The Japanese supercomputer Fugakujointly developed by Riken and Fujitsu, managed to maintain the ranked as the world’s fastest high-throughput machine for three consecutive periods (a year and a half) in the four main rankings.
For years there has been a battle to assemble the most powerful system. This career requires large investments in hardware to expand and renovate available equipment.
The set of processors that make up this supercomputer uses Fujitsu processors shielding the A64FX with 48 cores each and they use almost 16,000 of these processors with a total of 7.6 million processing cores.
This system was put at the service of Covid-19 therapy research, as part of the collaboration between Fujitsu Japan and researchers at the University of Tokyo Advanced Science and Technology Research Center.
It is developed jointly by Riken and Fujiitsu. Photo: DPA
On its commercial side it develops high performance simulations for industrial use cases with commercial applications.
This supercomputer has managed to stay in the first position of rankings as TOP500, HPCG, HPL-AI and Graph 500, for three consecutive periods, as highlighted by the multinational Fujitsu in a statement.
On this occasion, the results have been obtained with the full staff of Fugaku, which has with 158,976 nodes spread over 432 racks.
In the Top500, a list of the world’s highest performing supercomputers, Fugaku achieved a Linpack score of 442.01 petaflops.
The top ten supercomputers, according to Graph 500.
On HPCG, a performance rating for technology methods often used for real-world applications, it scored 16.00 petaflops.
In HPL-AI, which ranks supercomputers based on their high efficiency in calculating single and medium precision, commonly used in artificial intelligence applications, obtained a score of 2,004 exaflops.
Fujitsu also highlights Graph 500, which categorizes systems based on graph analysis performance, an important element in data-intensive workloads.
The first place in Graph 500 was achieved with a collaboration between Riken, Kyushu University, Fixstars Corporation and Fujitsu, where it obtained a score of 102,955 gigaTEPS.
The heart of this machinery is the Fujitsu A64FX microprocessor. This CPU is based on ARM version 8.2A architecture and adopts scalable vector extensions for supercomputers.
Fugaku claimed to be about 100 times more powerful than the K computer (i.e. a performance target of 1 exaFLOPS) and to have a high level of practicability in the world.
Regarding software uses both Linux and the McKernel lightweight kernel operating simultaneously, Side to side. The infrastructure on which both systems run is called the Heterogeneous Core Interface.
06/10/2020 Fugaku Japanese supercomputer RIKEN / FUJITSU RESEARCH AND TECHNOLOGY POLICY
The main high-performance simulations run on McKernel, with Linux available for all other supported services.
On May 23, 2019, RIKEN announced that the supercomputer would be called Fugaku, Named after an alternative name for Mount Fuji. A few months later, the logo was unveiled, depicting Mount Fuji, symbolizing “Fugaku high performance”.