Announcement

Collapse
No announcement yet.

AMD Just Squeezed More Workstation Performance Out Of Its RadeonSI Driver

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

  • AMD Just Squeezed More Workstation Performance Out Of Its RadeonSI Driver

    Phoronix: AMD Just Squeezed More Workstation Performance Out Of Its RadeonSI Driver

    While Vulkan is quickly taking over as the dominant graphics API for Linux gamers especially with the likes of DXVK and VKD3D-Proton mapping Direct3D atop Vulkan, OpenGL remains widely used by workstation software. It's also for workstation software where AMD's "PRO" closed-source OpenGL Linux driver has traditionally competed well (and outperformed) the open-source Mesa driver. But with all the recent changes, that's either a matter of the past or close to not being relevant with the latest Mesa enhancements...

    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
    How difficult would it be to reuse the radeonsi driver on Windows instead?
    ## VGA ##
    AMD: X1950XTX, HD3870, HD5870
    Intel: GMA45, HD3000 (Core i5 2500K)

    Comment


    • #3
      #amdfixvega

      Comment


      • #4
        by around 10% for SPECViewPerf's Siemens NX test case. A 10% performance boost for this merge request that has 379 lines of new code while removing 461 lines.
        379/461=0,82

        If there is pure reprogramming of the code, 18% less code meand +10% of performance ))

        Comment


        • #5
          Maybe the PRO driver will become a bundled "blessed" version of mesa

          Comment


          • #6
            Originally posted by darkbasic View Post
            How difficult would it be to reuse the radeonsi driver on Windows instead?
            RadeonSI has two important interfaces. One is the windowing system with GLX etc. That part could probably be solved by adding a lot of stuff for a new window system next to the others.

            The other parts is the graphics card through the AMDGPU Linux DRM interface. This is the really hard part, for many reasons. I guess if AMD tried to reuse it themselves they would at least have documentation for whatever the Windows equivalent is, but I bet it would need a lot of refactoring and complicated changes.

            Comment


            • #7
              Originally posted by Peter Fodrek View Post

              379/461=0,82

              If there is pure reprogramming of the code, 18% less code meand +10% of performance ))
              That's not a meaningful metric. You disregard the effect of using smarter data structures and smarter algorithms.
              If your comparison stood ground, bubble sort would be at least 4x faster than quicksort, that's the difference between their source code size.

              Comment


              • #8
                Originally posted by ernstp View Post

                RadeonSI has two important interfaces. One is the windowing system with GLX etc. That part could probably be solved by adding a lot of stuff for a new window system next to the others.

                The other parts is the graphics card through the AMDGPU Linux DRM interface. This is the really hard part, for many reasons. I guess if AMD tried to reuse it themselves they would at least have documentation for whatever the Windows equivalent is, but I bet it would need a lot of refactoring and complicated changes.
                Besides writing a new winsys, this will also require adapting AMD's LLVM shader compiler which RadeonSi uses to the Windows kernel driver. On Windows all AMD user mode drivers use the proprietary shader compiler.
                Last edited by user1; 10 September 2021, 08:15 AM.

                Comment


                • #9
                  Originally posted by zboszor View Post

                  That's not a meaningful metric. You disregard the effect of using smarter data structures and smarter algorithms.
                  If your comparison stood ground, bubble sort would be at least 4x faster than quicksort, that's the difference between their source code size.

                  It was just a joke using "old computer programing truth" that every code contains one line of code thtra is not needed at least ))

                  Comment


                  • #10
                    Originally posted by darkbasic View Post
                    How difficult would it be to reuse the radeonsi driver on Windows instead?
                    Very. It would mean writing a WDDM driver to bind RadeonSI to the card, and then rewiring everything to adapt to Windows' display model.

                    Comment

                    Working...
                    X