Post #2890670

@dschaub@mstdn.social @atpfm@mastodon.social @siracusa@mastodon.social it is the design of the thing called "Fabric", which shares memory between the CPU and GPU, and also the "Neural Engine" matrix math/ML chips. It's fresh in my mind because I just wrote a bunch of code that uses Metal or Core ML where appropriate. This is much different from writing CUDA, where you have to worry about whether it's worth sending things to the GPU.