Why does Niagara need CPU Access on a static mesh?

For GPU particle simulations that read mesh data (Sample Static Mesh, mesh-driven spawning), Niagara reads vertex/index data on the CPU side to upload to the GPU. Without ‘Allow CPU Access’ in the static mesh import settings, the data is GPU-only and Niagara fails at cook.

How do I enable Allow CPU Access?

Open the static mesh asset, find LOD 0 in the Details panel, and enable ‘Allow CPU Access’. Save. Re-reference in the Niagara emitter (it may show a warning until re-saved).

Does Allow CPU Access bloat memory?

Yes — keeps the index/vertex buffer in CPU memory in addition to GPU. For typical small particle meshes, the cost is negligible. For high-poly meshes, decide whether the perf is worth it; consider a simpler proxy mesh for particles instead.

Fix: Unreal Niagara Particle Mesh Cooked Package Fail

Quick answer: Open the static mesh asset, enable LOD 0 → Allow CPU Access. Save. Re-cook. The error is GPU particles needing CPU-side mesh data unavailable by default.

A packaged build fails to cook with: “NiagaraEmitter X: Particle mesh requires CPU access but is missing in cooked build.” The mesh works fine in PIE. Shipping cook trips on this and aborts.

CPU Access for GPU Particles

Niagara emitters that use mesh data on the GPU (Sample Static Mesh data interface, mesh-driven spawn, etc.) need access to the mesh’s vertex and index arrays from the CPU side to push to the GPU buffer. In editor builds, this data is always available. In cooked builds, it’s stripped unless “Allow CPU Access” is set on the mesh.

The Fix

Open the offending static mesh (the one your Niagara emitter samples).
In the right panel, find LOD 0.
Check Allow CPU Access.
Save the asset.
Re-run the cook. The error should clear.

Multiple LODs

If Niagara samples a specific LOD index, enable CPU Access on that LOD. Typically LOD 0 is enough; bake-down workflows may use higher LODs for cheaper sampling.

Memory Trade-Off

Allow CPU Access keeps the index/vertex buffer in main memory in addition to GPU memory. For a 5K-tri particle mesh, an extra ~150 KB — trivial. For a 500K-tri sculpt, ~15 MB — meaningful.

If memory matters, use a low-poly proxy specifically for particles: a 200-tri version that’s indistinguishable at particle size. Mark only the proxy with CPU Access.

Other Causes of the Cook Error

Mesh not referenced: the cooker only cooks meshes reachable from cooked-content roots. If the Niagara system isn’t reachable, the mesh might not cook even with CPU Access set. Confirm the Niagara system is in a shipping content path.
Strip in cooker filter: a custom IniSection or DirectoriesToNeverCook entry might exclude the mesh. Inspect Cook log for “skipping” messages.

Verifying

Re-cook. The error should disappear. Test the cooked build on target — particles should render correctly with the mesh emitter.

“GPU particles need CPU-accessible meshes. Toggle the flag, save, re-cook. Two minutes to fix.”

Audit all meshes used by Niagara systems once at project setup — saves catching this at the cook step every release.