When somebody pays somebody to write those frameworks. When nvidia sees a need, they can change CUDA over night to address it, and they pay people to do that.

When somebody pays somebody to write those frameworks. When nvidia sees a need, they can change CUDA over night to address it, and they pay people to do that. When you need to do the same in Vulkan, that’s a multi year process till your extension is “open”. A teaser here whose job is to get something done with ML has better things to do than going through that.