Announcement

Collapse
No announcement yet.

Gallium Nine Lands Big Optimization Around Dynamic System Memory Buffers

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

  • Gallium Nine Lands Big Optimization Around Dynamic System Memory Buffers

    Phoronix: Gallium Nine Lands Big Optimization Around Dynamic System Memory Buffers

    In addition to recent commits improving Mesa's Gallium3D Direct3D 9 "Nine" state tracker and addressing memory issues with 32-bit games, this D3D9 state tracker is now enjoying another performance optimization helping some games...

    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
    Does anyone know how Gallium nine performance compares to DXVK these days?

    Comment


    • #3
      Wined3d and DXVK do not implement yet the same optimisations for SYSTEMMEM and DEFAULT swvp.

      On the game (Phantasmat) that used to ran 1-2 fps before the optimizations and 34 fps after, dungeon said he got 4 fps with wined3d and 0.5 fps with DXVK.

      But a lot of the hard work is figuring out the expected behaviour for these buffers, and how to implement them in conformance with game expectations and in a performant way. So it's likely similar optimizations will be implemented in wined3d and DXVK in the near future.

      Comment


      • #4
        Same question here: Does anyone know how Gallium nine performance compares to DXVK these days?

        Comment


        • #5
          Originally posted by xcom View Post
          Same question here: Does anyone know how Gallium nine performance compares to DXVK these days?
          Both DXVK and wine-nine-standalone are pretty easy to install. Install one. Test. Uninstall. Install the other. Test. Share your results with the community.

          Comment


          • #6
            I wonder whether this would help GRANDIA HD, which is not as smooth as you would expect. Unfortunately using the standalone Nine with Proton is bit more tedious than it used to be.

            Comment


            • #7
              Originally posted by mannerov View Post

              Both DXVK and wine-nine-standalone are pretty easy to install. Install one. Test. Uninstall. Install the other. Test. Share your results with the community.
              This is the way.

              Comment


              • #8
                Originally posted by Chewi View Post
                I wonder whether this would help GRANDIA HD, which is not as smooth as you would expect. Unfortunately using the standalone Nine with Proton is bit more tedious than it used to be.
                It's not *that* hard. You just need to install it to the prefix and patch proton as below:
                Code:
                --- a/proton
                +++ b/proton
                @@ -545,9 +545,9 @@ class CompatData:
                
                if use_wined3d:
                dxvkfiles = ["dxvk_config"]
                - wined3dfiles = ["d3d11", "d3d10", "d3d10core", "d3d10_1", "d3d9"]
                + wined3dfiles = ["d3d11", "d3d10", "d3d10core", "d3d10_1"]
                else:
                - dxvkfiles = ["dxvk_config", "d3d11", "d3d10", "d3d10core", "d3d10_1", "d3d9"]
                + dxvkfiles = ["dxvk_config", "d3d11", "d3d10", "d3d10core", "d3d10_1"]
                wined3dfiles = []
                
                #if the user asked for dxvk's dxgi (dxgi=n), then copy it into place

                Comment


                • #9
                  Great to see Gallium nine still being worked on, Axel Davy does a great job.

                  Comment


                  • #10
                    Originally posted by user1 View Post
                    Does anyone know how Gallium nine performance compares to DXVK these days?
                    Well since no one wants to do some, here are my few tests on my rx480:

                    3DMark03: 88377 (dxvk) / 95451 (nine)
                    Unigine heaven:
                    DXVK (small artifact in scene 18 and 19) -> min fps 15.2, max fps 198.2, fps 96.7 score 2436
                    Nine -> min fps 31.5, max fps 209.5, fps 106.3, Score 2677

                    hl2 lost coast with fps_max at 999 (somehow it used to work before... now it replays very fast... but well) 264 fps (dxvk) vs 287 fps (nine)

                    FEAR (loading my save, GPU usage is at 100%): 338 fps (dxvk) / 333 fps (nine).

                    I'd say with both you are in good hands. Maybe nine plays a slightly bit better in CPU limited scenarios. BTW this was with a build with debug info for nine and not for dxvk, so nine was penalized there.

                    Comment

                    Working...
                    X