AWS Thinkbox Discussion Forums

Making better use of Idle time / non-full-workloads

I was wondering if there is a way of better utilizing the cpu and gpu while rendering and making the workload be at 100% as much as possible.

Lots of times, while rendering short frames, the e.g. preparation times are a ~third or more of the total task time, meaning 30s of 10% workload and then 60s of 100% workload. Now I could increase the concurrent tasks to say 2, but these 2 tasks then usually start simultaneously, and interfere with each other.

What I am looking for is a way of having a second concurrent task only use cpu/gpu workload when it is available, so for example in the 30s the first task is preparing.

I was thinking of having two slaves run and one slave have a lower windows-priority, but this does not seem to work.

Any thoughts?

Thanks!

The two Slaves approach is the best situation I’m aware of. Can you expand on why it didn’t work?

There’s some trickery when using Redshift if they’re not configured quite right (it tends to lock if multiple Redshifts are using the GPU), otherwise it should be fine as far as I’m aware.

Hi!
Yes I am trying to do this with Redshift. The Slave starts a commandline app, that I manually set to the lowest priority in the taskmanager. Though since this commandline is restarted on every new job,I have to reset the priority again. I am also not sure if the taskmanager priority setting has any effect on GPU priority, or if it is just to set the cpu priority on Win10?

It will be just CPU priority. You want to give each instance of Redshift exclusive access to the GPU or it can hard lock at the moment.

For setting the affinity, it will trickle down from the parent process, so set it on the Deadline Slave process. That’ll propagate to the sandbox, then Maya.

One customer once used Process Lasso (bitsum.com/). I can’t endorse it and I’m not sure it works, but I bring it up to say that programs exist to modify process priority automatically. Normally for CPU sharing we recommend modifying the CPU affinity, but you’re trying to use all the cores. :slight_smile:

Privacy | Site terms | Cookie preferences