ProHoster > Blog > internet news > Published 60 edition of the rating of the most high-performance supercomputers

Published 60 edition of the rating of the most high-performance supercomputers

The 60th edition of the ranking of the 500 most high-performance computers in the world has been published. In the new edition, there is only one change in the top ten - the Leonardo cluster, located in the Italian scientific research center CINECA, took 4th place. The cluster includes almost 1.5 million processor cores (CPU Xeon Platinum 8358 32C 2.6GHz) and provides performance of 255.75 petaflops with a power consumption of 5610 kilowatts.

The top three, as well as 6 months ago, includes clusters:

  • Frontier - Housed at the U.S. Department of Energy's Oak Ridge National Laboratory. The cluster has almost 9 million processor cores (AMD EPYC 64C 2GHz CPU, AMD Instinct MI250X accelerator) and provides performance of 1.102 exaflops, which is almost three times more than the second-place cluster (while Frontier's power consumption is 30% lower).
  • Fugaku - hosted by RIKEN Institute of Physical and Chemical Research (Japan). The cluster is built using ARM processors (158976 nodes based on Fujitsu A64FX SoC, equipped with a 48-core Armv8.2-A SVE 2.2GHz CPU). Fugaku delivers 442 petaflops of performance.
  • LUMI is hosted at the European Supercomputing Center (EuroHPC) in Finland and provides 151 petaflops of performance. The cluster is built on the same HPE Cray EX235a platform as the leader of the rating, but includes 1.1 million processor cores (AMD EPYC 64C 2GHz, AMD Instinct MI250X accelerator, Slingshot-11 network).

As for domestic supercomputers, the Chervonenkis, Galushkin and Lyapunov clusters created by Yandex dropped from 22, 40 and 43 places to 25, 44 and 47 places. These clusters are designed to solve machine learning problems and provide performance of 21.5, 16 and 12.8 petaflops, respectively. The clusters run Ubuntu 16.04 and are equipped with AMD EPYC 7xxx processors and NVIDIA A100 GPUs: the Chervonenkis cluster has 199 nodes (193 thousand AMD EPYC 7702 64C 2GH cores and 1592 NVIDIA A100 80G GPUs), Galushkin - 136 nodes (134 thousand AMD EPYC 7702 cores 64C 2GH and 1088 NVIDIA A100 80G GPUs), Lyapunov - 137 nodes (130 thousand cores AMD EPYC 7662 64C 2GHz and 1096 NVIDIA A100 40G GPUs).

The Christofari Neo cluster deployed by Sberbank dropped from 46th to 50th place. Christofari Neo runs NVIDIA DGX OS 5 (Ubuntu edition) and demonstrates performance of 11.9 petaflops. The cluster has more than 98 thousand computing cores based on an AMD EPYC 7742 64C 2.25GHz CPU and comes with an NVIDIA A100 80GB GPU. The second cluster of Sberbank (Christofari) moved from 80th to 87th place in the ranking over six months.

Two more domestic clusters also remain in the ranking: Lomonosov 2 - shifted from 262 to 290 place (in 2015, the Lomonosov 2 cluster occupied 31 place, and its predecessor Lomonosov in 2011 - 13 place) and MTS GROM - shifted from 318 to 352 place . Thus, the number of domestic clusters in the ranking has not changed and, as six months ago, is 7 systems (for comparison, in 2020 there were 2 domestic systems in the ranking, in 2017 - 5, and in 2012 - 12).

