I'll continue using it as long as the users are still split between NVIDIA and AMD cards. If AMD GPUs die out, CUDA it is. Or maybe give up and use a wrapper

I'll continue using it as long as the users are still split between NVIDIA and AMD cards. If AMD GPUs die out, CUDA it is. Or maybe give up and use a wrapper library for whatever 10 alternatives-only-supported-by-one-marginal-vendor there are. (like Apple)