We have a number of users experiencing driver errors with our GPU path tracer on RTX 3060. The path tracer builds quite a big command list (we use DirectX 12) and sometimes after executing it we get the “device removed” error. The reason of the removal is that the device hangs. The Windows events log gets errors like “Display driver nvlddmkm stopped responding and has successfully recovered.”.
We handle device lost errors according to MSDN recommendations and it sometimes works, but if the error happens a few times during the rendering (usually after 3-4 device re-creations) we cannot create that device anymore because it disappears from the devices list returned by DX.
This happens with the current drivers (536.23) and with the older ones (something about 531 or so), but didn’t happen about a year ago when we implemented this feature. I can’t really tell what was the driver version back then, though.
Unfortunately, there is no way I can provide a simple project that reproduces the problem, but if you guys could download and install the software, I would provide a sample scene and the instructions on how to crash the driver. This includes a software license, so please contact me directly using this profile email.
If there is a way to enable extra logging and get more information or you need some sort of logs or dumps - I am more than ready to help.
Thank you :)