"Japan Presets Ambitious Goal: Supercomputer Aiming for 1000-Fold Enhancement in Scientific Research and Global AI Advancement"
Japan is set to take a significant leap in computing technology with the development of Fugakunext, a next-generation supercomputer aiming to achieve zetta-scale computing performance. The national project is led by RIKEN, Japan's national research institute, and Fujitsu Limited, the nation's top technology company by market share.
Specifications and Goals
Fugakunext is designed to surpass its predecessor, Fugaku, which achieved a peak performance of 537 petaflops with around 160,000 interconnected nodes, each node having 48 Arm cores and 32 GiB high-bandwidth memory (HBM). Fugakunext targets pushing this performance into the zettaFLOPS regime, implying a computational capacity roughly three orders of magnitude higher.
The contract scope includes the design and building of the overall supercomputer system, its computing nodes, and CPUs, signaling a new architecture and hardware tailored for extreme-scale computing. The FUJITSU-MONAKA and FUJITSU-MONAKA-X CPUs, engineered for high performance and energy efficiency, feature Fujitsu's unique microarchitecture optimized for advanced 3D packaging and ultra-low voltage circuit operation.
Applications and Impact
The primary goal of Fugakunext is to support applications in large-scale scientific simulations and AI-enhanced research, including lattice QCD simulations, chiral fermion computations, tensor network algorithms, path optimization, and quantum computation. It aims to bridge high-performance computing (HPC) and emerging fields in quantum and AI.
If successful, Fugakunext is expected to dramatically accelerate the pace of scientific discovery in fields such as nuclear physics, particle physics, material science, and climate modeling by providing unprecedented computational power and efficiency. It will also advance efficient AI integration into HPC workloads to enable more intelligent simulations, predictive modeling, and algorithmic optimizations.
Investment and Timeline
The development of Fugakunext is backed by an investment of over $750 million, and the basic design phase is scheduled to run until 27 February 2026. The project forms part of Japan's broader "AI for Science" strategy, which integrates artificial intelligence with simulation technologies and real-time data to accelerate scientific discovery.
The growing demand for a flexible platform that supports large-scale computing resources, particularly as generative AI and other data-intensive technologies drive research and development, is being addressed by Fugakunext. The HPCI Steering Committee, established by MEXT, has emphasized the need for such a platform to address the rising demand.
Notable Achievements and Future Prospects
Fugaku, the predecessor to Fugakunext, rose to the upper ranks of the Top500 list of the world's fastest supercomputers. During the COVID-19 pandemic, Fugaku played a critical role in pandemic modeling. Four research teams are assessing the technical and scientific benefits of a zetta-scale supercomputer, and if successful, Fugakunext is expected to achieve speeds around 1,000 times faster than today's leading systems.
The core of Fugakunext will consist of the FUJITSU-MONAKA3 and its successor, the MONAKA-X CPUs, which are being developed on 2-nanometer technology. The development of Fugakunext is expected to enable seamless integration with GPUs and other accelerators, making the system adaptable for a wide range of uses, from artificial intelligence to intricate scientific simulations.
In summary, Fugakunext is carefully engineered to be Japan’s leap into zetta-scale computing with goals spanning advanced scientific computation, AI co-design, and quantum computing readiness, promising transformative impacts across academia, industry, and national technological capabilities.
Artificial intelligence will greatly benefit from the development of Fugakunext, as it is designed to support AI-enhanced research and advance efficient AI integration into high-performance computing workloads. The design and building of Fugakunext, a next-generation supercomputer aiming for zetta-scale computing performance, is being led by RIKEN and Fujitsu Limited, emphasizing the integration of artificial intelligence with simulation technologies and real-time data as part of Japan's broader "AI for Science" strategy.
