# \[5.8.4\] GPU hang (DXGI\_ERROR\_DEVICE\_HUNG) in Nanite MicropolyRasterize with Lumen on async compute

**URL:** <https://forums.unrealengine.com/t/5-8-4-gpu-hang-dxgi-error-device-hung-in-nanite-micropolyrasterize-with-lumen-on-async-compute/2857439>\
**Category:** Issues and Bug Reporting (Unreal Engine)\
**Tags:** unreal-engine, bug-report\
**Created:** [November 18, 2026, 6:21am UTC](https://forums.unrealengine.com/t/5-8-4-gpu-hang-dxgi-error-device-hung-in-nanite-micropolyrasterize-with-lumen-on-async-compute/2857439 "2026-11-18T06:21:45Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![sith\_au](https://d3kjluh73b9h9o.cloudfront.net/original/4X/2/8/5/28502927f1d97ba1051b3b47ccfe17233bf41ef0.png) [@sith\_au](https://forums.unrealengine.com/u/sith_au)\
**Post date:** [November 18, 2026, 6:21am UTC](https://forums.unrealengine.com/t/5-8-4-gpu-hang-dxgi-error-device-hung-in-nanite-micropolyrasterize-with-lumen-on-async-compute/2857439/1 "2026-11-18T06:21:45Z")

</div>

### Summary

GPU hang (DXGI\_ERROR\_DEVICE\_HUNG) in Nanite MicropolyRasterize during the virtual shadow map pass, only while Lumen diffuse indirect runs on async compute. Aftermath: Misaligned Address Error on a ClusterPageData Load4 in BitStreamReader\_Read\_RO. Happens at random in packaged builds and on editor startup. D3D12, RTX 3090, 5.8.4.

### What Type of Bug are you experiencing?

Rendering (Graphics / Niagara)

### Steps to Reproduce

Steps to Reproduce

1. Project with Nanite, virtual shadow maps, and Lumen using software tracing (r.RayTracing=0). Lumen diffuse indirect on async compute (the default, r.Lumen.DiffuseIndirect.AsyncCompute=1).
2. D3D12, NVIDIA RTX 3090 (seen on drivers 621.24 and 617.14).
3. Start a packaged Development build cold. The main menu map renders after startup movies (Electra) play.
4. Repeat. About 1 in 3 launches hangs within ~30 s. Opening the editor also hangs sometimes, almost always on the first viewport frame (frame 2). Not yet tried in a blank project.

### Expected Result

The map renders with no GPU hang.

### Observed Result

DXGI\_ERROR\_DEVICE\_HUNG, then the app exits. Aftermath dump:

Only active shader: MicropolyRasterize (NaniteRasterizer.usf). Graphics markers: Nanite::DrawGeometry \> PostPass in the VSM pass. Async compute markers: LumenScreenProbeGather \> UpdateWorldRadianceCaches.

One warp: Misaligned Address Error. It maps (with r.Shaders.Symbols=1) to uint4 Data = InputBuffer.Load4(Address); in BitStreamReader\_Read\_RO (BitPacking.ush). The other warps are stuck in BitAlignU32 in the same reader. Likely caller: DecodeTriangleIndices in ClusterRasterize (ClusterPageData at Cluster.PageBaseAddress + Cluster.IndexOffset), so the cluster data it read was bad.

No page fault.

* * *

Hang counts, cold launches:

- Defaults, driver 621.24: 13 of 44
- Defaults, driver 617.14: 2 of 10
- r.Lumen.DiffuseIndirect.AsyncCompute=0 or r.RDG.AsyncCompute=0: 0 of 13
- Startup movies skipped: 0 of 7
- r.RDG.AsyncComputeTransientAliasing=0: 1 of 8
- r.Nanite.Streaming.ReservedResources=0: 2 of 6

### Affects Versions

5.8

### Platform(s)

Windows

### For crash reports, include your callstack

No CPU callstack: it’s a GPU hang. Active GPU breadcrumbs at the hang: Graphics: ShadowDepths \> RenderVirtualShadowMaps(Nanite) \> Nanite::DrawGeometry \> NodeCull / ClusterCull (varies) AsyncCompute: DiffuseIndirectAndAO \> LumenScreenProbeGather \> UpdateWorldRadianceCaches \> TraceFromProbes (also CompositeTraces, DistanceFieldTracing or TranslucencyVolumeLighting in other hangs)

### Additional Notes

Aftermath dump, .nvdbg, the shader’s .dxil and .pdb, and a log -[https://drive.google.com/file/d/1jsRypHGp87Ql3k\_ez6gp6LwzCbJzgPUN/view?usp=sharing](https://drive.google.com/file/d/1jsRypHGp87Ql3k_ez6gp6LwzCbJzgPUN/view?usp=sharing)
