render-product-device-pinning

Pinning RenderProducts to specific CUDA-visible GPUs using the USD deviceIds attribute prevents cross-GPU contention. This device constraint alignment ensures reliable viewport picking in complex multi-GPU scenes.

Yes, USD supports assigning render workloads to specific CUDA devices by authoring the RenderProduct deviceIds attribute. This restricts rendering to designated CUDA-visible device indices and respects CUDA_VISIBLE_DEVICES remapping.

Constrain RenderProducts to specific CUDA-visible GPUs using the USD deviceIds attribute. This aligns rendering with CUDA_VISIBLE_DEVICES remapping to prevent cross-GPU contention and ensures deterministic CI runs.

Author the deviceIds attribute on the RenderProduct in USD before loading the scene. This pins the RenderProduct to a specific CUDA-visible GPU or set of GPUs, enforcing device constraints during rendering.

Restrict RenderProducts to CUDA-visible device indices when scenarios require deterministic rendering, reliable viewport picking, or simplified testing in multi-GPU environments. This prevents cross-GPU contention during rendering.