Hacker News Reader

moving large data around is the whole issue of deep learning.

No, it's not. You have to apply the data to the evolving network. Can you really do it at 320GB/s?

Good question. If you have enough cores - maybe? Does anyone have a precise idea?