Announcement

Collapse
No announcement yet.

RadeonSI With OpenGL 4 Showing Nice Performance Against Catalyst

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

  • #31
    Originally posted by vadimg View Post
    Thanks. So, in short, what's a problem with it? I've proposed that solution since 2011, though i didn't implement it myself. Is it still not considered good enough? Does it need more work, maybe known issues? Or just a lack of support?

    €Is the problem is that no one has force to push it though?
    €‹The shader cache really needs to be pushed, if it helps.
    Well there are a couple of issues listed in that post that need to be finished off, beyond that I'm not sure. I think the problem is really just that no one is working on it, Carl seems to have moved on from the Mesa project.

    Comment


    • #32
      First we need to have fewer shader variants, then we can expose binary shaders to the upper layers of Mesa, and then we can easily enable the shader cache.

      Comment


      • #33
        Originally posted by marek View Post
        First we need to have fewer shader variants, then we can expose binary shaders to the upper layers of Mesa, and then we can easily enable the shader cache.
        Hi, Marek!
        But what is a problem with shader variants? They can be exposed as a single binary, in theory. I.e. binary would contain all compiled variants and the keys. Is it problematic for some reason? Other than it's not implemented yet?

        Comment


        • #34
          I'm not working on the drivers for a long time since I've got proprietary job, but I'm still trying to use them sometimes for the games, and then I'm hitting freezes due to the long compilation time. I'm kinda solved it for me (or at least reduced) for r600g with custom backend, but I'm not ready yet to solve it the same way for radeonsi.
          Marek, do you think it's possible to solve it for radeonsi with the current backend? AFAIK catalyst uses LLVM as well (on windows too?), but there are are no such issues, and AFAICS it doesn't use any caches, just a multithreaded compilation. 

          Comment


          • #35
            Originally posted by jrch2k8 View Post
            i believe AMD should drop in catalyst(Next Version) support for Desktop GPUs and focus that driver only on RHEL with FireGL cards, move some catalyst developers out into the OSS drivers and focus on that for Desktop users. I mean Radeonsi already support most features the hardware offers(UVD,VDE, DPM, HDMI/S, DP, 2D accel, etc.) better than catalyst.

            Only missing items would be OpenCL 1.2+(should be close and 2.0 should be there when HMM lands) and Opengl 4.3+(4.2 is 99.99% done since the last feature has a patch in the works) that is pretty damn close(probably after compute shader 4.4 is just done too and most of 4.5 if not all)

            I believe in the long run this move will pay off the better and it will be cheaper for AMD too(and probably will boost radeonsi dev speed) instead on diving the effort in 2 semi complete drivers
            In the long run yes, but right now they can't. radeonsi is still very unstable on many GPUs. I've had https://bugs.freedesktop.org/show_bug.cgi?id=90481 on my HD 7970M too, several games in wine with nine hang the GPU, a webvr demo in firefox nightly recently produced a GPU lockup, The Talos Principle had similar lockups. Etc.

            Originally posted by dungeon View Post
            About CS:GO and Borderlands2, both runs fine with profiles on Catalyst... otherwise performance of those two are similar like in mesa, maybe even lower on some chips
            Sorry, what do you mean? Mesa performance is similar to catalyst performance in csgo?
            Here is a fresh comparison screenshot from wine 1.7.48 with nine: https://i.imgur.com/JN4TXUN.png. While natively it drops frequently to ~40fps and below it barely touches 60 fps with nine. And I think it may even have to do with sound, because it sort of crackles when the fps drops happen with wine.

            Comment


            • #36
              Originally posted by dungeon View Post

              No, those are just eON wrapper switches. That is just what should be done by benchmarker before he start benchmarking. He needs to make sure everything is at least visually correctly emited on the screen and not under any condition to benchmark mess on the screen.

