# Reproducible GPU crash in the parallel depth prepass after ~2 minutes 

**URL:** <https://forums.unrealengine.com/t/reproducible-gpu-crash-in-the-parallel-depth-prepass-after-2-minutes/2746440>\
**Category:** Issues and Bug Reporting (Unreal Engine)\
**Tags:** unreal-engine\
**Created:** [August 25, 2026, 5:35pm UTC](https://forums.unrealengine.com/t/reproducible-gpu-crash-in-the-parallel-depth-prepass-after-2-minutes/2746440 "2026-08-25T17:35:05Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![scarmahdi](https://d3kjluh73b9h9o.cloudfront.net/original/4X/2/8/5/28502927f1d97ba1051b3b47ccfe17233bf41ef0.png) [@scarmahdi](https://forums.unrealengine.com/u/scarmahdi)\
**Post date:** [August 25, 2026, 5:35pm UTC](https://forums.unrealengine.com/t/reproducible-gpu-crash-in-the-parallel-depth-prepass-after-2-minutes/2746440/1 "2026-08-25T17:35:06Z")

</div>

### Summary

Reproducible GPU crash in the parallel depth prepass after ~2 minutes: A packaged Development client faults on a render worker at a near-constant wall-clock time on two different machines, two GPUs, two driver versions, two RHIs and two engine versions. Thirteen candidate causes were eliminated with each console variable echoed back to confirm it applied.

### What Type of Bug are you experiencing?

Rendering (Graphics / Niagara)

### Steps to Reproduce

Reproduction  
A networked client rendering ~200 server-driven skeletal actors, 423 static props and 13 NPCs in a purchased marketplace level. Nine of nine attempts reproduced.

Launch a packaged Development build against a level with several hundred replicated actors receiving transform updates at ~9 Hz.  
Let it run. No player input is required — an idle session crashes identically to active play.  
Fault occurs at ~2:00, on a background worker, inside the RHI.  
Observed survival, one machine, consecutive runs  
1:56 1:59 2:00 2:00 2:01 2:02 2:04 2:05 2:05

### Expected Result

The client renders continuously without faulting. Parallel recording of the  
depth prepass should not dereference invalid memory.

### Observed Result

EXCEPTION\_ACCESS\_VIOLATION on a task-graph background worker inside the RHI,  
during parallel recording of the depth prepass, at a near-constant ~2:00 of  
wall-clock time.

Nine consecutive runs on one machine: 1:56, 1:59, 2:00, 2:00, 2:01, 2:02,  
2:04, 2:05, 2:05.

The trigger scales with TIME, not with frames or rendering load. Reducing the  
drawn skeletal-mesh count 6x RAISED the framerate, so the frame counter at the  
fault rose from 6,571 to 8,059 — while the wall-clock time stayed at 2:02.

It also crashes while the player is completely idle, identically to active  
play, so it is not driven by player input or camera movement.

Fault address is 0xffffffffffffffff on the desktop machine; 0x48 and 0x90 on  
the laptop.

-d3ddebug and -d3d12gpuvalidation both report ZERO errors and ZERO warnings.

### Affects Versions

5.8

### Platform(s)

Windows

### Upload an image

[UE58-crash-logs.zip](https://forums.unrealengine.com/uploads/short-url/mGJ00ed6k8NqnfphsSPQr2jZJ8p.zip) (7.23 MB)

### For crash reports, include your callstack

EXCEPTION\_ACCESS\_VIOLATION reading address 0xffffffffffffffff

BharataClient.exe!FD3D12ContextCommon::BindDiagnosticBuffer()  
BharataClient.exe!FD3D12DescriptorCache::SetSamplerTable()  
BharataClient.exe!FD3D12CommandContext::RHISetShaderParameters()  
BharataClient.exe!FRHICommandSetShaderParameters::Execute()  
BharataClient.exe!FRHICommand\<FRHICommandSetShaderParameters,  
FRHICommandSetShaderParametersString1589\>::ExecuteAndDestruct()  
BharataClient.exe!FRHICommandListBase::Execute()  
BharataClient.exe!FRHICommandListExecutor::FTranslateState::Translate\_ExecuteCommandList()  
BharataClient.exe!FRHICommandListExecutor::FTaskPipe::Execute()  
BharataClient.exe!TGraphTask\<TFunctionGraphTaskImpl\<void \_\_cdecl(enum ENamedThreads::Type,  
TRefCountPtr const &),0\> \>::ExecuteTask()  
BharataClient.exe!UE::Tasks::Private::FTaskBase::TryExecuteTask()  
BharataClient.exe!LowLevelTasks::FTask::ExecuteTask()  
BharataClient.exe!LowLevelTasks::FScheduler::ExecuteTask()  
BharataClient.exe!LowLevelTasks::FScheduler::WorkerLoop()  
KERNEL32.DLL!UnknownFunction

