63rd Edition of Supercomputer Rankings Released

by

Published 63rd issue rating 500 most high-performance computers in the world. In the 63rd edition of the rating, the clusters that occupied the first five places in the previous rating retained their positions:

The first place is occupied by the Frontier cluster, located at the Oak Ridge National Laboratory of the US Department of Energy. The cluster has 8.7 million processor cores (AMD EPYC 64C 2GHz CPU, AMD Instinct MI250X accelerator) and provides a performance of 1.206 exaflops. Used as an operating system HPE CrayOS (SUSE Linux Enterprise Server 15 edition).

Advertisement

The second place is occupied by the Aurora cluster, deployed at the Argonne National Laboratory of the US Department of Energy. Compared to the previous edition of the rating, the cluster increased the number of processor cores (CPU Xeon CPU Max 9470 52C 2.4GHz, Intel Data Center GPU Max accelerator) from 4.8 to 9.2 million. Performance increased from 585 petaflops to 1.012 exaflops. Aurora uses SUSE Linux Enterprise Server 15 SP4 as its operating system.

The cluster takes third place in the ranking Eagle launched this year by Microsoft for the Azure cloud. The cluster contains 2 million processor cores (CPU Xeon Platinum 8480C 48C 2GHz) and demonstrates peak performance of 561 petaflops. The cluster software is based on Ubuntu 22.04.

In fourth place is the cluster Fugaku, located at the 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 and runs Red Hat Enterprise Linux.

Fifth place is occupied by the LUMI cluster, located at the European Supercomputing Center (EuroHPC) in Finland and providing performance of 379 petaflops. The cluster is built on the same HPE Cray EX235a platform as the leader of the rating, but includes 2.2 million processor cores (AMD EPYC 64C 2GHz, AMD Instinct MI250X accelerator, Slingshot-11 network). HPE Cray OS is used as the operating system.

Advertisement

The new cluster took sixth place Alps, launched at the Swiss National Supercomputing Center. The cluster has 1,305,600 NVIDIA Grace 72C 3.1GHz processor cores and provides a performance of 270 petaflops.

As for domestic supercomputers, the clusters created by Yandex Chervonenkis, Galushkin And Lyapunov dropped from 36, 58 and 64 places to 42, 69 and 79 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 7 cores 702 64C 2GH and 1088 GPU NVIDIA A100 80G), Lyapunov – 137 nodes (130 thousand cores AMD EPYC 7662 64C 2GHz and 1096 GPU NVIDIA A100 40G).

The Christofari Neo cluster deployed by Sberbank dropped from 67th to 83rd place. Christofari Neo runs NVIDIA DGX OS 5 (Ubuntu edition) and demonstrates performance of 11.95 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 119th to 142nd place in the ranking over six months.

