Sending render jobs to Cluster renders on CPU not the GPUs?

[UmamiLandi]

Hello! 
We have a computer that is setup with x4 GPUs (RTX 8000) and when we send render jobs through the "Send to Cluster Queue", that computer starts rendering on the CPU and not the x4 GPUs.

Is there a way to fix that or did we miss something? 

[Christian_Garimberti]

Hi, do you activate the "GPU" rt on the master before to start the cluster?

And\or, what kind of license do you have on the 4gpu machine? Do you know that for 4GPU cluster render you need network 2 x "Pro" or 4 x "Core" licenses.

Best

Chris

[UmamiLandi]

Hi Chris!
Thanks for answering.

Yes, the GPU is enabled for ray-tracing before sending to the cluster queue and we have 2 x Pro licenses since each Pro can only do 2 GPUs. 

I should also add that when we remote into the Cluster workstation and render locally in VRED, we're utilizing all the GPUs, so licensing should work. 

[Christian_Garimberti]

Other "obvious" questions... Are enabled, or better, not disabled, the gpu's in the cluster configuration?

Do you see some errors or messages in the terminal?

Can you start a GPU rendering directly on the 4gpu's machine without use the cluster, just to see if that machine is working correctly?

[UmamiLandi]

Cluster Configurations says "All available" and no errors, everything looks normal.
Terminal doesn't give errors and the job gets sent to the Cluster workstation and it starts to render (CPU).

I can remote into the Cluster workstation and I when I do a local render test in VRED, all x4 GPUs are being used.

[Christian_Garimberti]

It looks strange that the cluster is running in CPU mode while you started a GPU Rt in the master...

For what i know in Vred it is not possible to mix the 2 RT modes...

Could you try, instead of use "Send to Cluster Queue"...

1. start the RT mode in your master
2. go the to cluster configuration and start the cluster
3. run the render with the "render" button

I'm not sure what "Send to Cluster Queue" do if you do not have a cluster server manager running

[UmamiLandi]

That worked, it uses all x4 GPUs that way but still only CPU when "Send to Cluster Queue" is used. 
Thing is, we want to setup Deadline for VRED, to send away render jobs and don't lock up our computers in VRED and let Deadline handle the priority of jobs coming in. 

Since this x4 GPU workstation currenly defaults to CPU in the "Send to Cluster Queue" inside VRED, I'll doubt it will work with GPUs using Deadline.

We can't be the only ones trying to GPU cluster through the Cluster Queue? 
I'm out of ideas...

If not creating a script that tells the VRED Cluster Service to start render with GPU, I don't know...

[Christian_Garimberti]

"Send to Cluster Queue" i don'think is used if you use Deadline. I suppose it uses his procedures and settings.

But i never used Deadline, so i don't know. Try to contact them...

"Send to Cluster Queue" is used if you install and manage your own Vred Cluster Server.

best

Chris

[UmamiLandi]

No that's correct so maybe I should try Deadline instead and see if that works with GPU.
I just thought that if I can get the VRED Cluster Queue work then Deadline should work.

Thanks for helping out! 
I'll get back when I've tested Deadline.