tile rendering

Deadline
1

Got it all working nice and easy, no problems there.
But here is what I’m looking into.

Rendering with redshift and GPUs.
Got two computers each with 4 GPUs in them as slaves.
Now when frame is submitted to rendering for example 3x3 tiles it submits 9 jobs each with single task.
Only way to render is to have all 4 GPUs rendering each task so it renders only 2 tiles at the time, each by one computer.
If I use 2 GPU per task and two concurrent tasks then, because there is only single task in each job it will rendering only with 2 GPUs and two others will be waiting as there is no other task to run concurrent one.
I know it is a bit messy but here is what I’m trying to get at.

I know that there should be option to run 2 slaves in each computer, and each slave having assigned 2 GPUs.
That way I would have efficiently 4 slaves runing, and each of them using 2 GPUs.
So how would I go with doing that and having every time each slave settings saved? :slight_smile:

So instead:
Comp1_Slave1- GPUs 0,1,2,3
Comp2_Slave1- GPUs 0,1,2,3

I would have
Comp1_Slave1- GPUs 0,1
Comp1_Slave2 - GPUs 2,3
Comp2_Slave1- GPUs 0,1
Comp2_Slave2 - GPUs 2,3

2

first sry somehow got double post. anyway. I know how to run additional slaves my question should be how to assign specific GPUs to each slave
Thanks

3

Hello,

Glad you found the multiple slave documentation. In Deadline 7.2 we have GPU affinity(docs.thinkboxsoftware.com/produc … u-affinity) which will allow you to assign specific slave instances to specific GPUs.