Multimatte not working on GPU

[peter_hoehsl]

Hi,

Multimatte based on material id is not working on GPU. Is it supposed to or is it not supported? It works fine on CPU.

Thanks,
Peter

[tsvetomira.girginova]

Dear peter_hoehsl, hello.

What are the material types that you are using?
For example, V-Ray Toon Mtl isn't supported. Please take a look also at the limitations described here and here.

Could you provide more details about your case?
Thanks in advance.

Best regards,
TS

[hsz]

Hi Tsvetomira,
Here is a case that doesn't work that isn't described above. Not sure if the same issue that Peter is having but this works on the CPU and not on the GPU

[tsvetomira.girginova]

Okay, will look at it. Thanks, hsz!
peter_hoehsl, feel free to share your case too if it is a different one.

[bdancer]

hsz, this setup looks simple, but it's actually a bit tricky for us; it works on CPU because of how "Rayserver Instancer" resolves materials, and Rayserver Instancer is CPU only, so for GPU different instancer is used (if you uncheck rayserver CPU result will match the unwanted GPU result)...
If it's an option for you to bake sphere to Alembic (and uncheck "pack-n-instance" in copy to points) it'll work.
I'll think of how this could be solved...

[hsz]

Got it - thanks for the workaround, Andrei! hopefully, you can find a solution.

[peter_hoehsl]

Sorry, I've been quite busy.

My issue is very similar to the above example except that I just get black. I'm using a normal Vray Material on normal geometry. Nothing packed and no particles. I can't share anything right now and will have to set up an example when I get a moment.

Object id works though.

[tsvetomira.girginova]

Okay, peter_hoehsl.
We are waiting for the scene in order to investigate further.

[james_gardnerpickett]

Im running into this same problem. using packed copies at sop level, houdini stays fast BUT material IDs not working when using CUDA mix of CPU and GPU. really need the speed up of CUDA! is there any news on this front?

[hsz]

There is a workaround that works at the moment. Have a look.