It was announced the 64th edition of the 500 supercomputers ranking most powerful in the world and in this new edition Various changes have occurred and quite interesting trends, since the United States leads the first positions, changing the trend of a few years ago where Asian clusters dominated.
Initially The inauguration of the "El Capitan" cluster stands out, which is ranked first, dethroning Frontier, which held the top spot for just over two years. Located at the U.S. Department of Energy's Livermore National Laboratory, this system incorporates 11 million AMD EPYC 24C 1.8GH cores Combined with AMD Instinct MI300X accelerators, achieving a performance impressive of 1.742 exaflops and running HPE Cray OS based on SUSE Linux Enterprise Server 15.
In second place is the Frontier, the former leader of the ranking, installed at the Oak Ridge National Laboratory. This system has been updated, increasing its number of cores from 8.7 to 9 million and its power from 1.206 to 1.353 exaflops, thanks to 64 GHz AMD EPYC 2C CPUs and AMD Instinct MI250X accelerators.
The cluster Dawn, from Argonne National Laboratory, ranks third. Equipped with 9.2 million cores Xeon CPU Max 9470 52C 2,4GHz and Intel Data Center GPU Max accelerators, achieves 1,012 exaflops. This system also uses SUSE Linux Enterprise Server 15 SP4 as its operating base.
Fourth place is occupied by Eagle, implemented by Microsoft for Azure, which uses Xeon Platinum 8480C CPUs with a total of 2 million cores and a peak performance of 561 petaflops. This cluster runs Ubuntu 22.04.
As for the other positions of the Top 10:
- In the HPC6 is in fifth position, from the Italian company Eni, which has 3 million AMD EPYC 64C cores and a capacity of 477 petaflops, managed with RHEL 8.9.
- Fugaku, the system that ranks sixth In the ranking, with 7.630.848 cores, Fugaku achieved an HPL benchmark score of 442 Petaflop/s, keeping it as the fastest system in the HPCG benchmark test with an impressive performance of 16 Teraflop/s.
- In the sThe Alps system is in seventh place., which recently received a significant update at CSCS in Switzerland. Alps is an HPE Cray EX254n system integrating 72-core NVIDIA Grace processors and the NVIDIA GH200 Superchip, with a Slingshot-11 interconnect.
- The LUMI system, also an HPE Cray EX, takes eighth place and is located at CSC's EuroHPC center in Finland. With a performance of 380 Petaflops.
- In the Leonardo is in ninth place, a system installed at CINECA, Italy, also supported by EuroHPC. Leonardo is an Atos BullSequana XH2000 system, using 8358-core 32 GHz Xeon Platinum 2.6 processors and 100 GB NVIDIA A4 SXM40 accelerators.
- The tenth and final system in the Top 10 is Tuolumne, another supercomputer located at the Lawrence Livermore National Laboratory in California, USA. This system is similar to the list leader, El Capitan, with an identical architecture, achieving a performance of 208.1 Petaflops on its own.
On the part of the trends that stand out of this new edition, the Yandex clusters such as Chervonenkis, Galushkin and Lyapunov have lost positions, now ranked 60th, 86th, and 99th, respectively. They are designed for machine learning and achieve performances between 12.8 and 21.5 petaflops. These systems use Ubuntu 16.04, AMD EPYC 7xxx CPUs, and NVIDIA A100 GPUs.
The United States continues to lead with 172 systems, representing 55.2% of total productivity, ora significant increase from 34.2% six months ago. In contrast, China shows a drop to 63 systems, generating just 2.7% of total output, down from 16% previously. Germany, Japan and Italy also recorded improvements, with Japan standing out in particular, where productivity rose from 5,8% to 8%.
In terms of operating systems, Linux remains the absolute standard since 2017. Within the distributions, Those based on generic Linux dominate with 38.2%, although they show a slight decline, while Ubuntu and Rocky Linux gain ground with 10.2% and 4%, respectively.
On the other hand, the minimum performance required to enter the ranking increased to 2.31 petaflops, and the threshold for the Top100 grew to 12.8 petaflops, reflecting the acceleration of technological development in this area. Geographic distribution remains concentrated in North America, Europe and Asia, with 181 supercomputers each in North America and Europe, and 143 in Asia. Africa maintains a sole presence on the list.
As for processors, Intel remains the leader with a 61.8% share, although AMD advances by reaching 32.4%. 64 and 24 thread cores are the most usedos, while the total number of cores in classified systems has grown significantly to 128,7 million.
Accelerators and coprocessors also show an increase, with 209 systems that employ them, highlighting NVIDIA as the undisputed leader with 183 systems. Among manufacturers, Lenovo remains in first place, followed by Hewlett-Packard Enterprise, while NVIDIA and Dell EMC strengthen their positions. InfiniBand and Ethernet continue to be the main interconnect technologies, with InfiniBand dominating in the number of systems, but Ethernet leading in terms of overall performance.
Finally, if you are interested in learning more about it, you can check the details in the following link