In the README for CUDA setup, it might be helpful to add information about CUDA versioning mismatch between nvidia-smi and nvcc --version.
The CUDA version reported by nvidia-smi does not report the installed toolkit version. The version reported by nvidia-smi is the maximum CUDA version that the driver supports.
If the CUDA version reported by nvidia-smi is lower than the one reported by nvcc, there's a possibility that the driver might not fully support all the features of the latest CUDA toolkit version. This could lead to a runtime error message.