CLUSTERATLAS
A field guide to supercomputing

ABCI 3.0

National Institute of Advanced Industrial Science and Technology (AIST)

CPU Accelerator Interconnect
i
Reading this map

ABCI 3.0 node architecture

ABCI 3.0 is operated by National Institute of Advanced Industrial Science and Technology (AIST) in Japan. 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
#19
HPL Rmax
145.1 PFlop/s
Theoretical peak Rpeak
181.49 PFlop/s
Ranked core count
479,232
Measured power
3596 kW
Energy efficiency
40.35 GFlop/s per watt

Cluster · H200 compute

Operator / site
National Institute of Advanced Industrial Science and Technology (AIST)
Country
Japan
Profile scope
Ranked system / partition
Data checked
14 September 2026
Manufacturer
HPE
Ranked core count
479232
Theoretical peak (PFlop/s)
181.49
Measured power (kW)
3596
Energy efficiency (GFlop/s per watt)
40.3503893214683
Operating system
Rocky Linux 9

H200 compute

Items marked ▧ come from the documented HPE Cray XD670 platform specification.

System / node model
HPE Cray XD670
Compute nodes of this type
766
CPU model
Intel Xeon Platinum 8558 · 48 cores
CPU sockets per node
2
Cores per socket
48
Physical CPU cores
96
Accelerator model
NVIDIA H200
Physical accelerators per node
8
Host / unified memory
2 TB DDR5
Accelerator memory
141 GB HBM3e per GPU
Network interfaces per node
8
Interface speed
200 Gb/s per port (InfiniBand NDR200)
Interface-to-accelerator topology
Eight InfiniBand NDR200 ports per compute node
Management / storage networking
One additional InfiniBand HDR port per node for storage
Local storage
2 × 7.68 TB NVMe
Form factor
5U single-node chassis · 6 Titanium power supplies (N+2)

Scope and uncertainty

AIST states that compute nodes communicate at full-bisection bandwidth but names no topology on its own pages; a three-tier fat-tree appears only in a paper by AIST staff, so no topology is recorded here. Node specifications follow the documented HPE Cray XD670 platform (5U single-node chassis, two sockets, eight SXM5 accelerators); operator-specific deviations are not published.

Documented fields: 22 / 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 — 6 / 7

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

H200 compute — 16 / 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 →

Data & sources