Google TensorFlow Research Cloud

Compute

95.21 H100e

Chips

4,096

First operational

May 17, 2017

Google TensorFlow Research Cloud is a large GPU cluster operated by Google, located in the United States, in the private sector. It came online around May 2017. At that moment it ranked number 1 among known clusters.

It is built on 4,096 Google TPU v2. That amounts to about 95.21 H100e of AI compute. The cluster draws roughly 2.69 MW. Its hardware is estimated at $39M.

Epoch AI rates the certainty of this record as likely. 1 cloud TPU = 4 TPU chips, according to the Forbes doccument and the FLOP counts we know for TPU v2

Full record

Owner: Google
Users: Google, Cloud
Sector: Private
Status: Existing
Country: United States
Primary chip: Google TPU v2
Primary chip count: 4,096
Total AI chips: 4,096
H100 equivalents: 95.21
Power capacity: 2.69 MW
Hardware cost: $39M
First operational: May 17, 2017
Rank when it came online: 1
Certainty: Likely