FugakuNEXT

TOP FugakuNEXT

-Pioneering the Future with AI and simulation-

The Role of the New Flagship Supercomputer, “FugakuNEXT”

In recent years, demand for computational resources has surged due to advances in simulation and data science, alongside the rapid rise of generative AI. At the same time, the needs for computational resources are becoming increasingly diverse, driven by the growing importance of new scientific research that integrates AI, simulation, automated experimentation, and real-time data.

Given this background, the new flagship supercomputer, FugakuNEXT is expected not only to further advance the simulation capabilities of conventional supercomputers but also achieve world-leading performance in AI. Its mission is to serve as a next-generation AI-HPC platform where AI and HPC work seamlessly together.

Development Structure for FugakuNEXT

To achieve world-leading performance in both AI and simulation, the development of FugakuNEXT will proceed through a three-party collaboration let by RIKEN, working together with Fujitsu and NVIDIA. This partnership brings together Fujitsu's expertise in high-performance ARM-based CPU and system integration, and NVIDIA's GPU technology and global ecosystem with the highest global market share in AI/HPC GPUs.

Please Click to open the Fujitsu website in a new tab.

In addition, software development, including system software, will be conducted through an open framework involving international collaboration, complementing the tripartite partnership. These combined efforts aim to deliver a competitive system achieving the world's first zetta-scale performance in AI (FP8) and to build a global ecosystem through worldwide deployment.

Please Click to open the NVIDIA website in a new tab.

FugakuNEXT Development Policy

The development of FugakuNEXT will be conducted by the three pillars of strategies: “Made with Japan,” “Technological Innovation,” and “Sustainability/Continuity.” These principles are based on the findings of the “Final Report on Next-Generation Computing Infrastructure” and the “Research Study on Next-Generation Computing Infrastructure”, as well as the experience and lessons learned from the development and operation of supercomputer “Fugaku”.

Based on these development principles, we will expand the frontiers of computation through a next-generation AI-HPC platform, create new science through “AI for Science” , secure Japan's leadership in advanced AI technology and computational infrastructure, and advance sustained R&D based on semiconductor and computational resource roadmaps. This will build a global FugakuNEXT ecosystem and further strengthen Japan's semiconductor industry and information infrastructure.

Operational Policy

FugakuNEXT will be deployed adjacent to the RIKEN Kobe site, with operations targeted to begin around 2030. We will establish an operational environment that minimizes any period of unavailability of computational resources during the system migration from “Fugaku, aiming to continuously provide world-class computational performance and resources in a stable manner. Furthermore, the system will pursue energy efficiency and low carbonization by combining operational technologies that promote the adoption of advanced cooling techniques and the use of renewable energy source. In addition, we plan to further evolve AI-driven operations and user support to ensure the sustainability and efficiency of the computational infrastructure, while expanding into new computational domains through integration with quantum computers, thereby providing a research environment accessible to all.

Research and Development Theme

Architecture Area

In the development of FugakuNEXT we will investigate trends in semiconductor manufacturing technology, packaging technology, and memory technology, while conducting research and development primarily related to hardware design, such as the microarchitecture of the CPU and accelerators as well as their memory subsystems, the compute node architecture, the interconnection networks for scale-up and scale-out, and the entire system. Specifically, we will co-design the hardware system together with the application group, taking into account the characteristics of the applications as well as conducting research on application performance modeling for the future system.

System software Area

To achieve the target performance and enhance the usability of FugakuNEXT, we will develop system software that fully unleashes the potential of the underlying hardware. The developed software will be released as open source, promoting continuous development and ecosystem growth through collaboration with international OSS communities. Furthermore, by organically integrating development environments, numerical and communication frameworks, and AI software, we aim to create a next-generation HPC–AI integrated platform.

Application Development Area

Our area conducts a series of research and development activities, including the development, support, and performance evaluation of HPC applications, as well as benchmarks, with a focus on co-design for next-generation HPC infrastructure namely Fugaku NEXT. We also aim to create fundamental technologies that accelerate and enhance the functionality of HPC applications by integrating simulation and advanced AI technologies. To effectively and efficiently promote these activities, we will develop and release various frameworks.

The FugakuNEXT Application Development Seminar

We are developing a highly efficient data center to support FugakuNEXT and exploring advanced operation technologies that make use of operational data and automation. By optimizing the data center and the supercomputer together, we aim to build a sustainable and reliable computing environment that is easy to use for everyone.