deep learning uses low precision floats sometimes as few as 8 bits are needed | Hacker News

Hacker News new | past | comments | ask | show | jobs | submit

login

perfmode on July 9, 2018 | parent | context | favorite | on: Improving the fast inverse square root (2010)

deep learning uses low precision floats

sometimes as few as 8 bits are needed

jadedhacker on July 9, 2018 [–]

I think gen 1 or gen 2 of the TPU explicitly supported short ints.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact