I really get the feeling with these models that what we need is a very memory-first hardware architecture that is not necessarily the fastest at crunching....

I really get the feeling with these models that what we need is a very memory-first hardware architecture that is not necessarily the fastest at crunching.... that seems like it shouldn’t necessarily be a terrifically expensive product