I'm currently doing some test animations of a single character wearing a subd surface dress with cloth simulation. On a single computer the render takes about 13 seconds per frame (GPU rendering) and is using about 20% of cpu and somewhere around 40% of GPU.
Will crowdrender will help with this case? The demos I've seen seem to show it splitting up each frame and sending part of it to each CPU. In my case it seems like most of the time is setup for rendering each frame, so splitting up a frame wouldn't be much help. Will crowdrender sense this when rendering an animation and send each FRAME to a different computer?
Also, what happens with Cloth simulations? I run the animation on the main machine to compute the cloth, then render the animation. Will crowdrender sync the cloth cache to the other machines or do I have to do something to bake and save it first?
Hi Greg,
Yes, you're correct, on both statements, crowd render at the moment is only able to do tile splitting and so each frame is rendered in parts by each computer. We're currently raising funds to replace our current load balancing system with a better one that will include the option to use frame splitting instead of tile splitting. You can read more about that here -> https://www.crowd-render.com/crowdfunding
Hope this answers your question :D
So just to clarify, you are saying that when rendering an ANIMATION crowdrender will not send different frames of the animation to different computers? ie: if I have 4 computers it will split frame 1 four ways, then split frame 2 four ways...etc and NOT send frame 1 to computer 1, frame 2 to computer 2...etc?
Wouldn't it be more efficient to do animations this way, particularly if the frame render times are relatively short?
Hi Greg,
Thanks for posting :)
Ok, so in the case of speedup, 13 seconds is pretty darn good. As you identify if most of the time spent on the render is just setup time, then splitting the render will only marginally reduce the render time. Each node in the system has to do the same setup work so that time can't be improved. The draw time may improve however, but if that is only a small portion of the render time then it likely won't help you much to distribute the render.
As for the cloth simulation, it depends on how you have generated the cache and where. If you have or can, the best way is to locate the cache inside the blend file. Crowdrender does not yet manage external dependencies, though future versions are planned that will.
This means that for now, with the latest version (as at the time of writing this post 016 is the latest), its best to have a packed cache which is in the blend file. The alternative to this would be to locate the cache on either a NAS (Network Attached Storage) device or share it from a designated file server on the network. Then each render node will reference the cache from that computer at render time, provided you have saved the blend file with the cache location set to the path which will be on either your NAS or file server.
Happy to coach you further if this is confusing at all. For the first timer in distributed rendering it can be a little daunting :)
All the best
James