54th edition 500 most high-performance computers in the world. In the new issue, the top ten has not changed. The cluster is in first place in the ranking by IBM at the Oak Ridge National Laboratory (USA). The cluster runs Red Hat Enterprise Linux, includes 2.4 million processor cores (using 22-core IBM Power9 22C 3.07GHz CPUs and NVIDIA Tesla V100 accelerators), which provide 148 petaflops of performance.
The American cluster takes second place , installed in the Livermore National Laboratory by IBM on the basis of a platform similar to Summit and demonstrating performance at the level of 94 petaflops (about 1.5 million cores).
In third place is the Chinese cluster , operating in the national supercomputing center of China, which includes more than 10 million computing cores and shows a performance of 93 petaflops. Despite similar performance, the Sierra cluster consumes half as much power as the Sunway TaihuLight.
In fourth place is the Chinese Tianhe-2A cluster, which includes almost 5 million cores and demonstrates a performance of 61 petaflops.
The cluster takes fifth place in the ranking , produced by Dell for the Texas Computer Center. The cluster runs CentOS Linux 7 and includes more than 448 thousand cores based on a Xeon Platinum 8280 28C 2.7GHz. The total size of RAM is 1.5 PB, and performance reaches 23 petaflops, which is 6 times less than the leader in the rating.
The most interesting trends:
- The new Russian cluster took 29th place in the ranking , launched by Sberbank. The cluster is built on the NVIDIA DGX-2 platform, uses a Xeon Platinum 8168 24C 2.7GHz CPU and has 99600 computing cores. SberCloud performance is 6.6 petaflops. The operating system is Ubuntu 18.04.01.
The second domestic cluster, Lomonosov 2, moved from 6rd to 93th place in the ranking over 107 months. Cluster in dropped from 365 to 465 place. The number of domestic clusters in the ranking over six months increased from 2 to 3 (in 2017 there were 5 , and in 2012 - 12);
- Distribution by number of supercomputers in different countries:
- China: 228 (219 six months ago). In total, Chinese clusters generate 31.9% of all productivity (six months ago - 29.9%);
- USA: 117 (116). Total productivity is estimated at 37.8% (a year ago - 38.4%);
- Japan: 29 (29);
- France: 18 (19);
- Germany: 16 (14);
- Netherlands: 15 (13);
- Ireland: 14 (13);
- UK: 11 (18);
- Canada 9(8);
- Italy: 5 (5);
- Singapore 4 (5);
- Australia, South Korea, Saudi Arabia, Brazil, Russia: 3;
- In the ranking of operating systems used in supercomputers, only Linux has remained for two and a half years;
- Distribution by Linux distributions (in brackets - 6 months ago):
- 49.6% (48.8%) do not detail the distribution,
- 26.4% (27.8%) use CentOS,
- 6.8% (7.6%) - Cray Linux,
- 4.8% (4.8%) - RHEL,
- 3% (3%) - SUSE,
- 2% (1.6%) - Ubuntu;
- 0.4% (0.4%) - Scientific Linux
- The minimum performance threshold for entering the Top500 has grown from 6 to 1022 teraflops in 1142 months (last year, only 272 clusters showed performance over petaflops, two years ago - 138, three years ago - 94). For the Top100, the entry threshold has increased from 2395 to 2570 teraflops;
- The total performance of all systems in the rating increased over the year from 1.559 to 1.650 exaflops (three years ago it was 566 petaflops). The system that closes the current ranking was in 397th place in the last issue, and 311th in the year before;
- The general distribution of the number of supercomputers in different parts of the world is as follows:
274 supercomputer is located in Asia (267 - six months ago),
129 in America (127) and 94 in Europe (98), 3 in Oceania; - As a processor base, Intel CPUs are in the lead - 94% (six months ago it was 95.6%), in second place is IBM Power - 2.8% (from 2.6%), in third place is AMD - 0.6% (0.4%), in fourth place is SPARC64 - 0.6% (0.8%);
- 35.6% (six months ago 33.2%) of all used processors have 20 cores, 13.8% (16.8%) - 16 cores, 11.2% (11.2%) - 12 cores, 11% (11.2%) - 18 cores, 7.8% (7% ) - 14 cores;
- 144 out of 500 systems (six months ago - 133) additionally use accelerators or coprocessors, while 135 systems use NVIDIA chips (six months ago there were 125), 5 - Intel Xeon Phi (there were 5), 1 - PEZY (1), 1 uses hybrid solutions (there was 1), 1 uses Matrix-2000 (1), 1 AMD Vega GPU (XNUMX months ago AMD accelerators were not used);
- Among cluster manufacturers, Lenovo took first place - 34.8% (a year ago 34.6%), second place
Sugon took the lead with 14.2% (12.6%), Inspur took third place - 13.2% (14.2%), fourth place is occupied by Hewlett-Packard - 7% (8%) and 7% (7.8%), followed by Atos - 4.6% , IBM 2.6 (2.4%), Fujitsu 2.6% (2.6%), Penguin Computing - 2.2% (1.8%), Dell EMC 2.2% (3%), Huawei 2% (1.4%), NVIDIA 1.2%. Five years ago, the distribution among manufacturers was as follows: Hewlett-Packard 36%, IBM 35%, Cray 10.2% and SGI 3.8%; - Ethernet is used to connect nodes in 52% of clusters, InfiniBand is used in 28% of clusters, and Omnipath is used in 10%.
When looking at overall performance, InfiniBand-based systems account for 40% of the Top500's overall performance, while Ethernet accounts for 29%.
At the same time, a new release of the alternative rating of cluster systems is available , focused on evaluating the performance of supercomputer platforms associated with the simulation of physical processes and tasks for processing large amounts of data, typical for such systems. Rating separately more and merged with the Top500 as energy efficiency is now in the main Top500 rating (based on the ratio of LINPACK FLOPS to power consumption in watts).
Source: opennet.ru