CUDA is not merely software control. Let's face it: it's tied to their architecture and evolves with their architecture. Ignore whatever patent and software cop

CUDA is not merely software control. Let's face it: it's tied to their architecture and evolves with their architecture. Ignore whatever patent and software copyright you are talking about, they are simply the best at implementing this programming model. If you port CUDA over and want high performance, you must build very similar GPUs. And can you beat NVIDIA on building their own GPU architecture without much space for innovation? And yes, this does mean that NVIDIA themselves is also facing increasingly absurd constraints and after a few generations CUDA as a programming model may not be sustainable any more.