Difference in VrayRT CPU vs CUDA

[Akira]

Hi,
I've checked the supported features list for VrayRT and did setup a simple scene with one VRmat and one FastSSS (raytraced solid) and a glass material as well as a carpaint shader.
I've rendered one time with VRayRT on CPU vs CUDA enabled.
The results are quite different.
What did I do wrong?
Thanks,
Akira

[tsvetan.geshev]

Can you post the scene here?

[Akira]

Here you go.

[tsvetan.geshev]

This look like a issue for our developers
I just reproduce the difference and for example the teapot, which uses VRayVRmatMtl renders very differently between RT/CPU and RT/GPU with different materials. I will make some more tests and will let you
know if this will be added to our bug tracking system or if it is fixed in the latest nightly builds.

[tsvetan.geshev]

Just make a note in our bug tracking system, so will keep you informed for any fixes regarding this issue.

[Akira]

I don't know, if it's important but the VRMat Library was created in Maya.
Only the teappt has the VRayVRMat applied.

[tsvetan.geshev]

Hello,

1. For the glass - increase the Trace Depth / GI Depth from the V-Ray RT tab in the settings. Keep in mind that you have to restart V-Ray in order these changes to have affect.
2. For the SSS - Single Scatter can be "None" or "Simple" on the GPU and "scatter gi" is always on. This is written in the log as warning, too.

[vlado]

1. For the glass - increase the Trace Depth / GI Depth from the V-Ray RT tab in the settings. Keep in mind that you have to restart V-Ray in order these changes to have affect.

You have to? I thought it works interactively, at least it does for me. If it doesn't you should make a bug report about it.

Best regards,
Vlado

[tsvetan.geshev]

To be more correct this is only for the nightly builds and it is reported already. For official builds there is no such problem with interactivity.

You have to? I thought it works interactively, at least it does for me. If it doesn't you should make a bug report about it.

Best regards,
Vlado

Also using higher values for reflection in the material will produce very close results when you compare CPU and GPU final renders.