Announcement

Collapse
No announcement yet.

A Major Open-Source ATI Linux Driver Update

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

  • #21
    Originally posted by BlueJayofEvil View Post
    For the Radeon HD 6000 series only up to the 6870 models are supported, correct?
    Has AMD mentioned any updates regarding the 6900 models yet?
    That's correct. When the rest of the 6xxx support was announced, Bridgman mentioned 6900 support would probably be 1-3 months, which would place it right about in time for the 2.6.39 kernel. But who knows, i haven't heard any information since then and it could go slower than expected.

    Comment


    • #22
      Originally posted by smitty3268 View Post
      That's correct. When the rest of the 6xxx support was announced, Bridgman mentioned 6900 support would probably be 1-3 months, which would place it right about in time for the 2.6.39 kernel. But who knows, i haven't heard any information since then and it could go slower than expected.
      Alex has modesetting and shadowfb running on HD69xx, but there's more work required to get acceleration going. If it looks like acceleration is going to take a long time we may push out the modesetting/shadowfb code first, otherwise we'll hold off and put them out together
      Test signature

      Comment


      • #23
        Originally posted by Svartalf View Post
        Depends on the applications and if you're doing something like Compiz. For some applications, it'll bring a good boost in performance. For others, not so much so (as you found in your testing...).
        If that's the case then it shouldn't be that hard for Michael to come up with a benchmark showing that pageflipping indeed 'dramatically speeds up the performance'.

        Because right now as I see it the only basis for his claims are that newer versions of mesa+kernel+ddx are faster in some benchmarks than stock Ubuntu 10.10, and then somehow he arrives at attributing these performance gains to pageflipping.

        How is a big mystery to me. Perhaps it was written in the stars, or his deck of tarot cards told him so .

        The way I see it is that the idea of pageflipping bringing performance gains is nice in theory, but doesn't work in practice, at least not with the open source drivers in my experience. I'd love to be proven wrong though.

        Comment


        • #24
          Maybe some of the devs can explain better why the gain is not that much. From my prespective it is because, framebuffer copy is GPU intensive operation, but the driver is CPU saturated, so GPU has plenty of time to do "unnessesary" things, such as framebuffer coping. Now fgrlx, is way more opitimized, so GPU has no time to do such vram mastrubations.

          Comment


          • #25
            Pageflipping is mainly useful for reducing memory bandwidth and decreasing latency in the 3D pipeline. If you aren't using pageflipping, you have to blit the back buffer to the front buffer which takes both memory bandwidth and 3D engine time. With pageflipping you just update the front and back buffer pointers. The DRI2 code in the xserver doesn't currently provide a mechanism to do non-vsynced pageflipping. This would provide maximum framerates at the expense of display tearing.

            Comment


            • #26
              Originally posted by agd5f View Post
              The DRI2 code in the xserver doesn't currently provide a mechanism to do non-vsynced pageflipping.
              You mean the display server enforces a sane model for updating the display? What a ridiculous idea.

              Comment


              • #27
                Most of the performance loss of 6.13.2 has disappeared with this. Props on that!

                6.14 is still not as fast as 6.13.1 was, so I guess there's still some tuning the VRAM heuristics left.


                [r700 hd4350 web-browsing, if you're curious]

                Comment


                • #28
                  Originally posted by Ex-Cyber View Post
                  You mean the display server enforces a sane model for updating the display? What a ridiculous idea.
                  It's much more user-friendly. I don't like that...

                  Comment


                  • #29
                    Originally posted by bridgman View Post
                    Alex has modesetting and shadowfb running on HD69xx, but there's more work required to get acceleration going.
                    So if I compile from git I could use modesetting? I'm at work right now so I can't check the git commits at the moment.
                    Acceleration I can wait for; it'll come soon enough. Catalyst works on my current install pretty well so I'm in no hurry.

                    Comment


                    • #30
                      Originally posted by BlueJayofEvil View Post
                      So if I compile from git I could use modesetting? I'm at work right now so I can't check the git commits at the moment.
                      We haven't pushed them out yet.

                      Comment

                      Working...
                      X