IBM Blue Vela

Compute

1K H100e

Chips

1,024

First operational

Jul 7, 2024

IBM Blue Vela is a large GPU cluster operated by IBM, located in the United States, in the private sector. It came online around July 2024. At that moment it ranked number 77 among known clusters.

It is built on 1,024 NVIDIA H100 SXM5 80GB. That amounts to about 1K H100e of AI compute. The cluster draws roughly 1.46 MW. Its hardware is estimated at $36M.

Epoch AI rates the certainty of this record as likely. There's 128x8=1,024 GPUs per pod, and it's implied that there's multiple pods, implying that there are thousands of GPUs, but this is all the information I have

Full record

Owner: IBM
Users: IBM
Sector: Private
Status: Existing
Country: United States
Primary chip: NVIDIA H100 SXM5 80GB
Primary chip count: 1,024
Total AI chips: 1,024
H100 equivalents: 1K
8-bit throughput: 2×10¹⁸ OP/s
Power capacity: 1.46 MW
Hardware cost: $36M
First operational: Jul 7, 2024
Rank when it came online: 77
Certainty: Likely