Seiran, the “Blue Gale”: a word from a poem by the monk Saigyō that stands for intellect and wind. The AI supercomputer of the AI Technology Center at Tokyo University of Technology.
Discover SEIRAN
Twelve DGX B200 nodes, 96 Blackwell GPUs. Specifications, rankings and the origin of the name.
Check eligibility and fees
SEIRAN is used per project. Each project needs a faculty member or equivalent as project leader.
Go to the user manual
Connection procedures and operating manuals are restricted to approved users.
- TOP500 world ranking
- #421
- November 2026 list
- HPCG world ranking
- #95
- November 2026 list
- NVIDIA DGX B200
- 96 GPUs
- 12 nodes (Blackwell)
- Base project allocation
- 300 GPU hours
- ¥100,000 per set
The machine
NVIDIA DGX B200 for large-scale computing.
Twelve DGX B200 systems with Blackwell-architecture GPUs, connected by InfiniBand NDR.
- GPU architecture
- Blackwell B200
- Interconnect
- InfiniBand NDR
- HPL measured
- 3.34 PFLOPS
- TOP500
- #421 worldwide

Name
The name “SEIRAN”
Named after a poem by the monk Saigyō. It pairs “blue,” the university’s symbol color, with the hope of raising a new wind of knowledge from Hachioji.
8 GPUs per node
NVIDIA DGX B200
Each node carries eight NVIDIA B200 GPUs. Built on the Blackwell architecture, it handles everything from AI training and inference to scientific computing.
Whole system
12 nodes · 96 GPUs
Twelve DGX B200 systems side by side. In the November 2026 lists, SEIRAN ranked #421 on the TOP500 and #95 on HPCG.
InfiniBand NDR
Inter-node network
The DGX B200 compute nodes are connected by InfiniBand NDR to support large-scale computation.
How it is used
Computing sized to your research.
The whole system has 12 nodes and 96 GPUs. As a rule, each job can use up to 4 GPUs; this is not a limit on how many GPUs a project can use at the same time. Specific conditions follow the official operating guidance.
Single jobs (1–4 GPUs)
For prototyping, model validation and small-to-medium training within a single node. Submit jobs through the Slurm scheduler or Open OnDemand.
- Up to 4 GPUs per job, as a rule
- Apptainer containers
Built for distributed computing
Twelve nodes connected by InfiniBand NDR. The resources assigned to each job follow the usage conditions and operational settings.
- InfiniBand NDR between nodes
- DGX B200 × 12 nodes
The full SEIRAN system (96 GPUs)
Ranked #421 on the TOP500 with a measured HPL performance of about 3.34 PFLOPS. Supports digital twins, AI-driven science, education and research.
- 96 NVIDIA B200 GPUs in total
- HPL measured 3.34 PFLOPS
News
Official announcements
- June 2026: #398 on the TOP500 and #88 on HPCG
- November 2025: first listed at #374 on the TOP500 and #83 on HPCG
- Introduction of SEIRAN and an AI/DX partnership agreement with Hachioji City
Run by AITC
AI Technology Center (AITC)
With SEIRAN as its core computing platform, the AI Technology Center (AITC) at Tokyo University of Technology supports cross-disciplinary research and the development of people.
- AI education
- Applied AI research
- Social implementation and ethics of AI
Using SEIRAN for your research.
Faculty, students and joint research projects at Tokyo University of Technology can check eligibility and fees for the GPU platform here.

