Graphics are a notorious case of CPU-bound applications. Adding IO boundedness to the equation seems like a bad idea.

I’m not saying it can’t be done, or that it can’t be fast. But it’s very much against the grain and needs more justification Than “NVMe is fast”