Announcement

Collapse
No announcement yet.

NVIDIA GTX 680 To GTX 1080 Blender OpenCL Benchmarks

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

  • NVIDIA GTX 680 To GTX 1080 Blender OpenCL Benchmarks

    Phoronix: NVIDIA GTX 680 To GTX 1080 Blender OpenCL Benchmarks

    With having published some Darktable OpenCL benchmarks at the beginning of the week with 20 different GPUs (plus more follow-up benchmarks), it didn't take long before requests came in to see some fresh OpenCL Blender benchmarks.

    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
    Please note that OpenCL is not popular in Blender due to shadows on transparencies not working. A direct comparison in render times probably cannot be made with these scenes. This is being worked on with no ETA. I have noticed that while being designed with CUDA in mind, Pascal is also quite impressive in OpenCL. I'd like to see how it compares to

    Comment


    • #3
      Originally posted by imperfectlink View Post
      Please note that OpenCL is not popular in Blender due to shadows on transparencies not working. A direct comparison in render times probably cannot be made with these scenes. This is being worked on with no ETA. I have noticed that while being designed with CUDA in mind, Pascal is also quite impressive in OpenCL. I'd like to see how it compares to
      To clarify, are you saying that shadows on transparencies specifically don't work on OpenCL rendering but do work rendering via CPU or CUDA? If so, that's a pretty weird anomaly.

      But yes, seeing CUDA results would be interesting, especially when compared to AMD OpenCL results.

      Comment


      • #4
        Originally posted by schmidtbag View Post
        To clarify, are you saying that shadows on transparencies specifically don't work on OpenCL rendering but do work rendering via CPU or CUDA? If so, that's a pretty weird anomaly.

        But yes, seeing CUDA results would be interesting, especially when compared to AMD OpenCL results.
        Yes, that's what I'm saying. It's not an anomaly since they are two separate kernels that don't have feature parity.

        Comment


        • #5
          The CUDA one shouldn't exist at all in an open source project.
          ## VGA ##
          AMD: X1950XTX, HD3870, HD5870
          Intel: GMA45, HD3000 (Core i5 2500K)

          Comment


          • #6
            Michael Remembering results for the BMW test file, I assume you left tile size untouched? It's probably one of the most important parameters. The default in this BMW file is way too low and so are the results.

            My Hawaii does this scene in 1:00 (1:40 currently with driver regression). GTX 1080 has ~50% more FLOPs, so there is clearly something wrong.

            Comment


            • #7
              Originally posted by schmidtbag View Post
              But yes, seeing CUDA results would be interesting, especially when compared to AMD OpenCL results.
              IMHO, priority should be first comparing OpenCL on different hardware and then add CUDA to the comparison eventually.

              Comment


              • #8
                https://code.blender.org/2016/11/ble...roject-status/
                Originally posted by schmidtbag View Post
                To clarify, are you saying that shadows on transparencies specifically don't work on OpenCL rendering but do work rendering via CPU or CUDA? If so, that's a pretty weird anomaly.

                But yes, seeing CUDA results would be interesting, especially when compared to AMD OpenCL results.
                Note: that won't be the case very much longer.

                Comment


                • #9
                  Originally posted by Marc Driftmeyer View Post
                  If you consider a year not much longer then yes. If you need to render with cycles today, don't buy AMD.

                  Comment


                  • #10
                    Originally posted by phoronix View Post
                    Phoronix: NVIDIA GTX 680 To GTX 1080 Blender OpenCL Benchmarks
                    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

                    > ProcessorIntel Xeon E3-1280 v5 @ 4.00GHz (8 Cores)
                    How high are the CPU scores by comparison?

                    Comment

                    Working...
                    X