GPU Render Backend
Select Render Backend on the Deformable Mesh Component to choose how the visible mesh is deformed and rendered. The gameplay component remains the same: hit registration, events, protection boxes and forwarded hits still belong to the DMC.
| Backend | RealtimeMesh enabled | Visible mesh |
|---|---|---|
| GPU | Yes or no | The plugin's own GPU renderer |
| CPU (Deprecated) | Yes, in the compiled plugin configuration | RealtimeMesh |
| CPU (Deprecated) | No | ProceduralMesh |
Enabling RealtimeMesh does not select GPU deformation. It controls the visual backend used in CPU mode. The CPU backend remains available for existing projects; Deprecated does not mean it has already been removed.
Setup
- Prepare a Deformable Mesh with the LODs you intend to use.
- Follow the existing component setup, including the source mesh's collision and hit events.
- Assign the prepared DataAsset and select GPU as the render backend.
- Test an impact in Play mode and check the visible mesh, scratch material and reset.
The GPU path uses prepared DataAsset data. It does not automatically extract a replacement mesh from the parent at runtime.
How it works
The GPU Deformation Subsystem collects hit commands and submits compute work to update vertex positions and damage colors. The visual mesh does not rely on a CPU ProceduralMesh or RealtimeMesh rebuild after every hit.
Batching does not turn all vehicles or attachment parts into one mesh or one GPU dispatch. Each DMC retains its own renderer and mesh-buffer ownership.
Moving vertex deformation to compute shaders reduces the need for CPU mesh rebuilds and is intended for workloads such as large meshes, many hits and multiple deformable vehicles. Actual gains depend on the workload and rendering cost; benchmark your own scene rather than assuming GPU is always faster.
Direction-Based Deformation
The GPU shader orients dents using the collision's impact direction and combines inward compression with asymmetric, lateral and noise-driven deformation. See Direction-Based Damage for collision normals, manual-hit fallback, repeated hits and the difference from CPU deformation toward an origin.
GPU mode does not silently switch to CPU deformation when a hit cannot be submitted yet. Eligible hits can wait for initialization and retry. A persistent retry warning requires checking the assigned data and initialization; it is not proof that a hit has finished rendering.
LOD support
| Visual backend | LOD support |
|---|---|
| GPU renderer | Uses the LODs available in the prepared DataAsset, independently of RealtimeMesh |
| CPU with RealtimeMesh | Supports prepared LODs |
| CPU with ProceduralMesh | LOD0 only |
Selecting GPU mode does not generate missing LOD data. Prepare the required LODs in the tool, and inspect damage at different distances as well as close up.
The GPU renderer does not deform Nanite clusters directly. It also does not introduce direct skeletal skinning/deformation: the source skeletal component can still provide vehicle animation, physics and collision while prepared mesh data supplies the deformable visual.
Materials and attachments
Geometric dents do not require a custom deformation material. Scratch masks still need a material that reads vertex damage colors.
Apply runtime material changes to Get Mesh Component, not only to the source mesh. GPU material/proxy updates preserve existing damage.
Attachments remain separate deformable components. A forwarded hit uses the receiving part's transform and Damage Zones.
Initialization and performance settings
GPU initialization work is staggered during spawn bursts, so the visual mesh may become ready after the actor spawns. Show Parent While Initializing can show the source visual during preparation; use it only when that source looks appropriate.
The legacy CPU LOD background/foreground controls do not configure GPU processing. Initialization and data-preparation profiling can still describe CPU work needed before GPU rendering. See Performance Profiling.
Performance depends on mesh size, LODs, hit workload, materials, the selected RHI and hardware. Choosing GPU is not a fixed-FPS guarantee.