              With radeonsi currently if someone run phoronix Bioshock benchmark and don't apply game switch or not run it under mesa env var... benchmark will still run fine with rendering completely broken and he will see huge numbers maybe which means nothing. Same goes to Catalyst without a switch (another one) there are rendering artifacts and benchmark also run slower.

              Of course I used the proper env var plus mesa patches to make Bioshock Infinite work on radeonsi.
              ## VGA ##
              AMD: X1950XTX, HD3870, HD5870
              Intel: GMA45, HD3000 (Core i5 2500K)

              Comment


              • #37
                Originally posted by haagch View Post
                @darkbasic How does csgo run for you? Stable 60+ fps? (I always test arms race and it drops to 40 fps all the time)
                I don't test it since quite a few time. Initially it was on par with catalyst (http://www.linuxsystems.it/2014/09/c...-par-catalyst/) but then it started being slower and slower, I don't know how it's now.
                ## VGA ##
                AMD: X1950XTX, HD3870, HD5870
                Intel: GMA45, HD3000 (Core i5 2500K)

                Comment


                • #38
                  Originally posted by haagch View Post
                  In the long run yes, but right now they can't. radeonsi is still very unstable on many GPUs. I've had https://bugs.freedesktop.org/show_bug.cgi?id=90481 on my HD 7970M too, several games in wine with nine hang the GPU, a webvr demo in firefox nightly recently produced a GPU lockup, The Talos Principle had similar lockups. Etc.
                  Croteam currently asking/tracking those sort of issues which seems to happen on any platform on any vendor GPU drivers, it is top pinned theme right now:

                  http://steamcommunity.com/app/257510...7867770168382/

                  Yes i have those sort of issue in menus when use mesa too, with SS3 too. But you can avoid it if you don't touch menus too much (or better don't apply anything) as game play is not affected just something very weird and locky happens in those menus

                  Sorry, what do you mean? Mesa performance is similar to catalyst performance in csgo?
                  Yes, CS:GO does not have Catalyst profile... but it can be boosted with hl2_linux profile. Without that you can usually see better perf on radeonsi for as example on Hawaii.

                  Here is a fresh comparison screenshot from wine 1.7.48 with nine: https://i.imgur.com/JN4TXUN.png. While natively it drops frequently to ~40fps and below it barely touches 60 fps with nine. And I think it may even have to do with sound, because it sort of crackles when the fps drops happen with wine.
                  Well nine is about DX9 variant of CS:GO... while native because it is newer should be somewhat slower and also is capped without using profile even with Catalyst.

                  What i am trying to say that is how it should be for now, without shader cache mentioned here, without awfull VGPR hickups softened at least, without threaded optimizations (which i think mesa opengl badly needed these days) and various little optimizations, bugs and regressions fixed... that is how it should be - similar like capped flat unprofiled Catalyst cases
                  Last edited by dungeon; 22 August 2015, 09:38 PM.

                  Comment


                  • #39
                    So comparison with more HUD settings. Not very scientific, because I simply played without reproducing anything, but to get an idea.

                    Shoots native: https://i.imgur.com/AbMkUbr.png
                    Shoots nine: https://i.imgur.com/rRcUIsX.png

                    Lake native: https://i.imgur.com/mJ9G7TW.png
                    Lake nine: https://i.imgur.com/gRLjAai.png

                    I haven't calculated the percentages, but from the looks of it the better performance could probably to large parts be simply scaled with better GPU load.
                    What exactly does the buffer wait time show? Because that one is a lot larger on opengl native.
                    With nine it also uses a lot less GTT and quite some bit less VRAM. Hm...

                    Comment


                    • #40
                      You know, try mesa 10.6 branch maybe there is some perf regression . I know of three commits which slowed down various things... well even that hack to raise pipe value which make Bioshock working also slowing down other games

                      So too much "works for me" patches oh my opensource in hurry GL 4 boys ... and that GL_ARB_copy_image also not needed because game has a disable switch, what to say
                      Last edited by dungeon; 22 August 2015, 10:28 PM.

                      Comment

                      Working...
                      X