The machine triples in power the second supercomputer on the list. Fugaku is the first of the Top 500.
There is a list called the Top500 that is launched twice a year taking advantage of the annual SC 17 Supercomputing Show, a fair that revolves around the world’s supercomputers – gigantic machines that can occupy an entire building -, new models and the most powerful.
Compiled by researchers from the Berkeley Lawrence National Laboratory, the University of Tennessee, and the cloud computing company Prometeus, the Top500 ranks the world’s supercomputers, placing the fastest performing mathematical calculations in the top positions.
Fugaku, the beast of Japan
For years, the United States and China have struggled to see who conquered the top spot in the Top 500. But today, neither of them surpasses the newcomer, Fugaku. Created by Fujitsu and the Japanese Riken Research Institute. Fugaku is a supercomputer located in Japan that has been recognized as the most powerful in the world, with 415.53 petaflops, a unit equivalent to one billion trillion floating point operations per second.
In this way, Japan with Fugaku has managed to unseat from the first position the supercomputer work of the IBM company, located in the Oak Ridge National Laboratory in Tennessee (United States), with 148.8 petaflops. Fugaku, which is located at the Riken Computer Science Center in Kobe (Japan), has a computing efficiency ratio of 80.87%, and has 360 ‘racks’ and 138,240 nodes with which in other measurements it has been able to achieve a maximum result of 13,400 teraflops.
More than triple the power
Directly compared to the Oak Ridge supercomputer, we see that Fugaku has up to three times the power. A madness of figures that lead him to lead the list of the most powerful through the use of processors with ARM architectures, with 48 A64FX cores and with a performance of more than one exaflop -equivalent to a thousand petaflops- in Artificial Intelligence tasks.
Fujitsu and Riken’s supercomputer has also become the first in two other rankings: HPGC, which measures the speed of data access in real applications; and HPL-AI, which measures performance in tasks such as machine learning to train Artificial Intelligence.