Curious: Why are all these neural network coprocessors (TPUs, Neural Engine, etc.) separate processors instead of just GPUs with better support for 8-bit ints and floats?
-
-
My guess is that GPU manufacturers didn't want to do that, so cloud providers made their own, without access to GPU IP.
-
But that paper is about NVIDIA GPUs. They still made their own separate block, and I don’t know why. Same with Apple—Apple makes their GPU, no reason they couldn’t have just enhanced it instead of adding a separate Neural Engine, right?
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.