Announcement

Collapse
No announcement yet.

Further Testing Shows More Hope For ATI Gallium3D

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

  • #16
    Awesome, at this pace.
    The 300g drivers will be 'finished' at the end of this year!
    That's fantastic.

    Kudos to all the developers. It was so nice to see the OSS being faster than the proprietary ones in one of the benchmarks. Hopefully, when the developers are done. This will be the case in all benchmarks.

    Comment


    • #17
      Originally posted by pingufunkybeat View Post
      r600g is faster than r600c in pretty much everything here. It's not nearly as optimised as r300g, though, and a proper shader compiler is still not done.
      you're probably right however the situation is different with an 4250 IGP strange thing is that the blob performance on this IGP is way worse than r600g.

      Comment


      • #18
        Seeing is believing

        Now that's why I keep saying libre drivers are always well worth all the waiting. Kudos to everyone involved. I'm using R600g from X.Org Edgers' repository and I can see some real progress there as well. Thanks for opening up, AMD - this way I'm yours forever (along with all my clients who care about more than getting their hardware to work immediately, no matter how).

        Comment


        • #19
          if i remember well the 300g was supposed to be the trainig ground for later drivers right? so keeping that in mind the r600g will surely follow soon. i just wonder why the more powerful r600 cards produce unplayable framerate. if the driver architecture is the same, how can that be? now the tests have been made on amd processors. do the results apply for intel processors aswell?

          besides. i hope these results will get the intel people to work on a gallium driver. why trying to avoid the the inevitable?

          will wayland produce even higher frame rates due to simpler nature?

          Comment


          • #20
            Corruption in 6.png / OpenArena chart:




            Typo on last page:

            The open-source Catalyst driver

            Comment


            • #21
              r600g is a different driver from r300g. The hardware architecture changed so much that a completely new driver was needed.

              Many of these things can be ported to r600g because there are similarities, and the developers understand the problem and how to solve it, but it's not exactly the same infrastructure, as far as I understand it.

              Also, AFAIK, r600 and later chips have a lot more horsepower, but are more complicated to program. r300g is already a mature driver, with lots of optimisations having gone into it. r600g is quite recent, and there's still lots of work to do.

              Comment


              • #22
                When testing the newest open-source bits ... we disabled the swap buffers wait support.
                Isn't this like cheating to make numbers look better? Shouldn't you be testing with out of the box settings only?

                Comment


                • #23
                  Do any of the Gallium drivers yet allow for tweaking GPU speed as well as configuring things like color correction, anti-alias preferences, etc unless otherwise specified and controlled by 3D apps? I was just wondering if all that Catalyst Control Center can do is exposed yet with the open source drivers or not. Extra fluff like that can wait until later of course, but still thought I'd ask.

                  Comment


                  • #24
                    Is there a tool that monitor Freq and Temp for ATI GPU that works with r300g?

                    Comment


                    • #25
                      Isn't this like cheating to make numbers look better? Shouldn't you be testing with out of the box settings only?
                      Correct me if I'm wrong, but I think with that feature enabled, frame rates would cap at the display refresh rate, which would be sort of useless for some of these tests.

                      Comment


                      • #26
                        Thermal info and chip clocking is handled by the drm not the 3D driver, see the bottom of this page:
                        http://wiki.x.org/wiki/RadeonFeature
                        for more info.

                        Comment


                        • #27
                          Originally posted by curaga View Post
                          Typo on last page:
                          The open-source Catalyst driver
                          That's not a typo. That sentence came from the future.

                          PS: Holy crap, the open-source drivers are getting fast!

                          Comment


                          • #28
                            Originally posted by jakubo View Post
                            if i remember well the 300g was supposed to be the trainig ground for later drivers right? so keeping that in mind the r600g will surely follow soon. i just wonder why the more powerful r600 cards produce unplayable framerate. if the driver architecture is the same, how can that be?
                            r300g has been worked on longer has more hw optimizations implemented (tiling, fast Z clears, etc.); r6xx+ support similar features, but they are not fully implemented yet. r300g also has a more optimized shader compiler. Additionally, there have recently been a lot of driver optimizations in r300g that reduce CPU usage; many of these should be ported to r600g eventually. Finally, r3xx-r5xx is DX9/GL2.x class hardware so it maps more directly to the GL2 APIs. r6xx+ is DX10/GL3+ hardware, so there the hardware doesn't match the older APIs as closely and is more complex (and flexible) in general; as such there are a lot more hoops to jump through in the driver to support those APIs.

                            Comment


                            • #29
                              How much do you guys expect the LLVM IR to increase performance before and after the TGSL ditch, compared to the current TGSL-only IR? Or is this some wild experiment that's not in the pipeline for a very long time?

                              Comment


                              • #30
                                llvm is not adapted to GPU and i don't see it being usefull. That being said i am pretty sure shader optimization won't improve much perf in benchmarks from this article.

                                Comment

                                Working...
                                X