1.1 quintillion operations per second: US has world’s fastest supercomputer

A systems engineer stands in a large room containing the Frontier supercomputer.
Enlarge / Systems engineer Matt Ezell, the procedure guide for the Frontier supercomputer at Oak Ridge National Laboratory.

The US has retaken the top rated place in the world supercomputer rankings with the exascale Frontier process at Oak Ridge Countrywide Laboratory (ORNL) in Tennessee.

The Frontier system’s score of 1.102 exaflop/s helps make it “the most effective supercomputer to ever exist” and “the initially legitimate exascale device,” the Top 500 task explained Monday in the announcement of its latest rankings. Exaflop/s (or exaflops) is shorter for 1 quintillion floating-point functions per second.

Frontier was additional than 2 times as rapid as a Japanese procedure that put second in the rankings, which are primarily based on the LINPACK benchmark that steps the “efficiency of a focused program for fixing a dense procedure of linear equations.”

“Primarily based on the latest HPE Cray EX235a architecture and equipped with AMD EPYC 64C 2GHz processors, the procedure has 8,730,112 whole cores, a electricity effectiveness ranking of 52.23 gigaflops/watt, and relies on gigabit Ethernet for knowledge transfer,” the Prime 500 group said. Frontier “is presently becoming integrated and tested at the ORNL in Tennessee, United states of america, exactly where it will be operated by the Section of Electrical power,” the group claimed.

While China could also have cracked the exaflop barrier, the US program is the first to have shown the speed in official Prime 500 screening.

Frontier also tops performance ranking

Frontier was also the second-most economical supercomputer in the new Environmentally friendly 500 rankings. The leading location in Environmentally friendly 500 was taken by a smaller model of the Frontier program at Oak Ridge, 1 with 120,832 cores as a substitute of the 8.7 million in the exascale system.

Getting next location in the Top 500 with a score of 442 petaflop/s was the prior best supercomputer, the Fugaku method at the RIKEN Heart for Computational Science in Kobe, Japan. When the Fugaku procedure has a theoretical peak higher than 1 exaflop/s, Frontier is the only program that has demonstrated it in Top rated 500 tests, the Best 500 announcement mentioned. ORNL stated that Frontier’s theoretical peak is 2 exaflop/s.

Fugaku was in initial position beginning with the June 2020 rating and stayed on top rated in November 2020, June 2021, and November 2021. The US had past held to start with spot in the 2 times-yearly rankings in November 2019 when Oak Ridge’s IBM-constructed Summit method posted a score of 148.6 petaflop/s.

DOE’s exascale challenge

The DOE in 2018 introduced a $1.8 billion task to establish at least two exascale supercomputers to assist a assortment of scientific analysis.

“The function to supply, install and exam Frontier began during the COVID-19 pandemic, as shutdowns close to the earth strained international source chains,” ORNL said this 7 days. “Extra than 100 customers of a public-private crew labored all around the clock, from sourcing millions of factors to ensuring deliveries of technique elements on deadline to cautiously putting in and screening 74 HPE Cray EX supercomputer cabinets, which include a lot more than 9,400 AMD-powered nodes and 90 miles of networking cables.”

Researchers will “obtain entry to the totally operational Frontier system afterwards this yr,” ORNL affiliate lab director Jeff Nichols stated, incorporating that it will be out there to “experts and engineers from all over the world.”

“Frontier users will design the overall lifespan of a nuclear reactor, uncover condition genetics, and create on recent developments in science and technologies to even more integrate synthetic intelligence with knowledge analytics and modeling and simulation,” ORNL also explained.

HPCwire’s coverage of the Major 500 list provided a even further description of Frontier’s composition:

Frontier is composed of 74 cupboards, every single weighing in at 8,000 kilos. 9,408 HPE Cray EX nodes are spread across these cupboards, with every single node run by an AMD “Trento” 7A53 Epyc CPU and quadruple AMD Instinct MI250X GPUs (totaling 37,632 GPUs throughout the system). The system has 9.2 petabytes of memory, break up evenly among HBM and DDR4, and it employs HPE Slingshot-11 networking. It is supported by 37 petabytes of node-local storage on major of 716 petabytes of middle-wide storage. The procedure is 100 per cent liquid-cooled employing heat (85° F) water, and in terms of space, the remaining Frontier procedure occupies 372 square meters.

HP Enterprise bought the supercomputing enterprise Cray in 2019.

China reportedly has exascale supercomputers

There are indications that China has two exascale supercomputers. But operators “of those people methods have not submitted examination effects for analysis” in the Top rated 500 position, maybe thanks to “tensions involving the United States and China,” The New York Instances wrote.

Officially, China’s top rated-rated method is the sixth quickest in the Leading 500. That system, identified as Sunway TaihuLight, was rated No. 1 four straight moments in 2016 and 2017 but has not up to date its score considering that. Yet another Chinese program termed Tianhe-2A held the leading place from 2013 to 2015 and is now ninth.

“There are rumors China has some thing” with exascale energy, but “there is very little official,” Jack Dongarra, a laptop scientist who made Linpack, told The New York Times. Dongarra is a professor at the College of Tennessee and holds a posture at ORNL.