Driver 365.19 on Windows 10 x64 with OptiX 3.9, CUDA 7.5 on GTX 750 crash!

The driver 365.19 crashes with CUDA memory errors. The driver 365.10 works great!

Is this a known issue in the 365.19 driver?

PS: I can’t test OptiX 4.0 Beta because there’s still no DirectX 11 interop.

Could you please add some more details about ways to reproduce this and the exact error message you’re seeing?
For example, is this happening with any of the OptiX 3.9.0 SDK pre-built sample executables?

Thanks for your reply.

It doesn’t happen with the pre-built samples, it happens with my code when I drop I new model on the window.

Initial it shows a ground plane and that works fine. When I drop a model (a .obj, .stl, whatever) this happens:

create a Geometry
create a GeometryInstance
set Geometry to the GeometryInstance
add the GeometryInstance to a GeometryGroup
add Material to the GeometryInstance

GeometryGroup and the Acceleration (“Trbvh”, “Bvh”) are already created.

The errors I get are:

Exception thrown at 0x00007FFC0E8A1F28 in ConsoleApplication3.exe: Microsoft C++ exception: optix::shared::CudaError at memory location 0x00000064778FE8B0.
Exception thrown at 0x00007FFC0E8A1F28 in ConsoleApplication3.exe: Microsoft C++ exception: optix::Exception at memory location 0x00000064778FF7F0.
Unknown error (Details: Function “_rtContextLaunch2D” caught exception: Encountered a CUDA error: result returned (714): Device stack error (stack error or exceeded stack size limit), [6619204])

I have the stack size normally set on 5000, I’ve tried 10000 but it still crash.

With previous drivers there was no issue. I’m in the process of simplifying my code and trying to find where things goes wrong.

It seems something goes wrong with the buffer created with m_context->createBufferFromD3D11Resource(RT_BUFFER_OUTPUT, pOptiXBuffer);

When the program closes it has problems destroying that:

Exception thrown at 0x00007FFC0E8A1F28 in ConsoleApplication3.exe: Microsoft C++ exception: optix::shared::CudaError at memory location 0x00000064778FF710.
Exception thrown at 0x00007FFC0E8A1F28 in ConsoleApplication3.exe: Microsoft C++ exception: optix::shared::CudaError at memory location 0x00000064778FF660.
Exception thrown at 0x00007FFC0E8A1F28 in ConsoleApplication3.exe: Microsoft C++ exception: optix::Exception at memory location 0x00000064778FFA10.
Unhandled exception at 0x00007FFC0E8A1F28 in ConsoleApplication3.exe: Microsoft C++ exception: optix::Exception at memory location 0x00000064778FFA10.

Just brainstorming, perhaps something about DirectX changed in the new driver?

I wish I could test DirectX interop with OptiX 4.0.

Thanks for the information.

Are you just shutting down the OptiX context or are you manually destroying all OptiX resources?
Did you unregister the buffer before destroying it?

If you can provide a minimal reproducer to reproduce this in-house that would help to investigate it.
We normally need an OptiX API Capture trace in failing state from the application for analysis. The shorter, the better.
Link to how to do that here https://devtalk.nvidia.com/default/topic/803116/?comment=4436953

I’m not sure if that alone would capture the DirectX 11 app behavior as well. If this cannot be reproduced with an OAC trace alone, we would need a minimal application reproducing the issue.

If everything else is correct and this is a CUDA or DirectX driver regression there is nothing you can do to solve this on your side. The pragmatic approach would be to stay on the driver which worked while the problem is fixed in a future driver based on your reproducer.