Crash in runnable thread Background Worker #3

GPU breadcrumbs at the fault:  
Breadcrumbs ‘Parallel’

- ParallelDraw (Index: 2, Num: 3)
- PrePass DDM\_AllOpaque (Forced by DBuffer)
- Scene
- RenderGraphExecute - /ViewFamilies
- SceneRender - ViewFamilies
- Frame 6571

### Additional Notes

ENVIRONMENT — reproduces on two independent machines:

Machine A: RTX 3070 Ti Laptop (8 GB), driver 610.88 (32.0.16.1088)  
UE 5.8.0 and 5.8.1 — survival 2:02 to 4:05  
Machine B: RTX 3080 desktop, driver 610.74 (32.0.16.1074)  
UE 5.8.1 — survival 1:56 to 2:05 across 9 runs

Both Game Ready and Studio driver branches tested on Machine A (NVIDIA ships  
both from the same 610.88 build; engine reported an identical version string).

THE BREADCRUMB FOLLOWS WHATEVER FORCES THE DEPTH PREPASS — this misleads:

stock → PrePass DDM\_AllOpaque (Forced by DBuffer)  
r.Nanite.ProjectEnabled=0 → PrePass DDM\_AllOpaque (Forced by DBuffer)  
r.DBuffer=0 → PrePass DDM\_AllOpaque (Forced by VirtualTexture)  
rhi.UseSubmissionThread=0 → PrePass DDM\_AllOpaque (Forced by DBuffer)

Disabling whichever subsystem forces the prepass simply hands that role to  
another, and the breadcrumb changes accordingly. Reading that as “the crash  
moved when I disabled X, therefore X is innocent” falsely exonerates every  
feature in turn. The fault follows the prepass itself, not its forcer.

ELIMINATED — every cvar echoed back via -ExecCmds to confirm it applied  
before the run was counted:

Nanite r.Nanite.ProjectEnabled=0 still crashes  
Deferred decals r.DBuffer=0 still crashes  
Submission thread rhi.UseSubmissionThread=0 still crashes  
RHI D3D12 and Vulkan (SM5, SM6) both crash  
Engine version 5.8.0 → 5.8.1 still crashes  
GPU / driver 2 GPUs, 2 drivers, GR + Studio still crashes  
Editor vs packaged both both crash  
Rendering workload 6x fewer meshes; foliage off still crashes  
Player activity idle vs continuous combat identical  
API misuse -d3ddebug, -d3d12gpuvalidation 0 errors

VULKAN — same project, same repro, ~4x longer survival (8:17 vs ~2:02),  
fault inside the driver:

nvoglv64.dll!UnknownFunction  
nvoglv64.dll!UnknownFunction  
UnrealEditor-VulkanRHI.dll!UnknownFunction

PROJECT SHAPE: a networked client rendering ~200 server-driven skeletal  
actors, 423 static props and 13 NPCs, receiving transform updates at ~9 Hz.  
~12 GB cooked content with Nanite, Lumen and virtual textures.

WHAT WE CANNOT RULE OUT: our own code. We eliminated object lifetime (all  
actor references are TObjectPtr or weak, verified with runtime counters) and  
rendering workload (above), and validation layers are clean — but this is a  
large project and we are not claiming certainty.

ATTACHED: 7 full logs, one per eliminated variable. Each contains engine init,  
GPU/driver banner, per-10s application telemetry, breadcrumbs and callstack.

Reproduced on both 5.8.0 CL 55116800 and 5.8.1 CL 56057345

# ======================================================================== RAW LOG EVIDENCE — 7 runs, one per eliminated variable

# UE 5.8 — GPU CRASH IN PARALLEL DEPTH PREPASS — CONSOLIDATED LOG EVIDENCE Seven runs, one per eliminated variable. Full logs available on request.

