• United States
Executive Editor

Fugaku still reigns as the world’s fastest supercomputer

Jun 28, 20214 mins
Data Center

The June 2021 TOP500 list of the fastest supercomputers has Fugaku ranked at number one, attaining the top slot for the third time and still three times faster than any other.

Credit: Shutterstock

Fugaku, the supercomputer built by Fujitsu, remains at number one in the TOP500 list of the fastest supercomputers in the world, where it is still three times faster than the nearest competition.

The contest for the fastest remains tight, with only one new entry into the top 10 on the latest list—Perlmutter, at the National Energy Research Scientific Computing (NERSC) Center at Lawrence Berkeley National Laboratory, which is part of the US Department of Energy. It joins the list at number five and bumps down numbers six through 10 from the previous list published in November 2020.(A system called Dammam-7 dropped off the top 10.)

The list is updated every June and November and has tracked the development of supercomputer performance and architecture since 1993. The speeds are based on the  High Performance Linpack (HPL) benchmark, which measures how well systems solve a dense system of linear equations

Among the top 10 in the latest ranking, five are in the US, three are in China and there is one each in Germany and Italy.

TOP500’s 10 fastest supercomputers, June 2021

Number 1: Fugaku

Fugaku – At 442 Pflop/s, the system hit the same speed as in November. It was built by Fujitsu, has 7,630,848 cores, and resides at the RIKEN Center for Computational Science (R-CCS) in Kobe, Japan.

Number 2: Summit

Summit – The fastest system in the US, Summit, scored 148.8 Pflop/s, the same as last November. It has 4,356 nodes, with two Power9 CPUs each with 22 cores each and six Nvidia Tesla V100 GPUs. The GPUs each have 80 streaming multiprocessors (SM). The nodes are linked together with a Mellanox dual-rail EDR InfiniBand network. Built by IBM, the system resides at the Oak Ridge National Laboratory (ORNL) in Tennessee.

Number 3: Sierra

Sierra – With an architecture very similar to Summit’s, this system achieved 94.6 Pflop/s, the same as last time. . It is built with 4,320 nodes with two Power9 CPUs and four Nvidia Tesla V100 GPUs, and is installed at the Lawrence Livermore National Laboratory in California.

Number 4: Sunway TaihuLight

Sunway TaihuLight – This system retains its ranking with the same score it achieved with the last ranking, 93 Pflop/s. Sunway TaihuLight is powered by Sunway SW26010 processors with 10,649,600 cores. It was developed by the National Research Center of Parallel Computer Engineering & Technology (NRCPC) in China and is installed at the National Supercomputing Center in Wuxi, China.

Number 5: Perlmutter

Perlmutter – The lone new entry to the top 10 list, this system is based on the HPE Cray Shasta platform, and a heterogeneous system with both AMD EPYC-based nodes and 1536 Nvidia A100-accelerated nodes. It achieved 64.6 Pflop/s.

Number 6: Selene

Selene – Bumped down from the number five slot last time, this system 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.

Number 7: Tianhe-2A

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.

Number 8: JUWELS Booster Module

JUWELS Booster Module – At 44.1 Pflop/s, this BullSequana system is the most powerful in Europe. It’s based on AMD EPYC processors with Nvidia A100s for acceleration and a Mellanox HDR InfiniBand as a network, which is similar to the Selene system. Built by Atos, it is installed at the Forschungszentrum Juelich (FZJ) in Germany.

Number 9: HPC5

HPC5 – This a PowerEdge-system build by Dell that achieves a performance of 35.5 Pflop/s using Nvidia Tesla V100s as accelerators and a Mellanox HDR InfiniBand as a network. It was installed by the Italian company Eni S.p.A.

Number 10: Frontera

Frontera – This Dell C6420 achieved 23.5 Pflop/s using 448,448 Intel Platinum Xeon cores and resides at the Texas Advanced Computing Center of the University of Texas.