US fights back with first exascale supercomputer

US fights back with first exascale supercomputer

Technology News |
By Nick Flaherty

The US is hosting the world’s first true exascale supercomputer, which also leads as the most energy efficient.

The Frontier system at Oak Ridge National Laboratory (ORNL) in the US is based on the latest HPE Cray EX235a architecture and equipped with AMD EPYC 64C 2GHz processors with 8,730,112 total cores, a power efficiency rating of 52.23 gigaflops/watt, and gigabit ethernet for data transfer.

A recent boost in performance for the Frontier system has allowed the machine to surpass the 1 exaflop barrier. With an exact HPL score of 1.102 Exaflop/s, Frontier is not only the most powerful supercomputer to ever exist – it’s also the first true exascale machine.

Related supercomputer articles

The top position was previously held for two years straight by the ARM-based Fugaku system at the RIKEN Center for Computational Science (R-CCS) in Kobe, Japan. Sticking with its previous HPL benchmark score of 442 PFlop/s, Fugaku has now dropped to No. 2. Fugaku’s theoretical peak is also above 1 exaflop, so it is technically an exascale machine, but Frontier is the only system able to demonstrate this on the HPL benchmark test.

Another change within the top ten is the introduction of the LUMI system at EUROHPC/CSC in Finland. Now occupying the No. 3 spot, this new system has 1,110,144 cores and has a HPL benchmark of nearly 152 PFlop/s. LUMI is also noteworthy in that it is the largest system in Europe.

The Adastra system at GENCI-CINES in France also joins at number 10. It achieved an HPL benchmark score of 46.1 Pflop/s and is the second most powerful machine in Europe, behind LUMI.

Frontier is currently being integrated and tested at the ORNL in Tennessee, USA, where it will be operated by the Department of Energy (DOE).

Summit, an IBM-built system at ORNL in Tennessee, USA, is now listed at the No. 4 spot worldwide with a performance of 148.8 Pflop/s on the HPL benchmark which is used to rank the TOP500 list. Summit has 4,356 nodes, each housing two Power9 CPUs with 22 cores 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, CA, USA, is at No. 5. Its architecture is very similar to the #4 systems Summit. It is built with 4,320 nodes with two Power9 CPUs and four NVIDIA Tesla V100 GPUs. Sierra achieved 94.6 Pflop/s.

Sunway TaihuLight is a system developed by China’s National Research Center of Parallel Computer Engineering & Technology (NRCPC) and installed at the National Supercomputing Center in Wuxi, China’s Jiangsu province, is listed at the No. 6 position with 93 Pflop/s.

Perlmutter at No. 7 is based on the HPE Cray “Shasta” platform, and a heterogeneous system with AMD EPYC based nodes and 1536 Nvidia A100 accelerated nodes. Perlmutter achieved 64.6 Pflop/s

Now at No. 8, Selene is an Nvidia DGX A100 SuperPOD installed inhouse. This uses an AMD EPYC processor with Nvidia A100 for acceleration and a Mellanox HDR InfiniBand as network and achieved 63.4 Pflop/s.

Tianhe-2A (Milky Way-2A), a system developed by China’s National University of Defense Technology (NUDT) and deployed at the National Supercomputer Center in Guangzhou, China is now listed as the No. 9 system with 61.4 Pflop/s.

The Adastra system installed at GENCI-CINES is new to the list at No. 10. It is the third new HPE Cray EX system and the second fastest system in Europe. It achieved 46.1 Pflop/s.

Across the top 500 systems, China stayed constant with 173 systems, while the United States dropped from 150 systems to 126.

Supercomputer interconnect

On interconnect technologies, Ethernet still won out with 226 machines, but this was a drop from 240 machines on the last list. Infiniband accounted for 196 on the current list, which increased from 180 previously. Omnipath stayed consistent with 40 machines on the list, while custom interconnects dropped from 34 connections in the last list to 32 connections on the current one. Like the last list, there were only 6 systems with proprietary networks.

The leading energy efficient system in the GREEN500 is the Frontier Test & Development System (TDS) at ORNL. With 120,832 total cores and an HPL benchmark of 19.2 PFlop/s, the Frontier TDS machine is basically just one rack identical to the actual Frontier system and has a power efficiency of 62.8 gigaflops/watt.

The whole Frontier system is at number 2, with a power efficiency at 55.23 gigaflops/watt, while the No. 3 spot was taken by the LUMI system with a power efficiency rating of 51.63 gigaflops/watt.

LUMI is part of a wider trend in HPC that is proving that power doesn’t have to be sacrificed in the name of efficiency. The No. 4 spot was held by Adastra,  the HPE Cray EX system at GENCI-CINES that took the No. 10 spot on the TOP500. It’s also the second fastest in Europe and has an efficiency rating of 50.03 gigaflops/watt.

Last year’s winner of the GREEN500 has moved down to the No. 5 spot, as the MN-3 system from Preferred Networks in Japan received a power efficiency rating of 40.90 gigaflops/watt. This is an improvement over the system’s previous score of 39.38 gigaflops/watt.

The HPL-AI benchmark seeks to highlight the convergence of HPC, and artificial intelligence (AI) workloads based on machine learning and deep learning by solving a system of linear equations using novel, mixed-precision algorithms that exploit modern hardware.

The Frontier system has demonstrated an HPL-AI benchmark of 6.86 exaflops, which more than triples Fugaku’s previous winner score of around 2 exaflops.

Related articles

Other articles on eeNews Europe



Linked Articles