Announcement

Collapse
No announcement yet.

Factors effecting render threads (buckets)

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

  • Factors effecting render threads (buckets)

    My co-worker's machine has the same processor as mine (3.06 GHz Xeon) and even less RAM than mine (his 1.5G and mine 2G) yet his machine is blowing mine away in head-to-head renders and he has 2 buckets going in every render where I have only one. He does have a diferent graphics card but I always thought that render was based on processor speed.

    Are there any settings I can change that will give me some speed or is this a hardware thing?

    (BTW we already tried going through our settings in Vray to duplicate them with the same results)

    Thanks,

    Mark

  • #2
    Re: Factors effecting render threads (buckets)

    Sounds like either hyper threading is off on your chip or is not an HT chip at all.

    we have a bunch of dell machines around the building, all appear to be the same but if you really look under the hood some are hyper threaded P4's and some arent.

    what do you get when you view your performance tab? I suppose there could be software related factors but this would be my first guess.





    Comment


    • #3
      Re: Factors effecting render threads (buckets)

      That would be it. Thanks.

      Comment


      • #4
        Re: Factors effecting render threads (buckets)

        The only place where you can really set how many threads will be used is during light cache calculations. Other than that vray should use all available threads for the rendering process. To set the number of threads used for lc calculations you should set the number of phases to the number of threads that you have available for rendering. However, LC calculations do not use buckets, so you wouldn't visually see any difference in the render process (although the vray log tells you that there are more threads to be used and of course the task manager)

        There also may be issues with memory as well. Basically if there is insufficient physical memory then windows uses virtual memory which is a certain amount of data that is written/read to/from a hard disk. Once physical memory is full, then the task uses virtual memory, but it will have to wait for and write the information that goes to the disk. This causes a loss of preformance simply because of the flow of information. Personally I have seen my quad core machine drop to 25% processing when the amount of information simply becomes to large for physical memory. Of course how much memory is available for a single task depends on not only the physical availability but also the amount of other information that is contained within memory, so simply having other programs running while rendering will take up available resources and slow things down.

        As far a Hyper-Threading is concerned, I am not really 100% sure if it benefits really extend to rendering. The whole concept behind HT is to use unavailable processer power to process multiple peices of information. This happens through a virtual separation of different render threads. Basically if a given task only requires 80% of the processors resources, then the other 20% is used for another task. Ultimately this only really expresses performance benefits when combining two tasks with each not requiring 100% of the processors resources. In the case of rendering, which will use all of the available processing resources, the performance increases may be negligible if any at all.
        Damien Alomar<br />Generally Cool Dude

        Comment


        • #5
          Re: Factors effecting render threads (buckets)

          just from the DR tests we've done on our network we do notice that non HT machines of same processor speed render slower than HT machines. I won't say that it's lore or fact but from our general experience there is a difference.

          A few years back we did some in house tests of our Intel chips verses our AMD chips. At that time the AMD's were not hyper threaded or dual cored and we noticed a huge difference in speed between the intels and the amds. The amds were far behind the intels. the only thing we could attribute the speed difference too was the hyper threading and perhaps the larger L2 cache on the intels. Now i know you're probably saying apples and oranges between the intels and amd's but at that time the AMD chips we used should have had the edge when it came to horsepower, but they lost.. While this is a different topic from HT vs Non HT 1 bucket vs 2 still holds true.



          Comment


          • #6
            Re: Factors effecting render threads (buckets)

            Originally posted by Travis
            just from the DR tests we've done on our network we do notice that non HT machines of same processor speed render slower than HT machines. I won't say that it's lore or fact but from our general experience there is a difference.
            I have never actually gone through and been able to do testing on my own end, so if you found a performance increase the I guess it does help. I wonder what the comparison would be between single core/single core w HT/dual core. Although the HT might be faster than the single core, I wouldn't think that it would be toe to toe with a comprable dual core, but I guess we will have to see. HyperThreading is not really used within new technology today, so its not really that big of a deal I guess.
            Damien Alomar<br />Generally Cool Dude

            Comment


            • #7
              Re: Factors effecting render threads (buckets)

              P4's with HT don't stand a chance against the dual cores. it's a slaughter. dual core has been the single greatest improvement in the speed of rendering if you ask me. It's more affordable than dual processors thats for dang sure. My only wish with dual cores is that windows was able to dedicate certain amounts of ram per core. That would be so nice. It's no biggie when rendering but when i have Rhino open, and a couple other cad apps open, everyone wants ram ram ram. sometimes there just isnt enough memory to go around when there is still plenty of processing power.

              Comment


              • #8
                Re: Factors effecting render threads (buckets)

                Vray automatically splits up the available ram equally between threads. In the systems rollout there is an option to set the dynamic memory limit. For static scenes (which for vfr make up 95-8% of scenes right now) this is over looked be cause the scene is precomplied for how it needs to be loaded into the scene (anytime you see "Static Raytrace Accelorator" thats what its doing). However if you are doing any work with displacement (or in the future motion blur, proxies, or fur) this generates dynamic geometry which then references that memory limit.

                Of course there's still the limitations of the amount of memory the program can access, how much the os can handle, how much can be stuffed on a motherboard and lastly how deep your pockets are. ;D

                Forgot to add the Spot3d link http://www.spot3d.com/vray/help/150R....htm#raycaster
                Damien Alomar<br />Generally Cool Dude

                Comment


                • #9
                  Re: Factors effecting render threads (buckets)

                  well when i render i want my engine to use every drop of everything that it can. i was really generalizing about multitasking with multiple cad apps open at one time. dual cores have given us this ability to have a butt load of apps running, but when they all want ram it can be slow going.

                  Comment


                  • #10
                    Re: Factors effecting render threads (buckets)

                    Will it be possible to use the DR so, that the slave is rendering only and the master is free for modelling and other stuff?
                    www.simulacrum.de - visualization for designer and architects

                    Comment


                    • #11
                      Re: Factors effecting render threads (buckets)

                      Originally posted by Micha
                      ...so, that the slave is rendering only and the master is free for modelling and other stuff?
                      that would be cool!
                      --j.andre.

                      Comment


                      • #12
                        Re: Factors effecting render threads (buckets)

                        Originally posted by Micha
                        Will it be possible to use the DR so, that the slave is rendering only and the master is free for modelling and other stuff?
                        I'm not sure if the V-Ray DR is set up like that at the moment. I can ask though
                        Best regards,
                        Joe Bacigalupa
                        Developer

                        Chaos Group

                        Comment

                        Working...
                        X