ND A100 v4

Compute

631 H100e

Chips

2,000

First operational

Jun 1, 2021

ND A100 v4 is a large GPU cluster operated by Microsoft, located in the United States, in the private sector. It came online around June 2021. At that moment it ranked number 12 among known clusters.

It is built on 2,000 NVIDIA A100. That amounts to about 631 H100e of AI compute. The cluster draws roughly 1.72 MW. Its hardware is estimated at $54M.

Epoch AI rates the certainty of this record as likely. They ran the high performance LINPACK benchmark on a pool of 164 8xA100 VMs, so the clusters can include at least 1312 A100s. This might be a slice of a larger cluster. Estimating as 2k since it says "thousands"

Full record

Owner: Microsoft
Users: Cloud
Sector: Private
Status: Existing
Country: United States
Primary chip: NVIDIA A100
Primary chip count: 2,000
Total AI chips: 2,000
H100 equivalents: 631
8-bit throughput: 1.2×10¹⁸ OP/s
Power capacity: 1.72 MW
Hardware cost: $54M
First operational: Jun 1, 2021
Rank when it came online: 12
Certainty: Likely