Yesterday, the 55th edition of the TOP500 list was published – the ranking of the world’s most high-performance supercomputers. The first place in it broke the system from Japan.
The Fugaku supercomputer showed 415.5 petaflops in the High Performance Linpack (HPL) test, 2.8 times ahead of the IBM Summit system that ranks second in the US. It is noteworthy that Fugaku uses 48-core single-chip Fujitsu A64FX SoC systems based on Arm architecture. In single or lower precision operations that are often used in AI tasks, including machine learning, Fugaku peak performance exceeds 1 exaflops.
The IBM Summit system mentioned above uses 22-core Power9 processors and Nvidia Tesla V100 accelerators. This supercomputer demonstrates a performance of 148.8 petaflops.
In third place is also an American supercomputer – IBM Serra, which is similar in architecture to IBM Summit. Its result is equal to 94.6 petaflops.
The Chinese Sunway TaihuLight system on 260-core Sunway SW26010 processors demonstrates the result of 93 petaflops.
The fifth position of the list with a result of 61.4 petaflops is occupied by the Chinese Tianhe-2A supercomputer based on Intel Xeon processors and Matrix-2000 coprocessors.
This is followed by the Italian HPC5 system based on Intel Xeon Gold and the Nvidia Tesla V100, built by Dell. Its result in the HPL test is 33.5 petaflops.
Seventh place belongs to the American Selene system on AMD Epyc processors and Nvidia Ampere A100 accelerators. Its result is 27.58 petaflops.
In America, there is the eighth system from the TOP500 list. This is a Frontera supercomputer based on Intel Xeon processors, gaining 23.5 petaflops in the test.
Interestingly, in ninth place is the second Italian system of the top ten. This is a Marconi-100 supercomputer based on IBM Power9 processors and Nvidia V100 accelerators. The performance of the Marconi-100 is 21.6 petaflops.
Closes the top ten installed in Switzerland Piz Daint system with a performance of 19.6 petaflops. This is a Cray XC50 supercomputer based on Intel Xeon processors and Nvidia P100 accelerators.
The best Russian system scores 6.67 petaflops in the HPL test, ranking 36th. This is a Christofari supercomputer based on the Xeon Platinum, Nvidia DGX-2 and Tesla V100.