RIKEN, Fujitsu, NVIDIA Collaborate on FugakuNEXT Supercomputer

RIKEN, in collaboration with Fujitsu and NVIDIA, has initiated a project to co-design FugakuNEXT, a successor to the Fugaku supercomputer, commencing after an announcement on August 22. This undertaking, supported by Japans MEXT, aims to create a hybrid AI-HPC system featuring FUJITSU-MONAKA-X CPUs paired with NVIDIA technologies via NVLink Fusion, and will focus on applications including earth systems modelling, disaster resilience, drug discovery, and advanced manufacturing. The collaboration will also encompass software development for mixed-precision computing and performance optimisation, with the intention of serving universities, government agencies, and industry partners nationwide.

A New Era in Japanese Supercomputing

RIKEN, in collaboration with Fujitsu and NVIDIA, is co-designing FugakuNEXT, the successor to the Fugaku supercomputer, to address critical research priorities for Japan. This collaboration was formally initiated at the FugakuNEXT International Initiative Launch Ceremony held in Tokyo, and enables partners to shape the system’s architecture. The effort will highlight Japan’s adoption of modern AI and showcase innovations for researchers and enterprises globally.

FugakuNEXT will be a hybrid AI-HPC system, integrating both simulation and AI workloads. The system will feature FUJITSU-MONAKA-X CPUs, paired with NVIDIA technologies via NVLink Fusion, a silicon technology enabling high-bandwidth connections. This design aims to deliver speed, scale, and efficiency in computational performance.

The new supercomputer will support applications such as automating hypothesis generation, code creation, and experiment simulation. Specific areas of focus include accelerating simulations using surrogate models and physics-informed neural networks, enhancing manufacturing processes through AI-driven design, and improving earth systems modeling for disaster preparedness and prediction. RIKEN, Fujitsu, and NVIDIA will jointly develop software, including tools for mixed-precision computing, continuous benchmarking, and performance optimization.

Backed by Japans MEXT (Ministry of Education, Culture, Sports, Science and Technology), FugakuNEXT will serve universities, government agencies, and industry partners nationwide. The development of Japanese Supercomputing FugakuNEXT signifies a strategic investment in Japans future, built on sovereign infrastructure and a commitment to scientific leadership.

Architectural Innovations and Technological Foundations

FugakuNEXT will utilize NVIDIAs whole software stack, including NVIDIA CUDA-X libraries such as NVIDIA cuQuantum for quantum simulation, RAPIDS for data science, NVIDIA TensorRT for high-performance inference and NVIDIA NeMo for large language model development, alongside other domain-specific software development kits tailored for science and industry. Innovations developed on FugakuNEXT could potentially serve as blueprints for systems worldwide.

The system will be built around FUJITSU-MONAKA-X CPUs, designed to be paired with NVIDIA technologies using NVLink Fusion, a new silicon technology that facilitates high-bandwidth connections between Fujitsus CPUs and NVIDIAs architecture. This combination aims to achieve speed, scale, and efficiency in computational performance.

FugakuNEXT will support a wide range of applications, including automating hypothesis generation, code creation and experiment simulation. Scientific research will be accelerated through the use of surrogate models and physics-informed neural networks, while manufacturing processes will benefit from AI-driven design optimisation. Furthermore, the system will aid in earth systems modeling, specifically for disaster preparedness and prediction related to earthquakes and severe weather events.

RIKEN, Fujitsu and NVIDIA will collaborate on software developments, including tools for mixed-precision computing, continuous benchmarking, and performance optimization. This collaborative effort underscores a commitment to refining computational processes and maximising system performance.

Applications and Strategic Objectives

FugakuNEXT will be a hybrid AI-HPC system, designed to combine simulation and AI workloads. The system’s architecture will feature FUJITSU-MONAKA-X CPUs paired with NVIDIA technologies using NVLink Fusion, a new silicon technology enabling high-bandwidth connections between Fujitsus CPUs and NVIDIAs architecture. This combination is intended to deliver speed, scale and efficiency.

FugakuNEXT will support a wide range of applications, including automating hypothesis generation, code creation and experiment simulation. Scientific research will benefit from accelerated simulations using surrogate models and physics-informed neural networks, while AI will be used to learn from simulations to generate efficient and aesthetically pleasing designs in manufacturing. The system will also aid in earth systems modeling, supporting disaster preparedness and prediction for earthquakes and severe weather events.

RIKEN, Fujitsu and NVIDIA will collaborate on software developments, including tools for mixed-precision computing, continuous benchmarking, and performance optimization. These collaborative efforts will refine computational processes and maximise system performance, potentially establishing blueprints for systems worldwide.