Using PCG with GPU Processing
- Beta. Sends PCG processing tasks to GPU to free CPU; efficient for point processing, Runtime Generation, static mesh spawning. Available on a small number of nodes (Copy Points, Static Mesh Spawner, Custom HLSL node). Connected GPU nodes execute together as a Compute Graph.
- "there is a CPU cost for transferring data between the CPU and GPU, and for preparing a compute graph for execution" → group GPU nodes, minimize transfers (yellow up/down arrow badges show uploads/downloads). GPU wins only "when there are enough points in the data to fully utilize the GPU hardware". GPU nodes + GPU Static Mesh Spawner = fast path for mesh spawning.
- Authoring guidelines: use Hierarchical Generation to run shared work on large/Unbounded grids; prefer GPU nodes (compute shaders); minimize CPU↔GPU transfers.
- Custom HLSL node: kernel types (Point Processor — once per point, output initialized from input; Point Generator; ...), HLSL Source Editor, debug options (Dump Cooked HLSL, Print Shader Debug Values, Trigger Render Capture).