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

0
41


Enlarge / Programs engineer Matt Ezell, the system lead for the Frontier supercomputer at Oak Ridge Nationwide Laboratory.

The US has retaken the highest spot on the earth supercomputer rankings with the exascale Frontier system at Oak Ridge Nationwide Laboratory (ORNL) in Tennessee.

The Frontier system’s rating of 1.102 exaflop/s makes it “essentially the most highly effective supercomputer to ever exist” and “the primary true exascale machine,” the Prime 500 venture stated Monday within the announcement of its newest rankings. Exaflop/s (or exaflops) is brief for 1 quintillion floating-point operations per second.

Frontier was greater than twice as quick as a Japanese system that positioned second within the rankings, that are based mostly on the LINPACK benchmark that measures the “efficiency of a devoted system for fixing a dense system of linear equations.”

“Based mostly on the newest HPE Cray EX235a structure and outfitted with AMD EPYC 64C 2GHz processors, the system has 8,730,112 whole cores, an influence effectivity score of 52.23 gigaflops/watt, and depends on gigabit Ethernet for information switch,” the Prime 500 group stated. Frontier “is presently being built-in and examined on the ORNL in Tennessee, USA, the place it will likely be operated by the Division of Vitality,” the group stated.

Though China might also have cracked the exaflop barrier, the US system is the primary to have demonstrated the velocity in official Prime 500 testing.

Frontier additionally tops effectivity rating

Frontier was additionally the second-most environment friendly supercomputer within the new Green 500 rankings. The highest spot in Inexperienced 500 was taken by a smaller model of the Frontier system at Oak Ridge, one with 120,832 cores as an alternative of the 8.7 million within the exascale system.

Taking second place within the Prime 500 with a rating of 442 petaflop/s was the earlier high supercomputer, the Fugaku system on the RIKEN Heart for Computational Science in Kobe, Japan. Whereas the Fugaku system has a theoretical peak above 1 exaflop/s, Frontier is the one system that has demonstrated it in Prime 500 testing, the Prime 500 announcement stated. ORNL stated that Frontier’s theoretical peak is 2 exaflop/s.

Fugaku was in first place beginning with the June 2020 ranking and stayed on high in November 2020, June 2021, and November 2021. The US had final held first place within the twice-yearly rankings in November 2019 when Oak Ridge’s IBM-built Summit system posted a rating of 148.6 petaflop/s.

DOE’s exascale venture

The DOE in 2018 announced a $1.8 billion venture to develop no less than two exascale supercomputers to help a variety of scientific analysis.

“The work to ship, set up and take a look at Frontier started throughout the COVID-19 pandemic, as shutdowns world wide strained worldwide provide chains,” ORNL said this week. “Greater than 100 members of a public-private workforce labored across the clock, from sourcing hundreds of thousands of elements to making sure deliveries of system components on deadline to rigorously putting in and testing 74 HPE Cray EX supercomputer cupboards, which embrace greater than 9,400 AMD-powered nodes and 90 miles of networking cables.”

Researchers will “acquire entry to the absolutely operational Frontier system later this 12 months,” ORNL affiliate lab director Jeff Nichols stated, including that it will likely be obtainable to “scientists and engineers from world wide.”

“Frontier customers will mannequin your entire lifespan of a nuclear reactor, uncover illness genetics, and construct on latest developments in science and expertise to additional combine synthetic intelligence with information analytics and modeling and simulation,” ORNL additionally said.

HPCwire’s coverage of the Prime 500 listing offered an extra description of Frontier’s composition:

Frontier consists of 74 cupboards, every weighing in at 8,000 kilos. 9,408 HPE Cray EX nodes are unfold throughout these cupboards, with every node powered by an AMD “Trento” 7A53 Epyc CPU and quadruple AMD Intuition MI250X GPUs (totaling 37,632 GPUs throughout the system). The system has 9.2 petabytes of reminiscence, cut up evenly between HBM and DDR4, and it makes use of HPE Slingshot-11 networking. It’s supported by 37 petabytes of node-local storage on high of 716 petabytes of center-wide storage. The system is one hundred pc liquid-cooled utilizing heat (85° F) water, and by way of house, the ultimate Frontier system occupies 372 sq. meters.

HP Enterprise bought the supercomputing firm Cray in 2019.

China reportedly has exascale supercomputers

There are indications that China has two exascale supercomputers. However operators “of these techniques haven’t submitted take a look at outcomes for analysis” within the Prime 500 rating, presumably resulting from “tensions between the USA and China,” The New York Instances wrote.

Formally, China’s top-ranked system is the sixth quickest within the Prime 500. That system, referred to as Sunway TaihuLight, was ranked No. 1 4 straight instances in 2016 and 2017 however hasn’t up to date its rating since. One other Chinese language system known as Tianhe-2A held the highest spot from 2013 to 2015 and is now ninth.

“There are rumors China has one thing” with exascale energy, however “there may be nothing official,” Jack Dongarra, a pc scientist who designed Linpack, informed The New York Instances. Dongarra is a professor on the College of Tennessee and holds a place at ORNL.



Source link

LEAVE A REPLY

Please enter your comment!
Please enter your name here