Announcement

Collapse
No announcement yet.

DR - How exactly rendering process splits beetwen stations?

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

  • DR - How exactly rendering process splits beetwen stations?

    Hi everybody!
    I would like to get a clear answer if possible.
    I have main station and four nodes - all dual xeons.
    Let's say a scene with Irradiance map/Primary and Light cache/secondary.
    In Light Cache Number of passes set to 10.
    When I hit the render button it starts this way:
    1 stage. - Building Light Cache;
    2 stage. - Prepass;
    3 stage. - Rendering Image.
    Only on a third stage I can clearly see that all five station doing their job.
    What about the first and second stage? Does their calculations splits the same way?

    Thanks in advance.
    Gleb.

  • #2
    hmmm...i cant recall if the lightcache can be calculated using DR yet. Also consider that your DR servers wont start rendering until they've opened the file, loaded bitmaps and all that. Sometimes when I have very large scenes, my DR servers wont actually start contributed until my local machine has rendered out a good portion of it.
    ____________________________________

    "Sometimes life leaves a hundred dollar bill on your dresser, and you don't realize until later that it's because it fu**ed you."

    Comment


    • #3
      LC is unfortunatly not distributed, the passes option in LC is to make sure you use all available threads on local PC (dual+HT) so set it to 4 if you have HT on.

      ir.map prepasses are distributed, so you should see the node work here.
      i recently changed my workstation and somehow this new one is mush slower posting the data to the nodes and if its just a quick testrender the prepasses is calculated before the nodes kick in.
      a little annoying.

      Comment


      • #4
        Forgot to mention - all my node stations are real-time monitored.
        When I start the render, almost emidiately perfomance monitor
        shows memory usage goes up to 500Mb and both cpu's up to 100% (with the graph sporadically dropping for few seconds).
        This way it stays till the final.
        So if the LCache calculates only on a main machine, what makes those nodes so busy?
        Anyway, thanks for info guys!

        Gleb.

        Comment


        • #5
          There is a way around the delay problem that works for me.

          After you finish the LightCache calculation stop the render. Load the LC from the saved LC file and start the render again. The nodes should start right away. They do when I do this. Bit of a pain, but saves a lot of time.

          Comment


          • #6
            I think they all caluclate their own LC independently?

            TR is right - you have to precaluclate the LC on your fastest machine.
            You can also uncheck "render final image" in the Vray Globals and set the IRR map at -5,-5 or -4,-4 - so it runs it real quick.

            You can of course, use the DR to render the good IRR map on the final pass.
            www.studio2a.co

            Comment


            • #7
              Each node will calculate its own light cache. If the nodes are much slower than the main machine, this may take more time than the rendering itself - in which case, the render nodes won't get any chance to help for the irradiance map and the final rendering.

              Best regards,
              Vlado
              I only act like I know everything, Rogers.

              Comment


              • #8
                Originally posted by dennis
                i recently changed my workstation and somehow this new one is much slower posting the data to the nodes and if its just a quick testrender the prepasses is calculated before the nodes kick in.
                a little annoying.
                .

                aha. that explains the above. any reason to not just calc it on one and pass it around.

                Comment


                • #9
                  Originally posted by dennis
                  aha. that explains the above. any reason to not just calc it on one and pass it around.
                  You mean manually or automatically? There are no problems to do this manually; as for automatically calculating the map on the main machine only - this may be coded as well, but it needs some thought.

                  Best regards,
                  Vlado
                  I only act like I know everything, Rogers.

                  Comment


                  • #10
                    i guess it doesnt have to be, since it will be 100% distributed in the next build right ?

                    Comment


                    • #11
                      I never copy the LC file to my other computers and it works fine. Perhaps they just finish at the same time but I doubt it. Just save the LC file to the scene dir. in max on your main machine and you should be good to go.

                      I do choose the "from file" when using the saved map and not the "switch to saved cache" option, that does not work.

                      Comment

                      Working...
                      X