CLUSTERATLAS
A field guide to supercomputing

CHIE-4

SoftBank Corp.

CPU Accelerator Interconnect
i
Reading this map

CHIE-4 node architecture

CHIE-4 is operated by SoftBank Corp. 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
#21
HPL Rmax
135.4 PFlop/s
Theoretical peak Rpeak
151.88 PFlop/s
Ranked core count
662,256

Cluster · NVIDIA DGX B200 · public profile

Operator / site
SoftBank Corp.
Country
Japan
Profile scope
Ranked system / partition
Data checked
14 September 2026
Manufacturer
Nvidia
Ranked core count
662256
Theoretical peak (PFlop/s)
151.88
Operating system
Ubuntu 24.04.2 LTS

NVIDIA DGX B200 · public profile

Items marked ▧ come from the documented NVIDIA DGX B200 platform specification.

System / node model
NVIDIA DGX B200
CPU model
Xeon Platinum 8570 56C 2.1GHz
CPU sockets per node
2
Cores per socket
56
Physical CPU cores
112
Accelerator model
NVIDIA B200 SXM 180GB
Physical accelerators per node
8
Host / unified memory
2 TB DDR5 (32 DIMMs, upgradable to 4 TB)
Accelerator memory
180 GB HBM3e per GPU · 1,440 GB total
Accelerator ↔ accelerator link
NVLink 5 · all GPU pairs
On-node fabric
NVLink 5 through 2 × 5th-generation NVSwitch, 14.4 TB/s aggregate GPU bandwidth
CPU ↔ accelerator link
PCIe Gen5
Network interfaces per node
8
Interface model
NVIDIA ConnectX-7 single-port InfiniBand
Interface speed
400 Gb/s InfiniBand per adapter
Interface-to-accelerator topology
One adapter per GPU
Management / storage networking
2 × BlueField-3 DPU dual-port (storage and in-band management)
Local storage
2 × 1.92 TB NVMe M.2 SSD in RAID 1 (OS) + 8 × 3.84 TB NVMe U.2 SSD in RAID 0 (data cache)
Form factor
10U rackmount · 6 × 3.3 kW power supplies (5+1 redundant) · 14.3 kW maximum

Scope and uncertainty

SoftBank states CHIE-4 carries over 4,000 NVIDIA Blackwell GPUs in NVIDIA DGX B200 systems, and that the build completed on 22 July 2025. It publishes no DGX-system or node count, so none is recorded. Node specifications follow the documented DGX B200 platform. Node specifications follow the documented NVIDIA DGX B200 platform (10U rackmount, 6 × 3.3 kW power supplies (5+1 redundant), 14.3 kW maximum); operator-specific deviations are not published.

Documented fields: 23 / 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 — 4 / 7

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

NVIDIA DGX B200 · public profile — 19 / 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