Hi - I’m developing an application that should run on both, windows and linux. My app works well on windows with the latest cuda 11.5 update 1 and the latest driver 497.09. Since I want to use WSL2 to cross-compile my app to run on linux I installed driver 510.06. Now the windows version doesn’t work anymore and I get an error “CUDA_ERROR_UNSUPPORTED_PTX_VERSION” if I try to load my JIT compile ptx onto the GPU with cuModuleLoadDataEx.
On Linux I run into problems much earlier. even the call to cuInit( 0 ) failed with “CUDA_ERROR_OUT_OF_MEMORY” using driver 510.06.
I followed the guides to install cuda and the wsl2 subsystem as close as possible, but I seems not to be able to get my app working on windows & WSL2 (linux).
I’m getting a very similar issue: my simple hello world CUDA program was compiling and successfully running on Windows while using driver 479.29. I apparently have to switch from this driver to be able to use CUDA WSL by installing driver 510.06, but then my program no longer runs successfully on neither Windows nor Linux.
Switching back to driver 479.29 solves the problem of running/developing my code but only on Windows: I am unable to make it cross-platform as I cannot easily test for compilation/running on Linux using WSL.
Please use the latest NVIDIA Windows driver 497.29 released in December from here. This is the latest official driver that supports WSL2. It’s much more recent than 510.06 and should have your issues resolved. Please let us know if otherwise.