RIKEN to Deploy Two NVIDIA-Accelerated Supercomputers for AI for Science and Quantum Computing
Japan's RIKEN is building two new supercomputers using NVIDIA GB200 NVL4 systems to advance AI for science and quantum computing. Together they bring 2,140 Blackwell GPUs into RIKEN's infrastructure, strengthening Japan's sovereign AI strategy and secure domestic compute for science, industry, and technology.
What's being built
- AI for science system: 1,600 NVIDIA Blackwell GPUs on GB200 NVL4, linked by NVIDIA Quantum-X800 InfiniBand. Focus areas include life sciences, materials science, climate and weather, manufacturing, and lab automation.
- Quantum-focused system: 540 NVIDIA Blackwell GPUs on GB200 NVL4 with Quantum-X800 InfiniBand for quantum algorithms, hybrid simulation, and quantum-classical workflows.
- Total capacity: 2,140 Blackwell GPUs across both systems.
Why this matters for researchers
For teams running large models, complex simulations, or data-heavy workflows, these systems provide scaled GPU acceleration with high-bandwidth interconnects suited for multi-node training and mixed-precision compute. The quantum-focused system creates a testbed for hybrid algorithms and near-term quantum-classical methods.
Because the systems are in-country, labs and industry partners in Japan gain secure access paths for sensitive projects and faster iteration cycles, with less dependency on overseas resources.
Connected to FugakuNEXT
These supercomputers will also act as proxy platforms for codesigning hardware, software, and applications for FugakuNEXT - the successor to Fugaku being codesigned by Fujitsu and NVIDIA. FugakuNEXT is planned to use FUJITSU-MONAKA-X CPUs paired with NVIDIA technologies via NVLink Fusion, targeting major application-level speedups over CPU-centric systems and future integration with production quantum computers.
Translation: teams can validate algorithms, optimize kernels, and de-risk scaling now, then carry learnings into FugakuNEXT ahead of its planned 2030 operation.
Software stack and method updates
- Floating-point emulation for Tensor Cores: RIKEN and NVIDIA are collaborating on FP emulation that lets traditional scientific codes benefit from Tensor Core throughput without sacrificing numerical intent.
- CUDA-X libraries: RIKEN plans to use CUDA-X's 400+ GPU-accelerated libraries, microservices, and tools to speed up HPC and AI for science applications across domains.
Timelines
- RIKEN systems: Operational in spring 2026.
- FugakuNEXT: Targeted for operation by 2030.
Practical steps for your lab
- Audit top workloads for GPU readiness: mixed precision paths, memory footprint, and node-to-node communication.
- Profile communication patterns under InfiniBand; prioritize overlap of compute and I/O and adopt asynchronous pipelines.
- Adopt CUDA-X libraries where possible before custom kernels; reserve hand-tuned kernels for clear hotspots.
- Prepare hybrid AI + simulation and quantum-classical baselines; define metrics and datasets now to compare across systems later.
- Containerize environments and CI test against multiple GPU architectures to reduce porting time.
Leadership perspectives
"RIKEN has long been one of the world's great scientific institutions, and today it stands at the forefront of a new era in computing," said Ian Buck, vice president of hyperscale and high-performance computing at NVIDIA. "Together, we're helping Japan build the foundation for sovereign innovation that will drive breakthroughs to solve the world's most complex scientific and industrial challenges."
"Integrating the NVIDIA GB200 NVL4 accelerated computing platform with our next-generation supercomputers represents a pivotal advancement for Japan's science infrastructure," said Satoshi Matsuoka, director of the RIKEN Center for Computational Science. "Our partnership will create one of the world's leading unified platforms for AI, quantum and high-performance computing, allowing researchers to unlock and accelerate discoveries in fields ranging from basic sciences to industrial applications for businesses and society."
Where to follow progress
Your membership also unlocks:
