Announcement

Collapse
No announcement yet.

The AMD R600 Gallium3D Driver Is Becoming Quite Fit

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

  • #21
    Originally posted by droste View Post
    the modes also depend on how many screens are attached to the card. (you will probably get higher frequencies in "low"-mode with a second screen attached)
    I've just tried deactivating and unplugging my external screen, same results.

    Comment


    • #22
      There is good explanation to manage pm in the arch wiki : https://wiki.archlinux.org/index.php...th_KMS_enabled

      Comment


      • #23
        Originally posted by allquixotic View Post
        Yeah, it's pretty fit and fast, once you:

        *Turn off SwapbuffersWait
        *Force off DRI vsync
        *Build the driver with -march=native -O3 -flto
        *Compile own kernel with full preemption model

        Unfortunately, even then, bug 37168 keeps me on Catalyst. Every few days I try the latest git master stuff and latest kernel to see if it fixes the bug, but nope. That makes r600g unusable for me.

        I could go back to mesa 7.10.2 where the bug doesn't manifest, but then the performance is horrid, and it doesn't work with the other 3D apps I run.

        It's kinda like how wine works: app "foo" runs fine with wine x.y.z, but with wine x.y.z+1, app "bar" runs fine but "foo" is broken. So you either `make install' a different version of wine each time you want to run one of the different apps, or you just give up and say, eh, too many regressions for me, goodbye.
        care to try building master without the extra flags? just in case

        /me can't seem to get onto SL from work, will have to play around from home at some point.

        Dave.

        Comment


        • #24
          Originally posted by Welsh Dwarf View Post
          I've just tried deactivating and unplugging my external screen, same results.
          Looks like that changed in recent kernel versions and the multimonitor modes are always selected. Now they are also the same for me (not depending on the monitor count).

          low:
          engine clock: 400 MHz
          memory clock: 1200 MHz
          voltage: 950 mV

          mid:
          engine clock: 600 MHz
          memory clock: 1200 MHz
          voltage: 1000 mV

          high:
          current engine clock: 850 MHz
          current memory clock: 1200 MHz
          voltage: 1200 mV

          but I used to have lower frequencies with single monitor in less recent kernel versions. e.g.:

          low:
          engine clock: 156 MHz
          memory clock: 300 MHz
          voltage: 950 mV

          Comment


          • #25
            Just out of interest, can anyone tell me what Catalyst does on top of what radeon does for power management?

            Comment


            • #26
              Originally posted by Welsh Dwarf View Post
              Just out of interest, can anyone tell me what Catalyst does on top of what radeon does for power management?
              I think that it's just lots of small optimisations, rather than specific chunks of functionality that are missing. It is much more aggressive with reclocking, which is difficult to program.

              Comment


              • #27
                Originally posted by Welsh Dwarf View Post
                Just out of interest, can anyone tell me what Catalyst does on top of what radeon does for power management?
                the R600 and up GPUs have a power management unit that needs to be programmed, AMD are looking into releasing source code to drive it, but its a lot of code, and I'm sure someone will claim its a big secret. Maybe someone should just RE it already.

                Dave.

                Comment


                • #28
                  OK, Thanks for the answer, that explains a lot

                  Comment


                  • #29
                    Originally posted by airlied View Post
                    the R600 and up GPUs have a power management unit that needs to be programmed, AMD are looking into releasing source code to drive it, but its a lot of code, and I'm sure someone will claim its a big secret. Maybe someone should just RE it already.
                    Well, that would be interesting to see, for sure. Powersaving and performance are the biggest remaining issues on r700 hardware, IMHO. Improved dynpm would be worth gold.

                    Comment


                    • #30
                      Originally posted by pingufunkybeat View Post
                      Well, that would be interesting to see, for sure. Powersaving and performance are the biggest remaining issues on r700 hardware, IMHO. Improved dynpm would be worth gold.
                      I would love for dynpm to work on multiple monitors.

                      Comment

                      Working...
                      X