The most interesting trends:

  • Distribution by number of supercomputers in different countries:
    • China: 162 (173 six months ago). In total, Chinese clusters generate 10% of all productivity (six months ago - 12%);
    • USA: 127 (127). The total performance is estimated at 43.6% of the entire rating performance (six months ago - 47.3%);
    • Germany: 34 (31). Total productivity - 4.5%;
    • Japan: 31(34). Total productivity - 12.8%;
    • France: 24 (22). Total productivity - 3.6%;
    • UK: 15 (12);
    • Canada 10(14);
    • Netherlands: 8 (6);
    • South Korea 8 (6)
    • Brazil 8 (6);
    • Russia 7 (7);
    • Italy: 7 (6);
    • Saudi Arabia 6 (6);
    • Sweden 6 (5);
    • Australia 5 (5);
    • Ireland 5;
    • Poland 5 (5);
    • Switzerland 4 (4);
    • Finland: 3 (4).
    • Singapore: 3;
    • India: 3;
    • Poland: 3;
    • Norway: 3.
  • In the ranking of operating systems used in supercomputers, only Linux has remained for six years;
  • Distribution by Linux distributions (in brackets - 6 months ago):
    • 47.8% (47.8%) do not detail the distribution;
    • 17.2% (18.2%) use CentOS;
    • 9.6% (8.8%) - RHEL;
    • 9% (8%) - Cray Linux;
    • 5.4% (5.2%) - Ubuntu;
    • 3.8% (3.8%) - SUSE;
    • 0.8% (0.8%) - Alma Linux;
    • 0.8% (0.8%) - Rocky Linux;
    • 0.2% (0.2%) - Scientific Linux.
  • The minimum performance threshold for entering the Top500 for 6 months was 1.73 petaflops (six months ago - 1.65 petaflops). Four years ago, only 272 clusters showed performance of more than a petaflop, five years ago - 138, six years ago - 94). For Top100, the entry threshold increased from 5.39 to 9.22 petaflops;
  • The total performance of all systems in the rating over 6 months increased from 4.4 to 4.8 exaflops (three years ago it was 1.650 exaflops, and five years ago - 749 petaflops). The system that closes the current ranking was in 458th place in the last issue;
  • The general distribution of the number of supercomputers in different parts of the world is as follows: 218 supercomputers are located in Asia (229 six months ago), 137 in North America (141) and 131 in Europe (118), 8 in South America (6), 5 in Oceania (5) and 1 in Africa (1);
  • As a processor base, Intel CPUs are in the lead - 75.6% (six months ago it was 77.4%), AMD is in second place with 20.2% (18.8%), and IBM Power is in third place with 1.4% (was 1.4%).
  • 22.2% (six months ago 20%) of all used processors have 24 cores, 15.8% (15%) - 64 cores, 14.2% (19.2%) - 20 cores, 8.4% (8.8%) - 16 cores, 7.6% (8.2% ) - 18 cores, 6% - 28 cores, 5% (5.4%) - 12 cores.
  • 177 out of 500 systems (six months ago - 167) additionally use accelerators or coprocessors, while 161 systems use NVIDIA chips, 9 - AMD, 2 - Intel Xeon Phi (from 5), 1 - PEZY (1), 1 - MN-Core, 1 - Matrix-2000;
  • Among cluster manufacturers, Lenovo took first place with 32% (six months ago 32%), Hewlett-Packard Enterprise took second place with 20.2% (19.2%), Inspur took third place with 10% (10%), followed by Atos - 8.6% (8.4%), Sugon 6.8% (7.2%), Dell EMC 3.6% (3.4%), NVIDIA 2.8% (2.8%), NEC 2.4% (2%), Fujitsu 2% (2.6%), MEGWARE 1.2%, Penguin Computing - 1.2% (1.2%), IBM 1.2% (1.2%), Huawei 0.4% (1.4%).
  • To connect nodes in 46.6% (six months ago 45.4%) of clusters, Ethernet is used, InfiniBand is used in 38.8% (39.2%) of clusters, Omnipath - 7.2% (7.8%). If we consider the total performance, then InfiniBand-based systems cover 33.6% (32.4%) of all Top500 performance, and Ethernet - 46.2% (45.1%).

In the near future, a new release of the Graph 500 alternative rating of cluster systems is expected to be published, focused on evaluating the performance of supercomputer platforms associated with simulating physical processes and tasks for processing large amounts of data inherent in such systems. The Green500, HPCG (High-Performance Conjugate Gradient) and HPL-AI rankings are combined with the Top500 and reflected in the main Top500 ranking.

Source: opennet.ru

Add a comment