As GPUs become more general purpose, do you expect there to be a split between hardware specialized for graphics and GPUs used for other applications like ML, etc.?
anag
How exactly do we design hardware for these special computations? Like, are these chips only processing the inputs in one way and doing so very quickly, then passing the vectors through?
As GPUs become more general purpose, do you expect there to be a split between hardware specialized for graphics and GPUs used for other applications like ML, etc.?
How exactly do we design hardware for these special computations? Like, are these chips only processing the inputs in one way and doing so very quickly, then passing the vectors through?