NVIDIA Unveils ‘Fugaku NEXT’: Japan’s Next-Gen AI Supercomputer Promising 100x Performance Boost

Japan is accelerating its investment in artificial intelligence (AI) technologies through a strategic partnership with industry leaders NVIDIA, Fujitsu, and RIKEN. This collaboration is set to create the nation’s most potent supercomputer, marking a significant milestone in computing power.

NVIDIA and Fujitsu Join Forces to Build a Zetta-Scale Supercomputer in Japan

As a vanguard in AI investment, Japan is working closely with NVIDIA and various corporate entities to bolster its technological prowess. Recent disclosures from NVIDIA’s blog indicate that RIKEN, a prominent research institute, will collaborate with both NVIDIA and Fujitsu to engineer a pioneering computing system. This project follows the legacy of the renowned ‘Fugaku’ supercomputer and marks a significant leap by integrating NVIDIA’s AI accelerators and Fujitsu’s MONAKA-X server CPUs.

System design and computing advancements by Fujitsu, Riken, and NVIDIA in scientific fields.

The upcoming machine, referred to as Fugaku Next, is poised to become Japan’s premier platform for both AI and high-performance computing (HPC).It supports diverse applications, from simulation to groundbreaking AI research, significantly advancing the scientific landscape. Projected performance metrics suggest that Fugaku Next aims for up to 600 exaFLOPS in FP8 sparsity, with further ambitions to reach a zetta-scale computing capability—an unprecedented achievement for any supercomputer globally.

Infographic: Road to 100x Application Performance by 2030 with FugakuNEXT — Image Credits: RIKEN

This groundbreaking supercomputer represents a synergistic hardware initiative between NVIDIA and Fujitsu, focusing on developing optimal CPU and GPU functionalities. The proposed architecture will incorporate NVLink Fusion technology for seamless onboard connectivity, paired with NVIDIA’s CUDA-X libraries tailored for specific applications, including quantum computing and large language models (LLMs).These advancements are expected to render Fugaku Next a unique asset in the realm of AI supercomputing.

Map of Japan indicating NVIDIA's 15-year collaboration with universities and supercomputers marked by dots.

Ian Buck, NVIDIA’s VP of HPC, highlighted during the announcement that the supercomputer would not be confined to existing Blackwell or Rubin architectures. Collaborating with Fujitsu opens doors for potential ‘newer inventions, ’ possibly introducing unique frameworks within the system. The two companies anticipate achieving a remarkable ‘100-fold’ performance enhancement over the original Fugaku, driven by innovations in architecture and simulation methodologies.

For further insights, visit the news source: PCWatchImpress.

Source & Images