render-product-device-pinning

To pin a USD RenderProduct to a specific GPU, you author the RenderProduct deviceIds attribute to restrict rendering to specific CUDA-visible device indices, ensuring deterministic rendering behavior in multi-GPU workflows.

GPU selection ambiguity occurs when RenderProduct deviceIds are not explicitly authored. You must restrict USD RenderProducts to specific CUDA-visible device indices while respecting CUDA_VISIBLE_DEVICES remapping to prevent the wrong device from being selected.

Yes, CUDA_VISIBLE_DEVICES remapping affects USD RenderProduct device pinning because the deviceIds attribute must map to the CUDA-visible indices, ensuring the renderer respects the correct remapped GPU constraints for ovrtx workflows.

You configure a multi-GPU CI rendering scene by authoring the RenderProduct deviceIds attribute so the picking workflow consistently targets CUDA-visible GPU 0, leaving Camera or RenderVar prims unaffected by the GPU selection constraints.

Limitations include respecting renderer constraints and ensuring correct deviceIds authoring, as failing to account for CUDA_VISIBLE_DEVICES remapping in multi-GPU environments can break the deterministic behavior of ovrtx rendering workflows.