So I have access to a computer cluster that has several cpu nodes that essentially have infinite RAM. However, I am using GPU with 12 GB RAM to render animations in Blender. I was wondering if it were possible to use the infinite RAM of the computer cluster for rendering but using the GPU RAM for computations within renders?
Also, if I am unable to use both the GPU and computer cluster RAM, would you recommend just using the cpu cluster with infinite RAM instead of GPU or is GPU that much more efficient with blender?
At this point in time, all the data required for rendering a scene on the GPU with Cycles is required to be resident in GPU RAM.
What is “effectively infinite RAM” supposed to mean here, exactly? In a cluster, there may be NUMA spread out over multiple nodes, but not all memory that is addressable is equally fast to access. You always want the fastest memory access possible, otherwise performance will suffer tremendously. Figure out which amount of memory is attached directly to each node and stay within that.
Whether GPUs or CPUs are faster of course depends on which CPUs and GPUs there are. I’d guess that for hardware used in a cluster, the speedup for GPU rendering might be 2-3x. It also depends on the scene. You’ll have to try it out to see if it’s worth the hassle.