Announcement

Collapse
No announcement yet.

GPU bucket performance with distributed rendering is really bad

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

  • GPU bucket performance with distributed rendering is really bad

    Tested on both vray next for 3ds max and standalone

    PC1 - 32core 128Gb RAM 1080TI
    PC2 - 8core 64Gb RAM 980TI
    10Gbit LAN, both pc same vray 4.1 and same nvidia 417.22

    4k frame on PC1 - 5m 12s
    4k frame on PC2 - 7m 35s

    4k frame PC1+PC2 - 6:09

    So adding extra machine with another GPU made the render slower. lol.

    I`ve tried dozens of render setting adjustments trying to figure out what`s going on, but nothing. Both GPUs are running at 100% load all the time(checked with GPUz), scene is only geometry and lights, no textures, only gray material, no displacement, no motion blur or dof. There is no excessive traffic over LAN or stuff like that.
    Running distributed, with just the PC2(use local host unchecked) works as expected. But when both of them run together, it`s, uh, eh..useless :/

    Last edited by psanitra; 28-12-2018, 10:13 AM.
    Noemotion.net - www.noemotion.net

    Peter Sanitra - www.psanitra.com

    Noemotionhdrs.net - www.noemotionhdrs.net

  • #2
    As it scales well here with the scenes that we have tested with, having the scene will help us greatly figuring out what is going on.

    Best,
    Blago.
    V-Ray fan.
    Looking busy around GPUs ...
    RTX ON

    Comment


    • #3
      I uploaded this sample scene for you, archive name is : TheCrew_HarryGarage_BenchScene_Ver01_GPU_VR.rar , hope you can find it on your ftp, pm me if something is not clear. good luck
      Noemotion.net - www.noemotion.net

      Peter Sanitra - www.psanitra.com

      Noemotionhdrs.net - www.noemotionhdrs.net

      Comment


      • #4
        Some of our QAs will check it up, matanov Margarita.Stoeva and get back to you.
        Thanks!

        Best,
        Blago.
        V-Ray fan.
        Looking busy around GPUs ...
        RTX ON

        Comment


        • #5
          It's reproducible here. Will get back after some investigation, thanks for the scene!
          If it was that easy, it would have already been done

          Peter Matanov
          Chaos

          Comment


          • #6
            BTW the DR version of the image seems a lot cleaner which sounds logical since it takes more time to render but something's definitely not right here.
            If it was that easy, it would have already been done

            Peter Matanov
            Chaos

            Comment


            • #7
              my suspicion was that you are calling standalone in server mode to pick up the distibuted job, and bucket GPU settings are not being passed correctly from max app to standalone..., but that`s just wild guess. You know better
              Noemotion.net - www.noemotion.net

              Peter Sanitra - www.psanitra.com

              Noemotionhdrs.net - www.noemotionhdrs.net

              Comment


              • #8
                I'm comparing local standalone vs DR, so that's not the case (unfortunately).
                If it was that easy, it would have already been done

                Peter Matanov
                Chaos

                Comment


                • #9
                  matanov Hi, just checking if you found out what the issue was and if it was fixed in any late builds/nighlies... thank you.
                  Noemotion.net - www.noemotion.net

                  Peter Sanitra - www.psanitra.com

                  Noemotionhdrs.net - www.noemotionhdrs.net

                  Comment

                  Working...
                  X