Google TPU v4 Pod

Compute

569 H100e

Chips

4,096

First operational

May 19, 2021

Google TPU v4 Pod is a large GPU cluster operated by Google, located in the United States, in the private sector. It came online around May 2021. At that moment it ranked number 13 among known clusters.

It is built on 4,096 Google TPU v4. That amounts to about 569 H100e of AI compute. The cluster draws roughly 2.99 MW. Its hardware is estimated at $38M.

Epoch AI rates the certainty of this record as confirmed. Used in July 2021 MLPerf benchmark, mentioned May 2021, likely operational beforehand

Full record

Owner: Google
Users: Google
Sector: Private
Status: Existing
Country: United States
Primary chip: Google TPU v4
Primary chip count: 4,096
Total AI chips: 4,096
H100 equivalents: 569
8-bit throughput: 1.1×10¹⁸ OP/s
Power capacity: 2.99 MW
Hardware cost: $38M
First operational: May 19, 2021
Rank when it came online: 13
Certainty: Confirmed