Skip to main content

About SEIRAN

System specifications, performance and the origin of the name of SEIRAN, the NVIDIA DGX B200 supercomputer at Tokyo University of Technology.

On this page

SEIRAN — The Blue Gale of Intelligence

“Seiran” is a word that appears in a poem by the monk Saigyō, composed when he visited Hachioji, and stands for intellect and wind. The English name “Blue Gale” expresses the new whirlwind of knowledge that AI brings.

SEIRAN is a high-performance GPU supercomputer introduced by Tokyo University of Technology in 2025. It supports research and education, from training and inference of generative AI to scientific computing.

Hardware configuration

SEIRAN consists of 12 NVIDIA DGX B200 systems based on the NVIDIA Blackwell architecture, with 96 GPUs in total.

The SEIRAN DGX B200 server racks
The racks / Hachioji Campus, Tokyo University of Technology

Compute

Compute nodes
NVIDIA DGX B200 × 12 nodesBlackwell architecture
GPUs
96 in total8 per node
GPU memory
180 GBAs listed on the TOP500
CPU
Intel Xeon Platinum 857056 cores, 2.1 GHz

Performance

HPL measured (Rmax)
About 3.34 PFLOPS
HPL theoretical peak (Rpeak)
About 3.66 PFLOPS
HPCG
About 102.43 TFLOPS
FP8 theoretical
About 0.9 EFLOPSA theoretical value under a different precision and conditions

Network

Inter-node links
InfiniBand NDRConnects the 12 nodes

At benchmark submission

OS
Ubuntu 24.04.3 LTS
TOP500 core count
15,552Not CPU cores alone

Benchmarks and rankings

SEIRAN is listed on the TOP500 and HPCG, the performance rankings for supercomputers.

TOP500 world ranking (November 2026)

#421

Previous list (June 2026): #398.

HPCG world ranking (November 2026)

#95

November 2026 published results

TOP500
Ranks supercomputers by their measured performance solving a dense system of linear equations with HPL.
HPCG
A benchmark that includes sparse-matrix computation, memory access and communication, which matter in scientific computing.

Research applications

Digital twins

Supports research that recreates spaces such as cities and data centers in a virtual environment and connects them to simulation and AI-based analysis and prediction.

AI for Science and scientific computing

Supports large-scale parallel computation in fluid dynamics, electromagnetic field analysis and numerical linear algebra, as well as AI for Science research such as Physics-Informed Neural Networks (PINNs) that combine physical models with machine learning.

Generative AI and large language models

Supports research that needs GPU resources, such as training, fine-tuning and inference of large language models.

Explainable AI (XAI)

Supports research on methods that analyze and visualize the grounds for AI decisions, and their application to engineering, medicine and other fields.

Read about research and activities

Official sources

Rankings and HPL/HPCG performance figures are from the November 2026 lists. AI theoretical performance and HPL measured performance use different computing conditions.