Skip to content
Arne Noori
Aboutarne.ai
← All artifacts
Google Tensor Processing Unit

Google Tensor Processing Unit

Image creditNorman P. Jouppi, George Kurian, Sheng Li, Peter Ma, Rahul Nagarajan, Lifeng Nai, Nishant Patil, Suvinay Subramanian, Andy Swing, Brian Towles, Cliff Young, Xiang Zhou, Zongwei Zhou, and David Patterson, via Wikimedia Commons (a later TPU v4 board)
CC BY 4.0 ↗

Google Tensor Processing Unit

Year
2016
Creator
Norman P. Jouppi and Google's TPU team
Company
Google
Location
Mountain View, California

Description

The Tensor Processing Unit, or TPU, is a chip that Google designed only for the math of neural networks. Google disclosed the TPU in May 2016, when the chips had already operated in its data centers for more than one year.

Importance

The TPU is a special chip for one type of work, and not a general processor such as a CPU or a GPU. In a 2017 paper, Google said that the TPU was on average 15 to 30 times faster than a CPU or GPU of that time. Apple also designed its own chip for the Mac, the M1, which is in this room.

Interesting fact

Google said that its engineers had the first TPUs in operation in its data centers only 22 days after they tested the first chips.

Sources

  1. Google supercharges machine learning tasks with TPU custom chip ↗Google Cloud Blog · 2016-05-18
  2. In-Datacenter Performance Analysis of a Tensor Processing Unit (Jouppi et al.) ↗arXiv · 2017-04-16
  3. Apple unleashes M1 ↗Apple Newsroom · 2020-11-10