Nuke Use GPU on Render

[Soal GIVORD]

Hello,

I woul like to know if it is possible tu use the GPU  of the clients when Nuke needs it, especially for heavy nodes like Zdefocus.  

Because right now, render times are far better when i render on a my local machine  than on the renderfarm even it it has a big GPU.

So i suspect it uses CPU only.

I checked "Use GPU if available" btw.

we are using Nuke 11.0v2 with RR 9.1.12

Thanks

Soal

[RR, Schoenberger]

Hi

 

I know that some Nvidia driver versions do not allow OpenGL if an application is running in background/service mode.
(Nvidia only, Not an issue with AMD)

But CUDA works.
I was not able to confirm if the filter uses CUDA or OpenGL.

 

 

Just to confirm:

You are not rendering on a second Job Thread on a client that does not have a GPU assigned?

 

 

 

You can try to do a test if the background mode is the issue.

Please open Windows service control.

Stop the RoyalRender service.
Start the rrClient in app mode via RR/win__rrClient.bat

 

 

 

 

 

Note that you can see the GPU usage at the end of an render log file:

 

 

 

regards,
Holger Schönberger

 

Please use the rrKnights Tavern
or our support system for new questions.

[Soal GIVORD]

Hello,

Thanks for your answer.

I confirm that Zdefocus (and surely all GPU based nodes in Nuke) is using CUDA.

The good news is that all I had to do was enable “GPU Requiered” on my job to have it compute on the GPU, whether in service or application mode.

The bad news is that it only uses one GPU out of two (see attachment), unlike Nuke interactive where it is using all GPUs.

I checked in rrConfig and both devices are checked on the client.

Do you have any clue on how i can force the job to use all GPUs?

Thanks again

Best regards

Soal

[RR, Schoenberger]

Hi

 

 

> The good news is that all I had to do was enable “GPU Required” on my job

Sorry, I have overseen that the commandline flag  --gpu is triggerd by this switch.

 

 

> only uses one GPU out of two (see attachment)

Just to verify:
Did you check the CUDA usage of the second GPU in Task manager?
As the screenshot shows the GPU overview in the Task manager only.

And that does not include CUDA.
(Or check the RR render log:  )

 

Perhaps it is related to the default Nuke preferences.
I have to see if “enable multi_GPU support” is enabled by default in Nukes preferences.

If it is not enabled by default, then we have to pre-create/modify the preference config files.

 

 

RR 9.1.16 has a GPU fix for a Windows issue. (RAM limitation affected VRAM as well, this was an issue for the second GPU with Redshift).
Perhaps you can update to see if it affects this render as well.

If you cannot update, you could use a separate RR folder with the new version and linked to your current RR folder.
And then start the new rrClient for a small test to see if it fixes this issue.

[Soal GIVORD]

Hello,

Thanks for your reply.

I think i found a solution.

Nuke has a commandline parameter to force multigpu usage (--multigpu).  

Because it seams preferences are not taken into account during farm rendering

So i added it during submission and it works! (see attachment)

Have a good day

Soal

[RR, Schoenberger]

Hi

 

Thanks for the information.

 

The next RR release will set this flag in its render config.

Pre-release version attached.

Please copy it into these 2 folders:

RR\render_apps\_config\

RR\render_apps\_config\rr_original

 

 

The next rrClient will have a new commandline switch <hasGPU> as well.

With this  “--gpu --multigpu” will be set if there is a Nvidia GPU available (or if you set “GPU Required” in the job).

[Soal GIVORD]

That's perfect! thank you