— 01 STOCK ---------------------------------------------------------  
Engine Version: 5.8.0-55116800+++UE5+Release-5.8  
Found D3D12 adapter 0: NVIDIA GeForce RTX 3080  
Driver Version: 610.74 (internal:32.0.16.1074, unified:610.74)  
CVAR UNDER TEST (echoed back to prove it applied):  
r.Nanite.ProjectEnabled:1  
visible meshes: mobs=196(60 drawn)  
TIMELINE:  
joined [2026.08.22-11.54.31:919]  
crashed [2026.08.22-11.56.33:195]  
reading address 0xffffffffffffffff  
GPU BREADCRUMB:  
[2026.08.22-11.56.33:193][819]LogRHI: Error: Breadcrumbs ‘Parallel’

- ParallelDraw (Index: 0, Num: 1)
- PrePass DDM\_AllOpaque (Forced by DBuffer)
- Scene
- RenderGraphExecute - /ViewFamilies
- SceneRender - ViewFamilies
- Frame 6817

— 02 NANITE DISABLED ---------------------------------------------------------  
Engine Version: 5.8.0-55116800+++UE5+Release-5.8  
Found D3D12 adapter 0: NVIDIA GeForce RTX 3080  
Driver Version: 610.74 (internal:32.0.16.1074, unified:610.74)  
CVAR UNDER TEST (echoed back to prove it applied):  
r.Nanite.ProjectEnabled:0  
visible meshes: mobs=196(60 drawn)  
TIMELINE:  
joined [2026.08.22-12.19.52:635]  
crashed [2026.08.22-12.21.54:370]  
reading address 0xffffffffffffffff  
GPU BREADCRUMB:  
[2026.08.22-12.21.54:370][573]LogRHI: Error: Breadcrumbs ‘Parallel’

- ParallelDraw (Index: 2, Num: 3)
- PrePass DDM\_AllOpaque (Forced by DBuffer)
- Scene
- RenderGraphExecute - /ViewFamilies
- SceneRender - ViewFamilies
- Frame 6571

— 03 DBUFFER DISABLED ---------------------------------------------------------  
Engine Version: 5.8.0-55116800+++UE5+Release-5.8  
Found D3D12 adapter 0: NVIDIA GeForce RTX 3080  
Driver Version: 610.74 (internal:32.0.16.1074, unified:610.74)  
CVAR UNDER TEST (echoed back to prove it applied):  
r.DBuffer:0  
visible meshes: mobs=195(60 drawn)  
TIMELINE:  
joined [2026.08.22-12.24.23:565]  
crashed [2026.08.22-12.26.25:330]  
reading address 0xffffffffffffffff  
GPU BREADCRUMB:  
[2026.08.22-12.26.25:330][867]LogRHI: Error: Breadcrumbs ‘Parallel’

- ParallelDraw (Index: 2, Num: 3)
- PrePass DDM\_AllOpaque (Forced by VirtualTexture)
- Scene
- RenderGraphExecute - /ViewFamilies
- SceneRender - ViewFamilies
- Frame 6865

— 04 SUBMISSION THREAD OFF ---------------------------------------------------------  
Engine Version: 5.8.0-55116800+++UE5+Release-5.8  
Found D3D12 adapter 0: NVIDIA GeForce RTX 3080  
Driver Version: 610.74 (internal:32.0.16.1074, unified:610.74)  
CVAR UNDER TEST (echoed back to prove it applied):  
rhi.UseSubmissionThread = “0”  
visible meshes: mobs=194(60 drawn)  
TIMELINE:  
joined [2026.08.22-12.28.56:773]  
crashed [2026.08.22-12.30.58:523]  
reading address 0xffffffffffffffff  
GPU BREADCRUMB:  
[2026.08.22-12.30.58:523][985]LogRHI: Error: Breadcrumbs ‘Parallel’

- ParallelDraw (Index: 6, Num: 8)
- PrePass DDM\_AllOpaque (Forced by DBuffer)
- Scene
- RenderGraphExecute - /ViewFamilies
- SceneRender - ViewFamilies
- Frame 6983

— 05 6x FEWER MESHES ---------------------------------------------------------  
Engine Version: 5.8.0-55116800+++UE5+Release-5.8  
Found D3D12 adapter 0: NVIDIA GeForce RTX 3080  
Driver Version: 610.74 (internal:32.0.16.1074, unified:610.74)  
CVAR UNDER TEST (echoed back to prove it applied):  
r.Nanite.ProjectEnabled:1  
visible meshes: mobs=195(10 drawn)  
TIMELINE:  
joined [2026.08.22-12.34.40:599]  
crashed [2026.08.22-12.36.42:413]  
reading address 0xffffffffffffffff  
GPU BREADCRUMB:  
[2026.08.22-12.36.42:413][61]LogRHI: Error: Breadcrumbs ‘Parallel’

