Supporting JHPC-quantum Project with a Next-Generation High-Performance Computing System

As high-performance computing (HPC) and quantum computing converge, research institutions worldwide are building hybrid platforms to tackle problems beyond today's computational limits. In Japan,  RIKEN Center for Computational Science (R-CCS) finalized the system configuration for the "JHPC-quantum" project in 2025, placing Japan at the forefront of this shift.

Global scientific research is entering a new era driven by heterogeneous computing and emerging quantum-classical simulation approaches. As AI model sizes and scientific simulation complexity continue to grow, single-architecture computing frameworks are increasingly insufficient to meet future research demands. Japan, a global leader in HPC, continues to lead with the Fugaku and is now accelerating progress toward a Quantum – HPC hybrid platform.

This hybrid platform brings together multiple supercomputers and quantum computers, including:

• Fugaku
• IBM Quantum System Two "ibm_kobe"
• Ion-trap quantum computer "Reimei"

These systems will be connected through a high-speed network and a unified software framework, creating one of the first integrated research environments capable of supporting quantum algorithm simulation, integrated application development, and hybrid-computing experimentation. The computing hub is planned to be located at RIKEN's Kobe site and is scheduled to be officially operational within Japan's fiscal year 2025.

Within this national-scale initiative that bridges quantum and classical computing, Giga Computing's assigned mission is to design and manufacture 135 computing nodes, each equipped with NVIDIA's latest-generation Grace Blackwell Superchip (GB200 NVL4), totaling 540 GPUs.

This flagship AI computing architecture, combining Grace CPUs with Blackwell GPUs, is designed specifically for large language model training, physics simulations, and hybrid quantum algorithm optimization. To maximize performance and ensure long-term system stability, the project addressed three major technical challenges:

✔️ Thermal Management and Reliability Under Extreme Power Density

Each GB200 NVL4 node operates at exceptionally high-power levels, making traditional air cooling inadequate for such thermal density. Giga Computing’s design implements a high-efficiency warm-water cooling architecture capable of operation at facility temperatures above 32°C, enabling GPUs to remain within their optimal thermal efficiency range even at peak power while maintaining performance and energy efficiency.

✔️ Modular, High-Density Design and Rapid Delivery

To ensure the platform could be completed within the planned schedule, servers use a highly modular structure and rapid assembly capability. With high-density GPU modules and a streamlined server design, the system integrates GPUs, CPUs, networking modules, and liquid-cooling cold plates into a unified package, reducing assembly and test time.

✔️ Deep Integration with NVIDIA High-Speed Networking

The system adopts NVIDIA Quantum X-800 InfiniBand networking with an optimized design for low-latency signal paths to ensure stable, synchronized performance during cross-node data exchange. Built-in management firmware supports NVIDIA NVLink for high-speed intra-node interconnects.

Giga Computing contributed key engineering solutions in this project, serving as a foundation for the Quantum–HPC hybrid platform.

Building on data-center and supercomputing liquid-cooling experience, the team implemented a high-reliability warm-water cooling module that meets RIKEN's requirements, featuring:

• High heat-flux cold plates: Engineered for GB200 NVL4 chips to handle thermal loads exceeding 1,000 W
• Serviceable design: Quick-disconnect, zero-leak connectors enable fast replacement and reduced operational cost
• Energy recovery potential: Cooling water outlet temperatures can reach 45°C, allowing potential reuse for building heating or integration with heat recovery systems

This solution aligns with RIKEN's energy saving policies and supports sustainable operations.

To support high-density installation and high-speed inter-server connectivity, the platform uses a rack-level design approach. One integrated system approach facilitates future expansion and maintenance. Each rack integrates:

• GB200 NVL4-based servers
• InfiniBand switches
• Coolant Distribution Unit (CDU) modules
• Power management and monitoring systems

This rack-level design enables rapid deployment at RIKEN's Kobe computing center and seamless connectivity to Fugaku and quantum systems.

The system delivers world-class compute capability exceeding 21 PFLOPS in FP64 and 5 EFLOPS in FP8. Through precise thermal management and power-distribution, the design aims to exceed typical energy-efficiency baselines by >15% while reducing reliance on cooling infrastructure.

The overarching goal is to drive innovation and performance responsibly, balancing compute capability with energy efficiency.

The success of this project relies on cross-organizational collaboration. ScaleWorX serves as the system integrator; DTS Corporation oversees project planning and testing; NVIDIA provides the Grace Blackwell architecture and InfiniBand; DDN supplies the high-performance file system; and Giga Computing designs and manufactures compute nodes.

This multidisciplinary, international collaboration highlights the broader trend toward HPC ecosystem integration, and further demonstrates how shared expertise accelerates complex platform delivery.

"We are honored to collaborate with Japan's leading scientific research institutions. Leveraging Giga Computing's extensive experience in high-performance system design and manufacturing, together with close collaboration with NVIDIA," said Freya Yu, APAC Business Director at Giga Computing Technology. "We are accelerating Japan's continued breakthroughs in world-class research. We hope to become a key driving force in the next-generation computing revolution through designs that balance performance, efficiency, and sustainability."

This Quantum–HPC hybrid platform is more than a supercomputer; it is an experimental environment that bridges quantum and classical computing. Through the SQC interface, an API under the JHPC-quantum initiative that connects quantum computers and HPC systems, manages data exchange, and schedules jobs, the platform operates alongside Fugaku, "ibm_kobe" , and "Reimei", allowing researchers to use a unified environment for:

• Quantum algorithm simulation
• Hybrid AI / quantum machine-learning training
• Materials science and molecular dynamics

This enables breakthrough advances in energy development, drug discovery, financial simulation, and fundamental physics research.

The compute layer provides the horsepower for these workloads and demonstrates how classical HPC and quantum simulation can be orchestrated in one system, opening new possibilities for computational research.

Participating in the construction of RIKEN’s Quantum–HPC hybrid platform is a significant milestone in bringing quantum and classical computing together at scale.

As quantum and HPC convergence continues to mature, the industry collaborators’ role is to support researchers with reliable systems and sustainable designs, allowing the platform to deliver real scientific results.

Category	Key Highlights
Compute Nodes	135 nodes, 540 GPUs (NVIDIA GB200 NVL4)
Networking	NVIDIA InfiniBand XDR, 3.2 Tbps bandwidth
Liquid Cooling	Warm-water cooling system (Warm-Water Cooling, >40°C)
Performance	FP64: 21+ PFLOPS / FP8: 5+ EFLOPS
System Integration	Connected with Fugaku, ibm_kobe, and Reimei
Leading Organization	RIKEN Center for Computational Science
Giga Computing Role	GPU node design & manufacturing, rack-level thermal and layout engineering
Sustainability Value	High-efficiency architecture, heat recovery potential, modular design