The world’s quickest supercomputer, by the numbers

A US supercomputer has formally taken the highest spot on the earth, in response to a current record by TOP500, a undertaking that tracks and ranks 500 of probably the most highly effective machines on the earth.

Coming in at primary is a Tennessee-based system referred to as Frontier, run by Hewlett Packard Enterprise, AMD, and the US Division of Power’s Oak Ridge Nationwide Laboratory, edging out the Fugaku system on the RIKEN Middle for Computational Science (R-CCS) in Kobe, Japan that beforehand held the title.

This Frontier system is “the results of greater than a decade of collaboration among the many nationwide laboratories, academia and personal trade, together with DOE’s Exascale Computing Mission, which is deploying the functions, software program applied sciences, {hardware} and integration essential to make sure influence on the exascale,” Oak Ridge Nationwide Laboratory mentioned in a press launch.

What can a supercomputer of this measurement and complexity do?  A presentation in March said that a number of the early science run on Frontier contains simulating Earth techniques to investigate biogeochemical cycles in addition to energy grid planning. As these supercomputers get bigger, specialists count on them to change into specialised at a handful of becoming functions.

Right here’s a have a look at what Frontier can do, by the numbers.

1.1 exaflops

The Frontier supercomputer boasts 1.1 exaflops of efficiency, which signifies that it’s in a position to deal with a few quintillion calculations per second, Oak Ridge Nationwide Laboratory mentioned in a press launch. That’s 10^18, or 1,000,000,000,000,000,000. Theoretically, Frontier can chart 2 exaflops at peak efficiency—that’s two quintillion calculations per second. That is 10 instances extra highly effective than the Oak Ridge Nationwide Laboratory’s earlier IBM Summit system, which had a peak efficiency of 200 petaflops, or 200,000 trillion calculations per second. Summit was adept at analyzing and evaluating genomes, in addition to utilizing deep studying to simulate world climate patterns. (Right here’s extra on what a “flop” and a teraflop are.)

The primary true exaflop machine

The Fugaku system in Japan, which held the highest place on the record for 2 years prior, maintains a efficiency of 442 petaflops per second, in response to TOP500. Although Fugaku might hit the 1 exaflop barrier at peak efficiency, Frontier has decidedly been topped the primary exascale machine. Intel and Argonne Laboratory’s Aurora was additionally anticipated to hit this milestone, however the system has reportedly been delayed as a consequence of manufacturing and expertise points. One other vital caveat: There’s discuss that two techniques in China have already moved previous this milestone silently, and their operators didn’t submit outcomes for analysis to the scientists who oversee TOP500, in response to The New York Instances.

[Related: This supercomputer will perform 1,000,000,000,000,000,000 operations per second]

52.23 gigaflops efficiency per watt

Supercomputers want loads of electrical energy to run. Nonetheless, Frontier and the Frontier Check & Growth System (TDS) at Oak Ridge Nationwide Laboratory are additionally two of probably the most energy-efficient supercomputers out of the group that TOP500 ranked, with the Frontier system having an influence effectivity of 52.23 gigaflops per watt. The Frontier TDS system—a smaller model of the supercomputer used for operating early assessments—which solely has 120,832 cores and a efficiency of 19.20 petaflops per second, is probably the most energy environment friendly system, with a ranking of 62.68 gigaflops per watt.

8,730,112 cores

Talking of cores, Frontier has about 8,730,112 of them able to processing parallel duties. Frontier is constructed from the HPE Cray EX supercomputers, which comprise AMD’s processors. All of that processing energy can be utilized in the direction of fixing elaborate issues and calculations, together with these wanted for synthetic intelligence techniques. “Along with modeling and simulating advanced scientific analysis, throughout organic, bodily and chemical sciences, with greater decision, Frontier can even allow dramatic breakthroughs in AI,” Hewlett Packard mentioned in a press launch. “At an exascale pace, Frontier’s customers can develop AI fashions which might be 4.5X sooner and 8X bigger.” Listed here are all of the programming languages, libraries, and instruments obtainable on Frontier.

700 petabytes of storage

Along with having learn speeds of as much as 75 terabytes per second, write speeds of as much as 35 terabytes per second, and with the ability to do round 15 billion random-read enter/output operations per second, Frontier’s storage element, referred to as Orion, can maintain about 700 petabytes of knowledge. To place it in perspective, Oak Ridge Nationwide Laboratory mentioned that represents 35 instances greater than the information held by The Library of Congress.

74 cupboards

Supercomputers have to divide up their huge {hardware} into totally different compartments. Frontier is made up of 74 supercomputer cupboards, 9,400 AMD-powered nodes, and 90 miles of networking cables. Every cupboard weighs round 8,000 kilos, or as a lot as a pick-up truck. In complete, the system incorporates round 9,408 CPUs and 37,632 GPUs (GPUs are significantly helpful for operating synthetic intelligence software program). It consumes 29 megawatts of energy. Researchers are anticipated to realize entry to the “totally operational” Frontier someday this yr.