Hi there,

I am using Deadline 10.x to manage Houdini simulation jobs on my local machine.
I have 2 x Nvidia Geforce RTX GPU cards, and have two slaves set up in Deadline, each with the GPUI affinity set to one of the GPUs (0 and 1).

What I would like to be able to do is kick off multiple Houdini OpenCL simulation jobs to Deadline, ie. Vellum and Pyro jobs, and have 2 jobs running simultaneously, one on each slave, each using one of the 2 GPU cards for the OpenCL simulation.

Is this possible?

It seems each Deadline pluigin is responsible for respecting the GPU affinity or not, and I'm not sure if the Houdini one does for OpenCL jobs.
I also saw some info online about setting certain ENV variables to control which OpenCL device is used, ie:
HOUDINI_OCL_DEVICETYPE = “GPU”
HOUDINI_OCL_VENDOR = “NVIDIA Corporation”
HOUDINI_OCL_DEVICENUMBER = 1

But I'm not sure how I could set those environment variables on a per-slave basis..

Any help will be super appreciated as this could really really streamline my sim workflow, especially as I start adding more GPUs to my system.

Thanks!

not sure if helpful for your situation, but there is this

Houdini 18.0.309 TOP Deadline scheduler now supports GPU affinity setting for OpenCL and Redshift nodes. A new GPU affinity Override section on the parm interface provides the functionality to set GPU affinity per job and worker.

so you can in theory override that per TOP node to make your jobs use different GPUs

Thanks Tomas, they actually added that in from a request of mine last week

..it was for an unrelated Redshift ROP issue, but if it works the same with OpenCL then awesome!