Announcement

Collapse
No announcement yet.

LLVMpipe vs. OpenSWR Software Rendering On A 40 Core / 80 Thread Tyan Server

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

  • LLVMpipe vs. OpenSWR Software Rendering On A 40 Core / 80 Thread Tyan Server

    Phoronix: LLVMpipe vs. OpenSWR Software Rendering On A 40 Core / 80 Thread Tyan Server

    With testing out a Tyan 1U server featuring dual Intel Xeon Gold 6138 CPUs, one of the uncommon test requests we have received but understandable given our audience is curiosity about the performance of OpenGL software rendering on this 40 core / 80 thread Xeon Scalable server when making use of Mesa's LLVMpipe software rasterizer and the newer OpenSWR driver from Intel.

    http://www.phoronix.com/vr.php?view=25103

  • #2
    OMG, it's nearly playable. Impressive!
    I wonder if there's much overhead / "bottleneck" in the dual-socket communication between the two CPUs. Ah, good old times when we had Quake 1 running on a 486 in 320x200. It was still playable. And pure SW rendering. Wasn't something like Half Life 1 the last game that was offering pure SW rendering (okay, maybe also some implementations of DooM source ports like zdoom)?
    Still, impressive at this may be, I'd probably prefer a GPU-"ASIC" for the job of rendering game graphics.
    Stop TCPA, stupid software patents and corrupt politicians!

    Comment


    • #3
      Now that's an interesting test!
      How's the image quality compared to a real GPU? Did you notice anything?

      Comment


      • #4
        Originally posted by mlau View Post
        Now that's an interesting test!
        How's the image quality compared to a real GPU? Did you notice anything?
        Didn't notice any rendering issues.
        Michael Larabel
        http://www.michaellarabel.com/

        Comment


        • #5
          This is exactly the benchmark I have been waiting for... if only AVX512 had been working. Will be waiting a bit more that the build issue gets resolved

          Comment


          • #6
            swr is going to be largely identical to llvmpipe in pixel-heavy workloads. I believe it's the vertex-heavy workloads where it shines due to its different approach to vertex processing parallelization.

            Comment


            • #7
              And how does Threadripper perform in comparison?

              Comment


              • #8
                Not impressed at all by the OpenArena numbers - I've seen similar framerates on my old desktop machine, although I haven't run the PTS test yet, just some single player shenanigans on various maps. But it doesn't seem to scale well. What was the CPU utilization like?

                Comment


                • #9
                  Typo:

                  Originally posted by phoronix View Post
                  AMD Eypc / Threadripper platforms for curiosity sake.

                  Comment


                  • #10
                    Is software Mesa multithreaded enough to make 40/80 core/thread perform better than a single core with a high frequency and ips?

                    Comment

                    Working...
                    X