AWS Thinkbox Discussion Forums

Moving on to new job while still finishing tasks

Deadline
#1

I have a RenderBox Machine with 4x 4090s.

I have a few questions. Using Redshift and Cinema 4D.

  1. I have 4 concurrent tasks usually. One GPU per task. If there is only one Task left on a Job. Then 3 GPUs will sit idle while waiting for that job to finish before they will move onto the next job. Is there a way to fix this so that the 3 that are done start a new job?

  2. One of my GPUs (GPU 2) Doesn’t seem to be working as well as the others. I’m trying to find ways to tell Deadline to only use GPU 0, GPU 1 and GPU 3

I’ve tried overriding the Worker properities with GPU affinity. However the result was only GPU 0 could successfully render and GPU 1 and 3 just failed jobs.

In the submission option there’s a option for “Select GPU Device” However I don’t know how to go about selecting them. I tried “0,1,3” I’m assuming it’s a list but How do I know exactly what to enter?

#2

Workers won’t pick up tasks from another Job until it’s done. You could run multiple Workers on the machine, one for each GPU.

Our Redshift integration supports GPU affinity, so you can restrict each Worker to a given GPU.

The only thing to keep in mind is the CPU will be shared across processes, keep on eye on things to be sure the machine doesn’t get overwhelmed.

The GPU number is just passed directly to Redshift, so whatever number Redshift thinks your problem GPU is is what you’d want to avoid.

Privacy | Site terms | Cookie preferences