Supercomputer debuts as world’s quickest exascale barrier

Frontier has arrived and ORNL is gearing up for science from day one. Credit score: Carlos Jones/ORNL, US Division of Power

The Frontier supercomputer on the Division of Power’s Oak Ridge Nationwide Laboratory in the present day gained the No. 1 spot because the world’s quickest on the 59th TOP500 record, with 1.1 exaflops of efficiency. The system is the primary to attain an unprecedented degree of computational efficiency generally known as exascale, a threshold of 1 quintillion computations per second.

Frontier has a theoretical most efficiency of two exaflops, or two quintillion calculations per second, making it ten occasions extra highly effective than ORNL’s Summit system. The system builds on ORNL’s in depth experience in accelerated computing and can allow scientists to develop applied sciences crucial to the nation’s vitality, financial and nationwide safety, serving to researchers clear up issues of nationwide significance that had been unimaginable to resolve simply 5 years in the past.

“Frontier ushers in a brand new period of exascale computing to resolve the world’s best scientific challenges,” mentioned ORNL Director Thomas Zacharia. “This milestone provides only a glimpse of Frontier’s unparalleled potential as a software for scientific discovery. It’s the results of greater than a decade of collaboration between nationwide laboratories, universities, and personal business, together with the DOE’s Exascale Computing Mission, which deploys purposes, software program applied sciences, {hardware}, and integration essential to make sure an affect on the exascale.

The rankings had been introduced throughout the International Supercomputing Conference 2022 in Hamburg, Germany, which brings collectively world leaders within the subject of excessive efficiency computing, or HPC. Frontier’s speeds exceeded these of some other supercomputer on the earth, together with ORNL’s Summit, which can be housed at ORNL’s Oak Ridge Management Computing Facility, a DOE Workplace of Science consumer facility.

Frontier, an HPE Cray EX supercomputer, additionally claimed the highest spot on the Green500 record, which charges the facility consumption and effectivity of commercially out there supercomputing methods, with a efficiency of 62.68 gigaflops per watt. Frontier rounded out the semi-annual rankings with first place in a more moderen class, mixed-precision computing, which assesses efficiency in codecs generally used for synthetic intelligence, with a efficiency of 6.88 exaflops.

Frontier’s supply, set up and testing work started throughout the COVID-19 pandemic, as shutdowns all over the world strained worldwide provide chains. Greater than 100 members of a public-private staff labored across the clock, from sourcing tens of millions of parts, delivering system elements on time, to putting in and painstakingly testing 74 cupboards. HPE Cray EX supercomputers, which embrace greater than 9,400 AMD-powered nodes and 90 miles of community cables.

“When researchers achieve entry to the totally operational Frontier system later this yr, it’s going to mark the fruits of labor begun greater than three years in the past involving a whole bunch of proficient individuals from the Division of Power and our business companions. at HPE and AMD,” ORNL Affiliate Lab Director for Computing and Computational Sciences, mentioned Jeff Nichols. “Scientists and engineers all over the world will put these extraordinary computing speeds to work to resolve a number of the most difficult questions of our time, and plenty of will start their exploration on day one.”

Frontier’s total efficiency of 1.1 exaflops interprets to over a quintillion floating level operations per second, or flops, as measured by the High Performance Linpack Benchmark Test. Every flop represents a attainable calculation, akin to addition, subtraction, multiplication or division.

Frontier’s early efficiency on the Linpack benchmark is greater than seven occasions that of Summit at 148.6 petaflops. Summit continues to be a formidable and extremely rated workhorse for open science, ranked quantity 4 on the TOP500.

Frontier’s mixed-precision compute efficiency reached roughly 6.88 exaflops, or greater than 6.8 quintillion flops per second, as measured by the Linpack-Accelerator High Performance Introspection, or HPL-AI, check. The HPL-AI check measures computational speeds in laptop codecs usually utilized by machine studying strategies that advance synthetic intelligence.

The detailed simulations that conventional HPC customers depend on to mannequin phenomena akin to most cancers cells, supernovae, coronavirus, or the atomic construction of components require 64-bit precision, a demanding type of laptop computation. Machine studying algorithms usually require a lot much less precision – typically as little as 32, 24 or 16 bits – and may benefit from particular {hardware} within the graphics processing models, or GPUs, that machines like Frontier depend on. to attain even sooner speeds.

ORNL and its companions proceed to execute the implementation of Frontier on schedule. Subsequent steps embrace continued testing and validation of the system, which stays on monitor for remaining acceptance and early scientific entry later in 2022 and open for full science in early 2023.

Frontier Info

The exascale efficiency of the Frontier supercomputer is made attainable by a number of the world’s most superior HPE and AMD applied sciences:

  • Frontier has 74 HPE Cray EX supercomputer cupboards, that are purpose-built to assist next-generation supercomputing efficiency and scale, when opened for early scientific entry.
  • Every node incorporates an optimized EPYC processor and 4 AMD Intuition accelerators, for a complete of over 9,400 CPUs and over 37,000 GPUs system-wide. These nodes present builders with simpler capabilities to program their purposes, because of the consistency enabled by EPYC processors and Intuition accelerators.
  • HPE Slingshot, the world’s solely high-performance Ethernet cloth designed for next-generation HPC and AI options, together with bigger, data-intensive workloads, to fulfill calls for for larger velocity and congestion management for apps run easily and enhance efficiency.
  • An I/O subsystem from HPE that can go reside this yr to assist Frontier and OLCF. The I/O subsystem contains an in-system storage layer and Orion, an enhanced central file system based mostly on Luster which can be the world’s largest and quickest single parallel file system based mostly on the Cray ClusterStor E1000 storage system. The system-integrated storage layer will use native compute node storage gadgets related by way of PCIe Gen4 hyperlinks to supply most learn speeds of over 75 terabytes per second, most write speeds of over 35 terabytes per second and over 15 billion random reads. learn enter/output operations per second. The Orion Core File System will present roughly 700 petabytes of storage capability and most write speeds of 5 terabytes per second.
  • As the following era supercomputing system and the world’s quickest for open science, Frontier can be vitality environment friendly, as a consequence of its liquid cooling capabilities. This cooling system promotes a quieter information heart by eliminating the necessity for a noisier, air-cooled system.

AMD technology will power a new supercomputer for the Department of Energy

Offered by
Oak Ridge Nationwide Laboratory

Supercomputer debuts as world’s quickest exascale barrier (2022, Might 30)
retrieved 30 Might 2022

This doc is topic to copyright. Aside from truthful use for functions of personal research or analysis, no
any half could also be reproduced with out written permission. The content material is offered for data solely.

Leave a Reply

Your email address will not be published.