The 56th edition of the ranking of the highest performing systems in the world has a record few newcomers
The next, 56th edition of the TOP500 list has been published, in which the most high-performance computers in the world are ordered by speed.
Japanese supercomputer Fugaku strengthens its leadership in the TOP500 list
Like six months ago , the list is topped by the Japanese supercomputer Fugaku based on Fujitsu A64FX SoCs based on Arm architecture. Moreover, an increase in the total number of cores from 7,299,072 to 7,630,848 made it possible to increase the result in the High Performance Linpack (HPL) test from 415.5 to 442 petaflops. This is three times the performance of the IBM Summit system on Power9 processors and Nvidia Tesla V100 accelerators, which is in second place with a score of 148.8 petaflops, unchanged in six months. The third place and also with the previous result of 94.6 petaflops continues to be occupied by the American supercomputer IBM Sierra, which is close in architecture to the IBM Summit.
The fourth place belongs to the Chinese supercomputer Sunway TaihuLight on Sunway SW26010 processors. As in the previous edition of the list, this system showed the result of 93 petaflops.
In fifth place is the American supercomputer Selene based on Nvidia DGX A100, installed by Nvidia. In June, he was seventh, but doubling the configuration allowed for an increase in performance from 27.58 petaflops to 63.4 petaflops.
The Chinese supercomputer Tianhe-2A on Intel Xeon processors and Matrix-2000 coprocessors dropped from fifth to sixth position, as its performance remained the same – 61.4 petaflops.
New in the top ten is the JUWELS Booster Module supercomputer, installed in Germany and resembling the aforementioned Selene system. The result of 44.1 petaflops was enough to take the seventh line of the list. Note that this supercomputer is part of a modular system architecture, the second module of which, based on Intel Xeon processors, took position 44 in the TOP500.
Eighth place is taken by the Italian HPC5 system based on Intel Xeon Gold and Nvidia Tesla V100. Its result is 33.5 petaflops.
This is followed by the American system Frontera on Intel Platinum Xeon processors, showing a result of 23.5 petaflops.
The Dammam-7 supercomputer from Saudi Arabia closes the top ten. It is formed by HPE Cray CS-Storm systems on Intel Gold Xeon processors and NVIDIA Tesla V100 GPUs. In the HPL test, its result is 22.4 petaflops.
The best of the two Russian systems in the TOP500 list is the Christofari supercomputer based on Intel Xeon Platinum, Nvidia DGX-2 and Tesla V100 processors. His result of 6.67 petaflops corresponds to 40th place. The second system – “Lomonosov-2” on a similar hardware base – takes 158th place.
The compilers of the rating note that the 56th edition has the least number of new systems in its entire history, counted from 1993. The total performance of all systems on the list for six months increased from 2.22 to 2.43 exaflops, and the threshold for entering the list rose from 1.23 to 1.32 exaflops.
