Hey guys,
I've done a lot of forum searching, but there seems to be not a ton of answers on what i'm trying to do.
I am developing a demo at work and am trying to use big 32 and 64 thread servers to do renders very quickly.
I have been playing around with the memory and multiprocessing settings a lot and have found settings that i really like to use.
When i am using a 8 thread processor with multiprocessing i can get my render time from 16 minutes down to 11minutes.
When i am using a 12thread processor with multiprocessing i can get my render time from 14minutes down to 8 minutes.
When i am using a 32 thread processor with multiprocessing i can get my render time from 15 minutes down to 3 minutes!!!
(The 64 thread results are inconsistent.)
So, i'm setting up a network render from a 8 thread part and sending it out to a couple other machines. On the 8 thread system i have 4 SSD's in RAID 0, and for the network i have a 10Gb switch with cat 6 cables. So drive and network speed is NOT an issue.
I've read on quite a few threads that there can be "slow skipping" when people are using multi-processing on each machine on a render farm. I find this to be true.
If i leave the project already rendered. I.E. all of the frames in the image sequence are in the output folder so all the machines will just skip over all of the frames.
-When i have multiprocessing OFF the machines detect those frames and skip over all 600 in about a second.
-When i have multiprocessing ON the machines take about 4 minutes just to skip over the frames. They don't even render anything it just takes 4 minutes for them to realize they are all already done!
Because of this slow skipping the machines do not do a multi-machine render with multiprocessing ON. It seems that it trumps the point of even having multi-machine rendering!
I would really like to be able to take advantage of all of the CPU power on all of my machines. I am trying to get the render that takes 8 minutes on my 12 thread part down to under 1 minutes. This SHOULD be possible by having 4 of my 32 thread machines all doing a network render, because one 32 thread machine can do that same load in 3 minutes.
Is there any way anybody has been able to figure out how to do this? I have a very fast network, very fast source drives, and very fast computers.
In short, all i really need is for the skipping to speed up with multiprocessing ENABLED. Because when it is enabled the skipping is the bottleneck.
I've noticed that Walter Soyka and Dave LaRonde (sorry if i mispelled it) have been a big help on this subject. If i could get either of your input that would be awesome. I've been at this for about a week and a half and i've finally turned to the forums!!!
Thanks in advance. Since i have to take off i will not proof read this but i apologize for any confusing parts.