We are using the command line to run an Unreal editor instance to render specific scenes using MRQ. We’re experiencing an issue where a worker thread occasionally crashes before reaching BeginPlay, creating empty render folders. This crash happens rarely, and when it does, rerunning the exact same command line resolves it.
Crash Detail
All of our crashes show the same error in the log:
LogLevel: ActivateLevel /Game/(ProjectName)/Levels/Sublevels/UEDPIE_0_L_LevelLoadNotify 0 0 1
[2026.03.10-02.00.45:243][ 2]LogThreadingWindows: Error: Runnable thread Background Worker #3 crashed.
[2026.03.10-02.00.45:247][ 2]LogWindows: Error: === Critical error: ===
[2026.03.10-02.00.45:248][ 2]LogWindows: Error:
[2026.03.10-02.00.45:248][ 2]LogWindows: Error: Unhandled Exception: EXCEPTION_ACCESS_VIOLATION reading address 0x0000017249f4a310
[2026.03.10-02.00.45:248][ 2]LogWindows: Error:
[2026.03.10-02.00.45:248][ 2]LogWindows: Error: [Callstack] 0x00007ffa30525f8f UnrealEditor-D3D12RHI.dll!FD3D12RayTracingPipelineState::FD3D12RayTracingPipelineState() [D:\\build\\++UE5\\Sync\\Engine\\Source\\Runtime\\D3D12RHI\\Private\\D3D12RayTracing.cpp:2765]
We have multiple levels that can be rendered, and this specific crash happens regardless of which level we’re trying to render.
Crash Frequency
This crash happens rarely, and when it does, rerunning the exact same command resolves it. In our efforts to reproduce it, we’ve observed a probability of around 2%–3%.
Command and Project Details
The Command we use is:
<Engine Location> <Project Location> <LevelToRender> -skipcompile -Multiprocess -NoLoadingScreen -FixedSeed -log -nohmd -Unattended -MRQInstance -deterministicaudio -audiomixer
Repro
This bug happens rarely, and we haven’t found a way to reproduce it consistently. However, we’ve found that limiting the number of cores used to load the project with -corelimit increases the probability of this crash occurring.
Hypothesis
From our testing and attempts to reproduce the bug, the problem appears related to level loading. Specifically, the raytracing shader worker thread is accessing resources that aren’t ready yet. Our experiments with limiting core counts increase the crash probability, which suggests a race condition between the main loading thread and the graphics worker thread.
[Attachment Removed]