News for hackers. Stuff that matters.

A comment on Benchmarking Google’s new TPUv2.

Re:Benchmarking Google’s new TPUv2 by PaulHoule on Friday February 23, 2018 @01:44PM (#16448333)
Does this take into account the fact that you might need fewer epochs if you reduce the batch size? (as is done for the CPU?)
[ Parent | on HN ]
Re:Benchmarking Google’s new TPUv2 by elmarhaussmann on Friday February 23, 2018 @02:35PM (#16448715)
Author here.

No, this is really only comparing the throughput on the devices. A thorough comparison should really focus on time to reach a certain quality - including all of the tricks available for a certain architecture.

[ Parent | on HN ]