Lawrence Berkeley NL NERSC Perlmutter

Compute

2.3K H100e

Chips

7,168

First operational

May 17, 2021

Lawrence Berkeley NL NERSC Perlmutter is a large GPU cluster operated by US Department of Energy, located in the United States, in the public sector. It came online around May 2021. At that moment it ranked number 2 among known clusters.

It is built on 6,144 NVIDIA A100 SXM4 40 GB, alongside NVIDIA A100 SXM4 80 GB. That amounts to about 2.3K H100e of AI compute. The cluster draws roughly 6.9 MW. Its hardware is estimated at $153M.

Epoch AI rates the certainty of this record as confirmed.

Full record

Owner: US Department of Energy
Users: Scientific research
Sector: Public
Status: Existing
Country: United States
Location: Lawrence Berkeley National Laboratory, Building 59, 1 Cyclotron Rd, Berkeley, CA 94720
Primary chip: NVIDIA A100 SXM4 40 GB
Primary chip count: 6,144
Secondary chip: NVIDIA A100 SXM4 80 GB
Total AI chips: 7,168
H100 equivalents: 2.3K
8-bit throughput: 4.5×10¹⁸ OP/s
Power capacity: 6.9 MW
Hardware cost: $153M
First operational: May 17, 2021
Rank when it came online: 2
Certainty: Confirmed