My sequential processing would be 100,000,000x faster but my memory bandwidth would be 10x less and overall processing 100x less. Sounds unpleasant, honestly https://twitter.com/intel/status/1106697022864711681 …
-
-
The back-of-the-envelope model is that each of 10^11 neurons maintains on average 10 bits of working state. So 1Tbps bandwidth.
-
I'll be the first to admit that this estimate could be orders of magnitude off in either direction, and also that "memory bandwidth" isn't super well-defined, but I still suspect the von Neumann bottleneck would be frustrating for a single-node Intel-based upload.
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.