Two more domestic clusters also remain in the ranking: Lomonosov 2 – moved from 370 to 406 place (in 2015, the Lomonosov 2 cluster took 31 place, and its predecessor Lomonosov in 2011 – 13 place) and MTS GROM – moved from 433 to 472 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:
    • USA: 171 (161 – six months ago). The total productivity is estimated at 34.2% of the entire rating productivity (six months ago – 32.8%);
    • China: 80 (104). In total, Chinese clusters generate 16% of all productivity (six months ago – 20.8%);
    • Germany: 40 (36). Total productivity – 8% (7.2%);
    • Japan: 29 (32). Total productivity – 5.8% (6.4%);
    • France: 24 (23). Total productivity – 4.8% (4.6%);
    • UK: 16 (15);
    • South Korea 13 (12);
    • Italy: 11 (12);
    • Canada 10 (10);
    • Netherlands: 9 (10);
    • Brazil 8 (9);
    • Saudi Arabia 8 (7);
    • Poland: 8 (4);
    • Sweden 7 (6);
    • Russia 7 (7);
    • Taiwan: 6 (5);
    • Australia 5 (6);
    • Norway: 5 (4);
    • Switzerland 5 (3);
    • Ireland 4 (4);
    • India: 4 (4);
    • Finland: 3 (3);
    • Singapore: 3 (3);
    • Czech Republic: 3;
    • Spain: 3 (3).
  • Since November 2017, only Linux remains in the ranking of operating systems used in supercomputers;
  • Distribution by Linux distributions (in brackets – 6 months ago):
    • 42.4% (44.6%) use Linux-based systems, but do not specify the distribution;
    • 16.8% (12.6%) – RHEL;
    • 9.4% (11%) CentOS;
    • 9.2% (9.6%) – Cray Linux;
    • 8.4% (7.8%) – Ubuntu;
    • 4.4% (4.4%) – SUSE;
    • 3% (2%) – Rocky Linux;
    • 1.2% (1%) – Alma Linux;
    • 0.2% (0.2%) – Amazon Linux;
    • 0.2% (0.2%) – Scientific Linux.
  • The minimum performance threshold for entering the Top500 over 6 months was 2.13 petaflops (six months ago – 2.02 petaflops). Six years ago, only 272 clusters showed performance of more than a petaflop, seven years ago – 138). For Top100, the entry threshold increased from 7.89 to 9.46 petaflops, and for Top10 – from 94.64 to 121.4 petaflops.
  • The total performance of all systems in the rating over 6 months increased from 7 to 8.2 exaflops (four years ago it was 1,650 exaflops, and six years ago – 749 petaflops). The system that closes the current ranking was in 457th place in the last issue.
  • The general distribution of the number of supercomputers in different parts of the world is as follows: 181 supercomputers are in North America (171 – six months ago), 157 in Europe (143), 147 in Asia (169), 9 in South America (10), 5 in Oceania (6) and 1 in Africa (1).
  • As a processor base, Intel CPUs are in the lead – 62.8% (six months ago it was 67.9%), AMD is in second place with 31.4% (28%), and IBM Power is in third place – 1.2% (was 1.4%).
  • 17.8% (six months ago 21.4%) of all used processors have 24 cores, 22% (21%) – 64 cores, 9% (10.6%) – 20 cores, 9.4% (7.4%) – 32 cores, 5.4% (6.2% ) – 16 cores, 5.6% (6%) – 18 cores, 5.2% (5.2%) – 28 cores, 5.8% (5%) – 48 cores, 4.4% – 56 cores, 2.2% (3%) – 12 cores . The total number of processor cores in all clusters of the rating increased over six months from 106.3 million to 114.6 million.
  • 196 out of 500 systems (six months ago – 185) additionally use accelerators or coprocessors, while 142 (155) systems use NVIDIA chips, 14 (25) – AMD, 1 (2) – Intel Xeon Phi, 4 – Intel DataCenter GPU, in 1 (1) – Matrix-2000.
  • Among cluster manufacturers, Lenovo took first place – 32.6% (six months ago 33.8%), Hewlett-Packard Enterprise took second place – 22.4% (20.6%), EVIDEN took third place – 9.8% (9.6%), followed by Dell EMC 6.8% (6.4%), Inspur – 4.4% (6.8%), NVIDIA 4.4% (3.6%), NEC 2.8% (2.4%), Fujitsu 2.8% (2.4%), MEGWARE 1.4% (1.4%), Microsoft Azure – 1.4% (1.2%), Penguin Computing – 1.4% (1.2%), Sugon 1% (1.8%), IBM 1% (1.2%), Huawei 0.4% (0.4%), Intel 0.4%.
  • InfiniBand is used to connect nodes in 47.8% (six months ago 43.8%) of clusters, Ethernet is used in 39% (41.8%) of clusters, Omnipath – 6.4% (6.6%). Looking at overall performance, InfiniBand-based systems account for 39.2% (41.4%) of the Top500's overall performance, while Ethernet accounts for 48.5% (44%).

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

Thanks for reading:

Advertisement