The thing is, parallel processing (i.e same instruction on multiple data pieces) is easy to implement for the most part in hardware. The issue is nobody wants

The thing is, parallel processing (i.e same instruction on multiple data pieces) is easy to implement for the most part in hardware. The issue is nobody wants to write low level software for a particular chip. Cuda is why NVIDIA is ahead despite AMD having comparable hardware. If you look at what its like to develop for MacOS or iPhone, its pretty clear that apple doesnt do this well. Even for their neural engine, I think you have to do some sort of entitlement thing to get code to run.