Google Hypercomputer TPU v5p pod

Compute

4.2K H100e

Chips

8,960

First operational

Dec 6, 2023

Google Hypercomputer TPU v5p pod is a large GPU cluster operated by Google, in the private sector. It came online around December 2023. At that moment it ranked number 18 among known clusters.

It is built on 8,960 Google TPU v5p. That amounts to about 4.2K H100e of AI compute. The cluster draws roughly 10.04 MW. Its hardware is estimated at $58M.

Epoch AI rates the certainty of this record as confirmed. The "Hypercomputer" is a cloud network, a TPU v5p pod seems to just be one component. Versions of this pod likely exist in multiple locations, possibly around the world

Full record

Owner: Google
Users: Google, Cloud
Sector: Private
Status: Existing
Primary chip: Google TPU v5p
Primary chip count: 8,960
Total AI chips: 8,960
H100 equivalents: 4.2K
8-bit throughput: 8.2×10¹⁸ OP/s
Power capacity: 10.04 MW
Hardware cost: $58M
First operational: Dec 6, 2023
Rank when it came online: 18
Certainty: Confirmed