CLUSTERATLAS
A field guide to supercomputing

Meta Research SuperCluster

Meta

CPU Accelerator Interconnect
i
Reading this map

Meta Research SuperCluster node architecture

Meta Research SuperCluster is operated by Meta in United States. This profile covers January 2022 first-phase snapshot.

Cluster interconnect: NVIDIA Quantum InfiniBand · 1,600 Gb/s per node (2022 report). Topology: Not verified.

DGX A100 · 2022 first phase

Node count
760
CPU
AMD EPYC 7742
CPUs per node
2
Accelerator
NVIDIA A100
Physical accelerators per node
8
Host / unified memory
Not verified
Accelerator memory
Not verified
Local accelerator links
NVLink / NVSwitch
CPU-to-accelerator link
PCIe 4.0
Storage
Not verified

Scope and uncertainty

January 2022 first-phase snapshot: 760 DGX A100 systems / 6,080 GPUs. Later expansion plans are not counted as installed hardware. CPU and on-node links follow the documented DGX A100 platform; installed memory variant is unverified.

Not verified means a detail has not been confirmed in this catalog. Diagrams are schematic; node indices are illustrative, not physical hostnames. Data checked 14 September 2026.

Sources

Browse all cluster architecture pages →

Data & sources