...wait, all the way to ASIC and only an order of magnitude improvement? (Or maybe they're being coy with the numbers?)
The deep learning hardware accelerator I worked on at Google is finally public!https://cloudplatform.googleblog.com/2016/05/Google-supercharges-machine-learning-tasks-with-custom-chip.html …
-
-
-
Yeah, they don't show raw latency/throughput numbers, which is unfortunate but not surprising if you look at Google infra papers.
End of conversation
New conversation -
-
-
aw, congrats. it's a super weird feeling to watch things go public (finally) from when you worked on them years ago!
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
very neat. Excited to finally know what you were working on :)
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Wow, you worked on it? My guess for the performance gain is simplification of FP arithmetic modules.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Out of curiosity, what were your memory tech. choices? HBM (seems increasingly popular lately) or something different?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
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.