cudafe++ died with status 0xc0000409 when switching to c++20 for nvcc

When I switch nvcc from using c++17 to c++20 through std flag, I get this error:

cudafe++ died with status 0xc0000409

All compiles and works as expected with 17, but when I switch to c++20, my compilation fails with the cryptic error.

I am running:

• CUDA Toolkit v12.5
• MVSC 17.10.1

Compiling host code with c++20 and passing c++17 flag to nvcc also works, and all my unit tests pass, but somehow, when I switch the nvcc flag, the compilation crashes.

What causes this behavior?

P.S.:

• no heavy use of templates, basically just to switch loops on and off
• no fancy options passed to nvcc also disabled them to see if they are causing issues
• I do a clean build in between the attempts and delete any cached files

CUDA 12.5 Update 1 resolves the issue.

• release notes
• NVIDIA Forum
• Download