We’ve determined that the issue described in this is due to two instances of C4D Commandline fighting to access files at the same timestep, because DL regularly distributes separate tasks to the GPUs in the same box at the exact same timestamp.
Is it possible to adjust a file so that DL would offset task distribution to GPUs on the same machine by as little as 1-2 seconds? It would save us an incredible amount of pain.
Thanks!
Luke
You could use throttling?
I’m not sure that would work in our scenario, as we don’t submit project files through the deadline server (they get placed on shared storage directly). Plus, that seems to be per-machine vs. per command line instance.
A good question would be: How is ‘Worker’ defined here? Individual machine? or the rendering instances on that machine?
I thought the file copy was just an example of usage, it could also be used to manage power usage to stop a cabinet full of kit all ramping up the power draw at the same time etc.
If you set these setting to 1 and then set a delay does this stop the workers running at the same time?
The tool tips seem quite specific in that regard, but I’ll give it a check.
At the moment I’m waiting for the farm to slow down so that I can run some more tests…
