- Fugakounext will combine the Fujitsu CPU and the NVIDIA GPUs in the next national Japan supercomputer
- The system is aimed at 600Eflops FP8 Performance with a 100x application performance gain
- Riken, Fujitsu and Nvidia see the project how to define a new AI-HPC standard
Japan is preparing its next national supercomputer, Fugakounext, through a collaboration between Fujitsu, Nvidia and Riken.
The system is planned for operation around 2030 and aims to combine artificial simulation and intelligence on a strictly integrated platform.
For the first time in a Japanese flagship project, GPUs will be used as accelerators. Nvidia will design (as expected) the GPU infrastructure, Fujitsu will handle the CPU and the integration of the system, and Riken will be involved in the software and algorithm work.
Feynman GPU
The result is expected to be an “AI-HPC platform” designed for science, industry and the discovery promoted by AI.
The performance objectives for the supercomputer are certainly ambitious. Fugakounext is designed to deliver more than 600Eflops of AI FP8 performance, which would make it the most powerful until now announced.
The system is also expected to achieve up to an increase of up to a hundred times in the application performance compared to Fugaku, while remains within approximately the same 40MW power budget.
Nvidia’s long -term road map points to Feynman’s GPU architecture (called by the theoretical physicist Richard Feynman) that arrives about 2028, so it could well play a role in Fugakunext food.
Fujitsu is developing a successor to its Monaka CPU for the project, tentatively called Monaka-X, with more nuclei, extended Simd capabilities and the Matrix of ARM computing engine for the inference of AI.
Together with the NVIDIA accelerators, the system is expected to execute great simulations together with the demanding workloads.
The hardware will not only deliver the objective profits, so the project will also support innovations such as substitute models, mixed precision arithmetic and neural networks informed by physics to accelerate performance and at the same time preserve precision.
Makoto Gonokami, president of Riken, said: “It is a great honor for Riken to collaborate with Fujitsu and Nvidia to advance the development of Fagakunext. Since ancient times, humanity has created civilizations and advanced societies through the science of computer science. in the transformation of the update. “
Ian Buck, vice president of Nvidia, added: “Fugakounext will offer Zettascale’s performance with almost 100 times faster application speeds, within the same energy footprint as its predecessor, accelerating research, increase industrial competitiveness and boost progress for people in Japan and worldwide.”
