World’s first exascale supercomputer is the world’s fastest

The first true exascale supercomputer, Frontier, is now the fastest in the world, toppling Fugaku, which held the title for the past two years, according to the latest TOPP500 list of the best performing supercomputers.

An exascale computer is one that can perform 10¹⁸ (one quintillion) floating point operations per second (1 exaFLOPS), and Frontier, installed at Oak Ridge National Laboratory in Tennessee, weighed in at 1.12 exaFLOPS.

Frontier also captured the title of most energy efficient supercomputer, generating 62.68 GFLOP per watt.

Frontier’s speed bumps down Fugaku at the RIKEN Center for Computational Science (R-CCS) in Kobe, Japan, from number 1 on the TOPP500 list last fall to number 2 now. Fugaku scored 442 peta FLOPS (PFLOPS) on the  High Performance Linpack (HPL) benchmark, which measures how well systems solve a dense system of linear equations.

TOPP500 notes that the theoretical upper peak of Fugaku is more than 1exaFLOP, but it did not demonstrate that speed in the HPL benchmark.

This is Frontier’s first listing in the top 10, and two other supercomputers, both in Europe also broke into the list. The LUMI system at EUROHPC/CSC in Finland weighed in at number 3, and the Adastra system at GENCI-CINES in France came in at number 10.

Here is the top 10 list.

#1—Frontier

An HPE Cray EX system run by the US Department of Energy, Frontier incorporates 3rd Gen AMD EPYC™ CPUs representing 8,730,112 cores that have been optimized for high-performance computing (HPC) and AI with AMD Instinct™ 250X accelerators and Slingshot-11 interconnects.

#2—Fugaku

Fugaku has 7,630,848 cores and was built by Fujitsu.

#3—LUMI

LUMI is an HPE Cray EX system at the EuroHPC center at CSC in Kajaani, Finland with a performance of 151.9 PFLOPS.

#4—Summit

An IBM system at Oak Ridge National Laboaratory in Tennessee, Summit scored 148.8 PFLOPS on the HPL benchmark. It has  4,356 nodes, each with two Power9 CPUs with 22 cores and six NVIDIA Tesla V100 GPUs, each with 80 streaming multiprocessors (SM). The nodes are linked by a Mellanox dual-rail EDR InfiniBand network.

#5—Sierra

Similar in architecture to Summit, Sierra reached 94.6 PFLOPS. It has 4,320 nodes with two Power9 CPUs and four NVIDIA Tesla V100 GPUs and is housed at the Lawrence Livermore National Laboratory, in California. CA, USA

#6—Sunway TaihuLight

Sunway TaihuLight is a machine developed by National Research Center of Parallel Computer Engineering & Technology (NRCPC) in China and is installed in the city of Wuxi. It reached 93PFLOPS on the HPL benchmark.

#7—Perlmutter

The perlmutter system is based on the HPE Cray Shasta platform, and is a heterogeneous system with both AMD EPYC-based nodes and 1536 Nvidia A100-accelerated nodes. It achieved 64.6 Pflop/s.

#8—Selene

Selene is an Nvidia DGX A100 SuperPOD based on an AMD EPYC processor with Nvidia A100 for acceleration and a Mellanox HDR InfiniBand as a network. It achieved 63.4 Pflop/s and is installed in-house at Nvidia facilities in the US.

#9—Tianhe-2A (Milky Way-2A)

Powered by Intel Xeon CPUs and NUDT’s Matrix-2000 DSP accelerators, it has 4,981,760 cores in the system to achieve 61.4 Pflop/s. It was developed by China’s National University of Defense Technology (NUDT) and is deployed at the National Supercomputer Center in Guangzhou, China.

#10—Adastra

Adestra, installed at GENCI-CINES, is new another HPE Cray EX system and the second fastest system in Europe. It achieved 46.1 Pflop/s.