Japan You don’t want to be lagged in supercomputing, so it is already Preparing the successor of your Fugaku superorderwho has called Fugakunextand that will develop through a collaboration between Fujitsu, Nvidia and Rikenthe largest institution in Japan dedicated to research and development tasks.
If everything goes according to the plans of the three parts, Fugakounext will enter into operation over 2030, and will be characterized by, among other things, mixing simulation and artificial intelligence on a platform with a very high level of integration.
In addition, for the first time in a high -level project that Japan leads, GPUS will be used as accelerators. Hence the presence of Nvidia in the group of entities and companies that will develop it. Thus, Nvidia will take care of the design of her GPUS infrastructure, and Fujitsu will take over the integration between the CPUS and the system. Riken’s work in the development of Fugakounext will be more related to the part dedicated to software, as well as the development of algorithms.
When it is working, it will be a super-tider for high performance computing tasks that imply work with AI, that is, it will be an AI-HPC platform prepared for tasks and research related to both science and different sectors of the industry.
The objectives that have set their builders for their performance are quite ambitious, since they want it to exceed 600 Exaflops of the IA FP8 yield, which would make it the supercomputer for the most powerful of those that have been announced so far for the future. In addition, it is expected that its performance with applications with respect to Fugaku multiplies by 100, but with approximately the same energy consumption that Fugaku currently has.
As for its architecture, everything indicates that it will use the Feynman architecture of Gpus, whose name is a tribute to the physicist Richard Feynman, which is not yet available. It is estimated that it will arrive in 2028, on time for its integration in Fugakounext.
Your CPU will not be one of the currently available. Fujitsu is developing a specific one, successor to his monoka chip. It is unknown what its definitive name will be, and for now it is known as Monoka-X. It will have an ARM matrix computing engine for AI inference, expanded simd capabilities and more nuclei than the current one. Its combination with NVIDIA accelerators will make the system capable of managing large -scale simulations, as well as supporting workloads that need a high computing power.
In spite of the computing power it will offer, only with hardware the improvements that project participants have been marked as a goal, so they will also rely on innovations related to software and algorithmic are not going to be achieved. As subrogated models, precision arithmetic myst and physics -based neural networks to accelerate performance without neglecting precision.
Makoto Gonokami, Presidente de Rikenhe pointed out that «It is a great honor for Riken to collaborate with Fujitsu and Nvidia in the progress of the development of Fugakounext. Since ancient times, humanity has built civilizations and advanced societies thanks to the science of computer science. Today, the appearance of AI, advanced semiconductors and quantum computers is causing a discontinuous transformation into a computational science, a change of historical paradigm that opens completely new horizons«.
Vivek Mahajan, Executive Director, Vice President, Technical Director and Responsible for the Fujitsu Limited Systems Platform He has commented on his part that «Fujitsu has the honor of being part of this collaboration. Taking advantage of our avant-garde technology Made in Japan Fujitsu-Monoka CPU and the experience developed through systems construction projects such as Fugaku, we will contribute to the establishment of a next-generation next-generation computational infrastructure in the world, working in close collaboration with Riken and our associated suppliers«.
Besides, Ian Buck, Vice President of High Performance Computer and Hyperscale of Nvidiarecalled that «Japan is a world technological leader who promotes excellence in manufacturing, robotics and engineering, while focusing on resilience to natural disasters and population aging, and AI and HPC are in a unique position to help address these complex challenges. Thanks to the collaboration between Riken, Fujitsu and Nvidia, Fugakounext will offer a zetaescala performance with almost 100 times faster application speeds, with the same energy consumption as its predecessor, which will accelerate research, promote industrial competitiveness and promote the progress of people in Japan and worldwide«.
