I have access to a large cluster with 20+ GPUs and I would like to send jobs to render in parallel, or at least split up the frames to send to each node.
I get access to a single virtual desktop to make use of up to 10 nodes, each with 12 CPUs and 2 tesla GPUs (K20, M2070Q, or M2070 - it depends on what’s available at the time).
This means setting it up to render via the command line and farming it out to the nodes.
The render bit seems to work fine as per the standard blender command line rendering, but it turns out that all nodes render a different instance of the same thing at the same time, or they all crash except one which renders the single frame.
Am I trying to do something that’s not possible?
If it should be possible, any thoughts on how to do it?