Fatal GPU error caused by displacement shader

- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report
I encountered a fatal error when rendering a scene with displacement shaders on a GPU (Windows 10, RTX 2060, Driver 462.31). Maya 2020 + MtoA 4.2.2 (Arnold Core 6.2.1.0). The full debug log is here: arnold-gpu-crash.txt
00:00:06 4143MB ERROR | [gpu] an error happened during rendering. OptiX error is: Unknown error (Details: Function "_rtContextLaunch2D" caught exception: Encountered a CUDA error: cudaDriver().CuEventSynchronize( m_event ) returned (700): Illegal address, file: <internal>, line: 0) GPU 0 had 4161MB free before rendering started and 2550MB free when crash occurred GPU errors are sometimes due to a GPU not having enough remaining free memory. To see if this is what happened here, try simplifying your scene or running on a GPU with more free RAM to see if it solves the crash. Otherwise, upgrading to the latest nvidia gpu driver and Arnold core (available from www.arnoldrenderer.com) might fix the crash
I tried re-prepopulating the GPU cache, downgrading the GPU driver to 461.40, cleaning up the scene, disabling auto bump and ignoring displacement shaders in Render Settings. Sadly, none of them works.
Weirdly, this problem only occurs after I added the fifth displacement shader. If I delete the last material and the last displacement shader, the scene renders just fine with the previous four displacement shaders.
I have seen some previous questions mentioning the same issue, but they are all several months old. Attaching them here, just for everyone's information:
- https://answers.arnoldrenderer.com/questions/28860/displacement-not-working-gpu.html
- https://answers.arnoldrenderer.com/questions/28317/cuda-error-when-using-displacement-on-the-gpu.htm...
- This one mentions that there is a hotfix in MtoA 4.1.1.1 for a related issue core#10050. I guess I will try this version of MtoA next... (Update: surprisingly downgrading to MtoA 4.1.1.1 solved this issue. There might be a regression in later versions that exposes this issue again.)
Any help/workarounds will be hugely appreciated!