Hi there,
I’ve been trying to make Flame use my most powerful GPU for better performance, but I can’t get it to use any GPU other than the one on which the monitors are connected.
I tried this environment variable in Flame Setup:
But
Has anyone got any clues on how to make this change,
Thanks 
Depending on how the code is written it may not be possible. If written to only use one GPU and also to be able to move textures for processing and display via the GPU, it may have to the same one.
I haven’t used Linux Flame in a while, but doesn’t Linux Flame only use the primary gfx card (usually for the monitors) for everything, and only Background Reactor can use the second card it finds?
That environment variable you mention (CUDA_VISIBLE_DEVICES) sounds like an NVIDIA variable, and may be used by a developer if they chose to do so. If Autodesk doesn’t support it, it would probably have no effect.
See this link for details on the env var: CUDA Pro Tip: Control GPU Visibility with CUDA_VISIBLE_DEVICES | NVIDIA Technical Blog
Linux doesn’t utilise CUDA. Flame uses Vulkan.
I have seen this question asked before and I have never seen an answer. Long and short is, from my understanding, the GPU card your monitor is plugged into will be the primary GPU and that’s what Flame will use.
Flame Linux certainly leverages the proprietary NVIDIA driver, which would strongly suggest that certain processing pieplines leverage CUDA resources.
I suspect that the machine learning tools rely on some CUDA processing pipelines.
Vulkan and CUDA processing are not mutually exclusive, you may want to render / display in Vulkan but use some CUDA to accelerate pixel processing like filtring, etc.
I think you’re correct that the primary GPU will be the one driving your display.
