Selene

NVIDIA Corporation

Node type

CPU Accelerator Interconnect

Reading this map

Selene node architecture

Selene is operated by NVIDIA Corporation in United States. This profile covers Ranked system / partition. Specifications are listed below; anything the operator has not published is itemised in the documented-fields list.

Published benchmark (TOP500 · June 2026)

HPL rank: #45
HPL Rmax: 63.46 PFlop/s
Theoretical peak Rpeak: 79.22 PFlop/s
Ranked core count: 555,520
Measured power: 2646 kW
Energy efficiency: 23.98 GFlop/s per watt

Cluster · DGX A100 platform

Operator / site: NVIDIA Corporation
Country: United States
Profile scope: Ranked system / partition
Data checked: 14 September 2026
Manufacturer: Nvidia
Ranked core count: 555520
Theoretical peak (PFlop/s): 79.22
Measured power (kW): 2646
Energy efficiency (GFlop/s per watt): 23.98337112622827
Operating system: Ubuntu 20.04.1 LTS
Network topology: Three-level fat-tree (leaf, spine, core) with 850 switches

DGX A100 platform

Items marked ▧ come from the documented NVIDIA DGX A100 platform specification.

System / node model: NVIDIA DGX A100
Compute nodes of this type: 560
CPU model: AMD EPYC 7742 64C 2.25GHz
CPU sockets per node: 2
Cores per socket: 64
Physical CPU cores: 128
Accelerator model: NVIDIA A100
Physical accelerators per node: 8
Host / unified memory: 2 TB (32 × 64 GB)
Accelerator memory: 80 GB per GPU
Accelerator ↔ accelerator link: NVLink / NVSwitch
On-node fabric: NVLink 3 through NVSwitch, 600 GB/s GPU-to-GPU
CPU ↔ accelerator link: PCIe 4.0
Network interfaces per node: 8
Interface model: NVIDIA ConnectX-6 / ConnectX-7 (platform configuration)
Interface speed: 200 Gb/s per HCA (HDR InfiniBand)
Interface-to-accelerator topology: One adapter per GPU
Management / storage networking: Two additional 200 Gb/s HDR InfiniBand HCAs per node dedicated to storage
Local storage: 2 TB OS drive plus 30 TB internal NVMe SSD (8 × 3.84 TB)
Form factor: 6U rackmount

Scope and uncertainty

Selene was expanded after launch: NVIDIA's Hot Chips 2020 tutorial describes 280 DGX A100 systems, and the Megatron-Turing NLG paper by NVIDIA and Microsoft authors describes 560. The ranked entry is the expanded configuration — the TOP500 core count doubles between the June and November 2020 lists — so 560 is recorded. Host memory, accelerator memory and node-local storage follow NVIDIA's own SPEChpc submission for Selene. Node specifications follow the documented NVIDIA DGX A100 platform (6U rackmount); operator-specific deviations are not published.

Documented fields: 27 / 29

Every field this catalog checks. Fields listed above have a value; the remainder are shown here. A field marked absent is not publicly available in the sources we could verify, which does not mean the hardware is missing. Fields that cannot apply, such as benchmark results for an unranked system, are excluded from the total.

Cluster — 7 / 7

Manufacturer
Ranked core count
Theoretical peak (PFlop/s)
Measured power (kW)
Energy efficiency (GFlop/s per watt)
Operating system
Network topology

DGX A100 platform — 20 / 22

System / node model
Compute nodes of this type
CPU model
CPU sockets per node
Cores per socket
Physical CPU cores
CPU note
Accelerator model
Physical accelerators per node
Host / unified memory
Accelerator memory
Accelerator ↔ accelerator link
On-node fabric
CPU ↔ accelerator link
Network interfaces per node
Interface model
Interface speed
Interface-to-accelerator topology
Aggregate scale-out per node
GPUDirect RDMA
Management / storage networking
Local storage
NVMe data disks
Max remote data disks
Uncached disk IOPS
Uncached disk throughput (MBps)
Form factor

Sources

Browse all cluster architecture pages →