Announcement

Collapse
No announcement yet.

Why Mesa 9.2 Doesn't Work For All Linux Users

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

  • Why Mesa 9.2 Doesn't Work For All Linux Users

    Phoronix: Why Mesa 9.2 Doesn't Work For All Linux Users

    While Mesa 9.2 has some performance improvements and many new features, this open-source 3D graphics library isn't cut for everyone...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Originally posted by phoronix View Post
    Phoronix: Why Mesa 9.2 Doesn't Work For All Linux Users

    While Mesa 9.2 has some performance improvements and many new features, this open-source 3D graphics library isn't cut for everyone...

    http://www.phoronix.com/vr.php?view=MTQ0MTM

    What you can do to help article with links pointing to a TODO or tasks page imminent.

    Comment


    • #3
      Mentioning the version in the title is a bit confusing it makes it seam like 9.1 was better. A more correct title would have been: "Why Mesa Still Doesn't Work For All Linux Users".

      Also what non-mesa opengl software implementation has more features then LLVMpipe?

      Comment


      • #4
        Another day another whiny article. None of the OSS driver devs EVER said that the OSS drivers would ever be more then 60-80% of the performance of the closed source drivers due to all of the things that they can't release and don't have the man power to reverse engineer this decade.

        Comment


        • #5
          Originally posted by Kivada View Post
          Another day another whiny article. None of the OSS driver devs EVER said that the OSS drivers would ever be more then 60-80% of the performance of the closed source drivers due to all of the things that they can't release and don't have the man power to reverse engineer this decade.
          I think it's less of a "can't release" and more of a "lots of missing small and very situational optimizations/hacks" issue.
          I'm pretty sure there isn't much left to "reverse engineer" in the R600 stack. It's mostly tedious optimization work that needs to be done.

          Comment


          • #6
            Don't forget

            Don't forget about this guy, he did crowd funding to implement KHR_debug. That's going to help push Mesa towards OpenGL 4.3, one less required feature to add. Here is his Github where you can follow his implementation. Implementing KHR_debug in Mesa 3D Graphics Library

            Comment


            • #7
              Originally posted by Kivada View Post
              Another day another whiny article. None of the OSS driver devs EVER said that the OSS drivers would ever be more then 60-80% of the performance of the closed source drivers due to all of the things that they can't release and don't have the man power to reverse engineer this decade.
              Then the most we can expect is the performance of a 7850 when buying a 7950 ? I would rather use the closed source drivers even if they are not "pure".

              Comment


              • #8
                Originally posted by wargames View Post
                Then the most we can expect is the performance of a 7850 when buying a 7950 ? I would rather use the closed source drivers even if they are not "pure".
                Those estimates were averages, not "most you can expect". Nobody is saying "this is all you'll get", rather this was an estimate ~6 years ago based on what we expected to be do-able given the anticipated (at the time) size of community & AMD development efforts.

                I asked our architects what to expect assuming a simple shader compiler/translator and code size/complexity which could be both implemented and maintained by a small group of developers, which in turn meant that there would be relatively few "special case" optimizations. The estimate was 60-70% of Catalyst performance on average.

                Change any of those assumptions and you change the estimated performance, of course. Vadim and Marek have already "violated" a couple of assumptions, in a good way, and we have more AMD developers working on the driver than what we assumed in the initial estimate. The combination of llvm and GCN architecture should also help once we get past the learning curve (and AFAICS llvm has a big honkin' learning curve).
                Last edited by bridgman; 22 August 2013, 09:51 AM.
                Test signature

                Comment


                • #9
                  A few GPUs already runs faster with Radeon then Catalyst, at least for some situations.

                  Comment


                  • #10
                    fglrx or radeon?

                    depends the situation, with a ati hd 4850 radeon is better solution if you don t need opengl 3.3 right now, with kernel 3.11 the things are good

                    Comment

                    Working...
                    X