Installing on CUDA will definitely come with some difficulties. We should provide good instructions to minimize pain here.
It would e.g. be great if we could provide instructions on enabling installation via conda / from source / docker (we already do).