Announcement

Collapse
No announcement yet.

AMD's R300 Gallium3D Driver Is Looking Good For 2011

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

  • I might add that I also use r600g...

    But the thing is if r300g can be as fast as fglrx then it should be possible for r600g, too. Although I don't really think that any of them could be faster.

    Comment


    • Originally posted by HokTar View Post
      I might add that I also use r600g...

      But the thing is if r300g can be as fast as fglrx then it should be possible for r600g, too. Although I don't really think that any of them could be faster.
      I've tried r600g once recently (as I've had PC with HD4850 for testing) and althrough it's speed was not comparable to fglrx (r600g was faster on HD4850 than r300g on my Radeon 9600 in Xonotic but not as much as hardware is) but all the effects in game worked without problems (even on ultimate settings, which needed very much RAM) and the game was mostly playable in 1280x1024 (monitor could not handle more) on such settings.
      r600c was about (or even less than) half of the r600g performance and had some display corruption problems.

      Comment


      • Originally posted by xeros View Post
        Even without these patches current r300g in Mesa 7.10 git is something about the same (or even better in some cases and a bit less in others) in performance compared to fglrx 9.3
        I can give you some examples where r300g is slower than fglrx. There's some unresolved slowliness and some unimplemented optimizations. If we had more manpower...

        Comment


        • The fact that r300g might be catching up with fglrx is just incredible. I was really not expecting it (not this soon anyway).

          I've just retested OpenArena with r600g + colour tiling. There are some visual artefacts, and I had one crash, but I get a solid 60 fps (capped to monitor refresh) at 1920x1080 now.

          This is really impressive improvement.

          Comment


          • Originally posted by marek View Post
            I can give you some examples where r300g is slower than fglrx. There's some unresolved slowliness and some unimplemented optimizations. If we had more manpower...
            That's the point: some.
            The lone fact that in some/many/most cases it is already on par with fglrx is truly remarkable.

            Thank you!

            P.S.: I seriously hope that you _will_ work on r600g, too!

            Comment


            • I'm sure that a few of us can chip in if you need a r600/r700 card to work with!

              Comment


              • BTW, the amount and frequency of your commits significantly decreased in the past few weeks.
                If you don't mind me asking, is there a specific reason for that?
                - Do you have less time?
                - Are you implementing something in a separate branch?
                - Are the current problems so difficult that it simply takes this much time to solve?

                Of course I might be totally off...

                On a sightly separate note:
                What can we still expect from r300g? Will it have some more features like page flipping, hyper z, msaa which has not yet been completely implemented?

                Comment


                • Originally posted by pingufunkybeat View Post
                  I'm sure that a few of us can chip in if you need a r600/r700 card to work with!
                  While this is certainly true I do expect AMD to ship anybody any kind of requested hardware to be able to develop their drivers.

                  Comment


                  • Yeah, for people with a proven track record, so do I.

                    Comment


                    • Originally posted by pingufunkybeat View Post
                      I'm sure that a few of us can chip in if you need a r600/r700 card to work with!
                      I have such a card, but thanks.

                      Originally posted by HokTar View Post
                      BTW, the amount and frequency of your commits significantly decreased in the past few weeks.
                      If you don't mind me asking, is there a specific reason for that?
                      - Do you have less time?
                      - Are you implementing something in a separate branch?
                      - Are the current problems so difficult that it simply takes this much time to solve?
                      Just less time, because I am a full-time student and on top of that I have a girlfriend. I still work on drivers as time permits.

                      Originally posted by pingufunkybeat View Post
                      What can we still expect from r300g? Will it have some more features like page flipping, hyper z, msaa which has not yet been completely implemented?
                      HyperZ and MSAA are the most likely, and some new OpenGL extensions, but there is more important work to do. Page flipping does not need any code in r300g. First I want to make some tests in piglit because there are some features which we don't know if they work (I think they don't). There's some work I'd like to do in st/mesa to support more texture formats (there are needless fallbacks for some) and accelerate buffer copies (ARB_copy_buffer and ARB_pixel_buffer_object). Having one- and two-channel texture compression (3DC/LATC/RGTC) would also help to reduce bandwidth in the games which use it, e.g. ETQW. I think I am going to dedicate more time to working on st/mesa, which all gallium drivers will benefit from, so that we can better utilize some hardware features in the drivers.

                      Concerning HyperZ, there's at least one bug in the current implementation that might result in corrupted rendering, and also hieararchical zbuffering doesn't work on RV530 even though the hw can do it. HyperZ can be enabled by setting the environment variable RADEON_HYPERZ=1 in case you want to test it. The driver should then print "radeon: HyperZ: YES". Most bandwidth-limited applications that heavily use zbuffering *and* use it properly should be faster.

                      Comment

                      Working...
                      X