Well, one problem is that you're reading an NVIDIA marketing post on an NVIDIA blog talking in particular about the lowest levels of the stack targeting NVIDIA

Well, one problem is that you're reading an NVIDIA marketing post on an NVIDIA blog talking in particular about the lowest levels of the stack targeting NVIDIA hardware. Higher level abstractions can and do just work across different hardware backends (not as well as we'd like, but we have some thoughts on how to improve that).