I am working with some software that has been written assuming multiple GPUs present on the system. It is crashing for me with error message:
[09:39:34] src/storage/storage.cc:71: Check failed: e == cudaSuccess || e == cudaErrorCudartUnloading CUDA: invalid device ordinal
I have only one GPU card. Is there any way to trick CUDA so that it would only use that single card, rather than failing?
I can make changes in the source code to use the single device only, but I cannot commit such change to repository so currently need to do all testing on the branch, pulling changes periodically.
My OS is 4.18.0-18-generic #19~18.04.1-Ubuntu.