's ok)
                    2. I'm getting messages like this in dmesg:
                    Code:
                    [11583.075963] radeon 0000:01:00.0: ffff88007bd72000 reserve failed for wait

                    Comment


                    • I wish power management would work with the expected results of lower power consumption in desktop use, alas, it still doesn't work.

                      First encounter of the problem was back in March with drm-next kernel from git and I filed a bug report then:

                      https://bugs.freedesktop.org/show_bug.cgi?id=27206

                      Now with profiles activated, the problem persists. I filed a bug report to the Cooker QA of Mandriva, which is the distribution I am currently using:

                      https://qa.mandriva.com/show_bug.cgi?id=60705

                      I wonder whether anyone with a Thinpad T60p and its FireGL V5200 has the same problem or whether there are actual reports of this hardware working with the latest power management stuff and actually reducing power consumption below 25 watts.

                      Comment


                      • The bug reports you linked to aren't actually about power management "not working" (ie not reducing power consumption), they both relate to screen artifacts when running in the "low" profile. Is your problem best described as "even when running with low profile I am not seeing a power reduction", or is it more like "I have not measured power consumption when running on low because of the display artifacts" ?

                        If you are still seeing the display artifacts with the latest code could you update the fdo ticket ? That's the one the driver developers will see, not a mandriva ticket.

                        Comment


                        • Low mode is known to cause display problems on some laptops which is why I initially disallowed it when displays were active and only later added it as an option for users to force low mode despite potential problems. auto switches between mid and high depending on ac state. Unless low is explicitly forced on, the driver will only use it when the displays are off (dpms active).

                          Comment


                          • If you are having problems with low, try mid instead.

                            Comment


                            • Originally posted by agd5f View Post
                              If you are having problems with low, try mid instead.
                              Thanks for the advice.

                              I tried new_pll={0,1} at "low" setting without any change.

                              Mowever, "mid" works fine now and actually reduces the GPU temperature by roughly 4 degrees. This is both true with my Mandriva setup and the recent Fedora 14 Alpha, so there has been some trmendous work going on and I'd like to thank you for that.

                              I don't want to open a new thread for this, because it seems rather specific to my hardware, but despite many people praising gallium 3D performance, on F14 (and previous experiments with mesa git), it's pretty abysmal on my FireGL V5200. No game works (Wolfenstein, Serous Sam), Google Earth stutters like hell and while usually I use mplayer with -vo gl for 720p movies with gallium 3D enabled this simply stutters along at 1-3 fps.

                              I understand that some people are really happy with gallium 3D performance, so I wonder if there is anything I can do to have this same wonderful experience?

                              Comment


                              • low/mid are not valid options for new_pll. new_pll has nothing to do with power management.

                                Comment

                                Working...
                                X