New TOP500 HPC List: Frontier Extends Lead with Performance Upgrade

Print Friendly, PDF & Email

After a flurry of new competitors in 2022 at the top of the TOP500 list of the world’s most powerful supercomputers, the first list of 2023 – issued here in Hamburg this morning at the ISC conference – reveals the same top 10 systems in the same order. Still, the HPC community will no doubt examine the internal details within the top 10 (there are some changes there) and the make-up of the overall list, along with updates to the GREEN500 and other sub-categories. It’s what you do.

The Frontier supercomputer at Oak Ridge National Laboratory, the world’s first certified exascale-class system (the Chinese are known to have stood up exascale systems before Frontier but performance numbers have not been submitted to the TOP500), remains no. 1. Interestingly, Frontier’s LINPACK (HPL) number increased from 1.02 ExaFLOPS last November to 1.194 ExaFLOPS now, a 17 percent increase. We assume this benchmark improvement is due to the tuning process Frontier is undergoing at Oak Ridge as the system is in the final stages of installation and user readiness.

Frontier is an HPE Cray EX235a architecture powered by AMD EPYC CPUs and Instinct MI250 accelerators. It has 8,699,904 cores and an energy efficiency rating of 52.59 Gflops/watt (ranking it no. 6 on Green500).

Remaining at no. 2 on the TOP500: the Fugaku system at the Riken Center for Computational Science in Kobe, Japan. The ARM-powered system held steady with an HPL score of 0.442 ExaFLOPS.

The LUMI system at EuroHPC/CSC in Finland – AMD-powered – entered the list last June and has an HPL score of 0.3091 ExaFLOPS. It is the largest system in Europe.

The Leonardo system at EuroHPC/CINECA in Bologna, Italy, is at the no. 4 spot. Supplied by Atos (Eviden) and based on BULL SEQUANA XH2000, it also saw upgrades that achieved an HPL score of 0.239 ExaFLOPS compared with its previous score of 0.174 ExaFLOPS.

The rest of the top 10 are older systems of varying vintages dating back to 2016. They include:

IBM-built Summit (no. 5) at Oak Ridge with a performance of 148.8 petaFLOPS. Summit has 4,356 nodes with two POWER9 CPUs of 22 cores each and six NVIDIA Tesla V100 GPUs, each with 80 streaming multiprocessors (SM). The nodes are linked together with a Mellanox dual-rail EDR InfiniBand network.

Sierra, a system at the Lawrence Livermore National Laboratory is at No. 6. With an architecture similar to Summit, it is built with 4,320 nodes with two POWER9 CPUs and four V100 GPUs and achieved 94.6 petaFLOPS.

Frontier supercomputer

Sunway TaihuLight, a system developed by China’s National Research Center of Parallel Computer Engineering & Technology and installed at the National Supercomputing Center in Wuxi, is at no. 7 with 93 petaFLOPS.

Perlmutter at no. 8 is based on the HPE Cray “Shasta” platform and a heterogeneous system with AMD EPYC-based nodes and 1,536 NVIDIA A100 GPU nodes. Perlmutter achieved 64.6 petaFLOPS.

At no. 9 is Selene, an NVIDIA DGX A100 SuperPOD installed inhouse at NVIDIA.  The system is based on AMD EPYC processors with NVIDIA A100s and a Mellanox HDR InfiniBand as network. It achieved 63.4 petaFLOPS.

Tianhe-2A (Milky Way-2A), a system developed by China’s National University of Defense Technology, is listed at no. 10 with 61.4 PetaFLOPS.

Much like the previous list, China and the United States earned most of the entries on the TOP500. The U.S. increased its lead from 126 machines to 150, while China dropped from 162 systems to 134 – though it must be noted that China has not for several years submitted exascale benchmark numbers. In terms of continents, Asia as a whole saw 192 machines on the list, North America has 160 systems, and Europe has 133.

In terms of system interconnects, ethernet is still ahead despite dropping from 233 machines to 227. Infiniband interconnects increased their presence from 194 machines to 200, and Omnipath dropped from 36 machines to 35. Custom interconnects saw a massive increase from 4 systems to 31.

Henri supercomputer


The no. 1 spot on the GREEN500 is again the Henri system at the Flatiron Institute in New York with an energy efficiency of 65.40 Gflops/Watt. Improvements to the system allowed it to jump from the no. 405 spot to no. 255 on the TOP500 with an HPL score of 2.88 petaFLOPS – an increase over last list’s score of 2.038 petaFLOPS. Henri is a Lenovo ThinkSystem SR670 with Intel Xeon Platinum CPUs and NVIDIA H100 GPUs.

No. 2 was achieved by the Frontier Test & Development System (TDS) at Oak Ridge with an energy efficiency rating of 62.20 Gflops/Watt. Frontier TDS is a single rack identical to the Frontier system and has an HPL score of 19.2 petaFLOPS.

The No. 3 spot was taken by the Adastra system in France. It is an HPE Cray EX235a system with AMD EPYC and AMD Instinct MI250X.

Frontier is at no. 6 with an energy efficiency rating of 52.59 Gflops/Watt.