Chip advances require fresh ideas, speakers say

Re:Chip advances require fresh ideas, speakers say by jacksmith21006 on Tuesday February 13, 2018 @06:21PM (#16371855)

Can you provide a source on the memory starvation for the TPUs? Plus are we talking generation 1 or 2 or both?

Did not seem to be an issue in the paper with the first generation.

https://cloud.google.com/blog/big-data/2017/05/an-in-depth-l... An in-depth look at Google's first Tensor Processing Unit (TPU ...

[ Parent | on HN ]

Re:Chip advances require fresh ideas, speakers say by pjscott on Tuesday February 13, 2018 @08:42PM (#16372720)

In the paper on the first-generation TPU, in section 7. They estimate that there would have been impressive gains in speed, both absolute and per-watt, if they'd had enough design time to give it more memory bandwidth:

https://drive.google.com/file/d/0Bx4hafXDDq2EMzRNcy1vSUxtcEk...

[ Parent | on HN ]

Re:Chip advances require fresh ideas, speakers say by baybal2 on Tuesday February 13, 2018 @07:01PM (#16372118)

This is what was intensively discussed by first trial users, and later admitted by google tpu team themselves on hot chips 29

Even with monsterous HBM2 memory, they still have it.

It is probably hard to predict what matrix set to prefetch when you deal with a neural net. So you have cache misses there too

[ Parent | on HN ]