Tesla Training Cluster

Compute

1.3K H100e

Chips

4,032

First operational

Aug 20, 2021

Tesla Training Cluster is a large GPU cluster operated by Tesla, located in the United States, in the private sector. It came online around August 2021. At that moment it ranked number 9 among known clusters.

It is built on 4,032 NVIDIA A100. That amounts to about 1.3K H100e of AI compute. The cluster draws roughly 3.46 MW. Its hardware is estimated at $138M.

Epoch AI rates the certainty of this record as likely. I'm assuming these are A100s because H100s weren't out yet at the time and that is what the other cluster mentioned has

Full record

Owner: Tesla
Users: Tesla
Sector: Private
Status: Existing
Country: United States
Primary chip: NVIDIA A100
Primary chip count: 4,032
Total AI chips: 4,032
H100 equivalents: 1.3K
8-bit throughput: 2.5×10¹⁸ OP/s
Power capacity: 3.46 MW
Hardware cost: $138M
First operational: Aug 20, 2021
Rank when it came online: 9
Certainty: Likely