- ParallelDraw (Index: 4, Num: 5)
- PrePass DDM\_AllOpaque (Forced by DBuffer)
- Scene
- RenderGraphExecute - /ViewFamilies
- SceneRender - ViewFamilies
- Frame 8059

— 06 FOLIAGE OFF ---------------------------------------------------------  
Engine Version: 5.8.0-55116800+++UE5+Release-5.8  
Found D3D12 adapter 0: NVIDIA GeForce RTX 3080  
Driver Version: 610.74 (internal:32.0.16.1074, unified:610.74)  
CVAR UNDER TEST (echoed back to prove it applied):  
r.Nanite.ProjectEnabled:1  
visible meshes: mobs=194(60 drawn)  
TIMELINE:  
joined [2026.08.22-12.37.35:355]  
crashed [2026.08.22-12.39.37:131]  
reading address 0xffffffffffffffff  
GPU BREADCRUMB:  
[2026.08.22-12.39.37:131][839]LogRHI: Error: Breadcrumbs ‘Parallel’

- ParallelDraw (Index: 1, Num: 2)
- PrePass DDM\_AllOpaque (Forced by DBuffer)
- Scene
- RenderGraphExecute - /ViewFamilies
- SceneRender - ViewFamilies
- Frame 6837

— 07 LATEST BUILD ---------------------------------------------------------  
Engine Version: 5.8.1-56057345+++UE5+Release-5.8  
Found D3D12 adapter 0: NVIDIA GeForce RTX 3080  
Driver Version: 610.74 (internal:32.0.16.1074, unified:610.74)  
CVAR UNDER TEST (echoed back to prove it applied):  
r.Nanite.ProjectEnabled:1  
visible meshes: mobs=193(60 drawn)  
TIMELINE:  
joined [2026.08.22-14.13.39:927]  
crashed [2026.08.22-14.15.35:508]  
reading address 0xffffffffffffffff  
GPU BREADCRUMB:  
[2026.08.22-14.15.35:508][880]LogRHI: Error: Breadcrumbs ‘Parallel’

- Nanite::BasePass
- NaniteBasePass
- BasePass
- Scene
- RenderGraphExecute - /ViewFamilies
- SceneRender - ViewFamilies

========================================================================  
SYMBOLICATED CALLSTACK (packaged Development build):  
BharataClient.exe!FD3D12ContextCommon::BindDiagnosticBuffer()   
BharataClient.exe!FD3D12DescriptorCache::SetSamplerTable()   
BharataClient.exe!FD3D12CommandContext::RHISetShaderParameters()   
BharataClient.exe!FRHICommandSetShaderParameters::Execute()   
BharataClient.exe!FRHICommand\<FRHICommandSetShaderParameters,FRHICommandSetShaderParametersString1589\>::ExecuteAndDestruct()   
BharataClient.exe!FRHICommandListBase::Execute()   
BharataClient.exe!FRHICommandListExecutor::FTranslateState::Translate\_ExecuteCommandList()   
BharataClient.exe!FRHICommandListExecutor::FTaskPipe::Execute()   
BharataClient.exe!TGraphTask\<TFunctionGraphTaskImpl\<void \_\_cdecl(enum ENamedThreads::Type,TRefCountPtr const & \_\_ptr64),0\> \>::ExecuteTask()   
BharataClient.exe!UE::Tasks::Private::FTaskBase::TryExecuteTask()   
BharataClient.exe!LowLevelTasks::TTaskDelegate\<LowLevelTasks::FTask \* \_\_ptr64 \_\_cdecl(bool),48\>::TTaskDelegateImpl\<`LowLevelTasks::FTask::Init<`UE::Tasks::Private::FTaskBase::Init’::`2'::<lambda_1> >'::`15’::\<lambda\_1\>,0\>::CallAndMove()   
BharataClient.exe!LowLevelTasks::FTask::ExecuteTask()

---

<div class="post-metadata">

**Author:** ![LucasRz](https://d3kjluh73b9h9o.cloudfront.net/original/4X/2/8/5/28502927f1d97ba1051b3b47ccfe17233bf41ef0.png) [@LucasRz](https://forums.unrealengine.com/u/LucasRz)\
**Post date:** [August 31, 2026, 12:36pm UTC](https://forums.unrealengine.com/t/reproducible-gpu-crash-in-the-parallel-depth-prepass-after-2-minutes/2746440/2 "2026-08-31T12:36:20Z")

</div>

Hi Scarmahdi, I’ve been having the same problem for the past four days. If you find any solution, please share it. I’ll do the same if I find one. Thanks!
