A slave per GPU ?

Deadline
1

Hey,

Is it possible to have a Deadline slave per GPU instead of per machine?

I have a rendernode with 3 GPUS in it, but when I select to use 1 GPU per task (under the Redshift options in the submission panel) the job takes three times longer to complete.

I don’t want 3 GPUs doing one task or 1 GPU doing one task and the other 2 idle. I would like to get all 3 GPUs working on three separate frames individually.

How do I do this?

Thanks!

2

Sure. GPU affinity per Slave:
docs.thinkboxsoftware.com/produc … u-affinity

We also have Redshift GPU “mappings” in Maya / Softimage, but it sounds like you have already discovered this:
docs.thinkboxsoftware.com/produc … ic-options

3

Cheers Mike. That’s perfect. I actually figured that out right after I made this post, but I was stuck in a meeting.

Interestingly, the render times are not much different.

19min50 for a job using 3 GPUs per task
20min30 for a job using 1 GPU per task

I guess it depends on the render… If it was a memory heavy render, then it would probably be a bad idea having three tasks run at once on one machine…

Thanks

4

What exact version are you running of Deadline? We did have a few bugs when controlling GPU affinity, so you might just be hitting one of those.
Also, what application is this? Maya?

5

I think the difference is just due to the startup time…

It’s from Softimage - using Deadline 8.0.9.1

6

Ah, ok. Feel free to post the render logs or send them into support for review. However, the time stamps should make it pretty clear where if any slowdown is occurring for you.