Hi Epic team,
We are evaluating the GPU execution path of the PCG Static Mesh Spawner in UE [version / changelist], targeting [platform and RHI, e.g. Windows / DX12 / SM6].
Based on our testing and our understanding of the current documentation, this path generates a
UPCGProceduralISMCompoent, whereas the CPU path generates conventional ISM components. The GPU path is attractive for its spawning performance, but the generated instances appear to lack support for several features available through the CPU path:
- Collision and physics queries.
- Navigation geometry contribution and NavMesh generation.
- Ray tracing representation.
- Lumen participation.
Our use case requires large numbers of PCG-generated static mesh instances to participate in both gameplay systems and lighting. We would like to retain the performance benefits of GPU spawning while supporting these features.
Could you please clarify the following?
1. Current support and configuration
Are these expected limitations of the GPU Static Mesh Spawner in this UE version, or are there supported settings or workflows that enable any of these features for UPCGProceduralISMCompoent?
For ray tracing and Lumen specifically, could you clarify which paths are supported, if any? We are interested in ray tracing effects, Lumen Hardware Ray Tracing, and Lumen Software Ray Tracing, including any differences in geometry representation or scene participation.
2. Recommended implementation or workaround
Is there a supported way to retain GPU-side point processing and instance generation while also providing the representations required for collision, navigation, ray tracing, and Lumen?
For example, does Epic recommend a hybrid workflow where GPU-generated instance transforms are used to populate conventional ISM/HISM components, or where separate gameplay and rendering representations are maintained? If so, is there an existing API, extension point, or example implementation we should follow?
We understand that some approaches may require GPU-to-CPU readback or additional CPU-side work. Guidance on the expected synchronization costs and performance trade-offs would be helpful.
3. Architectural constraints
Which of these limitations are inherent to the current GPU-driven instancing architecture, and which are features that could be supported through additional engine integration?
If engine modifications are currently required, could you point us toward the relevant systems or recommended extension points?
4. Roadmap
Does Epic plan to add support for these features to the PCG GPU Static Mesh Spawner / UPCGProceduralISMCompoent?
If there is an active plan, could you share the intended scope and any target release or approximate timeframe that can be disclosed? If a timeframe is not available, knowing whether this is planned, under investigation, or currently outside the intended scope would help us choose an implementation strategy.
We would also appreciate knowing whether Epic expects GPU spawning to reach feature parity with the CPU path for these use cases, or whether the two paths are intended to serve different requirements over the longer term.
Thank you for any guidance.