GPU Offloading to CPU

[NorthlogicVFX]

Hey there

I would like to receive some clarification on this:

"CPU fallback

In case your scene won’t fit into your GPU RAM limits, you can still render on CPU."

Does this mean, like the text straight on implies, that the scene will fall back completely to the CPU, or does it break the scene into chunks so that the parts "overflowing" from the GPU goes to the CPU? And in which instances this should work? I have tried a scene which will render on CPU but crash on hybrid rendering. Thanks

[savage309]

What is meant is that if you cant render your scene with V-Ray GPU on GPUs, you can still use your CPU as a CUDA device (select only it in the device list, named “C++/CPU”). It will render and the result will be exactly the same as on the GPU. Using only the CPU will be slower, but it is a fallback.

Best,
Blago.

[duykhang]

What is meant is that if you cant render your scene with V-Ray GPU on GPUs, you can still use your CPU as a CUDA device (select only it in the device list, named “C++/CPU”). It will render and the result will be exactly the same as on the GPU. Using only the CPU will be slower, but it is a fallback.

Best,
Blago.

When we render a "GPU Scene" on Hybrid mode with CPU only - working as a Cuda Device, will the scene (geometry, textures...) load to ram - not VRAM ? and it solves the issue of Out of memory ( when the scene does not fit on Vram of GPU )?
Otherwise, in this case when CPU work as a Cuda device with GPU algorithm, how about the speed rendering compare to CPU ( V-ray Next )? Is there any test about this?

[savage309]

If you use only the C++/CPU, only CPU RAM will be used (the GPU VRAM in this case is not used).
If you use both CPU and GPUs, each device will use its own memory (CPU RAM, GPUS their VRAM).

Speed should be comparable.

Best,
Blago.

[NorthlogicVFX]

Thanks for the replies!