//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>
The U.S. announced the world’s first exascale computer this week, Frontier, powered by AMD chips. Oak Ridge National Laboratory’s Frontier system led the Top500 list at 1.1 exaflops, making it the first to break the exascale barrier.
The system dials down energy consumption and more than doubles the second–ranked system on the Top500 list, according to experts at Top500 and Green500.
The Frontier system also topped the Green500 list, delivering 62.68 gigaflops/watt power–efficiency from a single cabinet of optimized 3rd Gen AMD Epyc processors and AMD Instinct accelerators. Frontier’s mixed–precision computing performance reached 6.86 exaflops, as measured by the High–Performance Linpack–Accelerator Introspection (HPL–AI) test.
Since 1993, Top500 has listed the world’s most powerful computing systems. The list is compiled with the help of high–performance computing (HPC) experts, scientists, manufacturers, and the internet community.
The Top500 and Green500 lists show a growing preference for AMD silicon. On the Top500, AMD powers 94 systems, an increase of 95% year–over–year. AMD Instinct MI200 accelerators made their first entry to the Top500 in seven systems.
“Innovation and delivering more performance and efficiency for supercomputers is critical to addressing the world’s most complex challenges,” said Forrest Norrod, senior vice president of the Data Center Solutions Group at AMD, in a press statement. “Epyc processors and AMD Instinct accelerators continue to push the envelope in HPC, providing the performance needed to advance scientific discoveries.”
Growing HPC applications include materials science, energy issues, and national security. Leading chip foundry Taiwan Semiconductor Manufacturing Co., which fabricated the AMD chips used in the Frontier system, said that HPC early this year became the largest segment of its business.
Oak Ridge is one among many, including commercial businesses such as Google, making wider use of HPC.
“Our collaboration with AMD has been critical for us to ensure that we deploy the world’s leading platform for computational science,” said Bronson Messer, director of science at Oak Ridge Leadership Computing Facility.
Since its founding in May 1992, the Oak Ridge, Tennessee, facility has developed supercomputers for the scientific community on behalf of the U.S. Department of Energy (DOE), which includes nuclear weapons as part of its bailiwick. Scientists have used Oak Ridge systems to solve problems in areas involving biology, advanced materials, climate, and nuclear physics.
For years, nations around the world have been in a race to announce the first exascale computer.
China’s National Supercomputing Center in Wuxi may have first broken the exascale barrier in March 2021 on its Sunway Oceanlite system powered by chips from Phytium, according to unverified reports. China has used its Phytium–based supercomputers to design the world’s first hypersonic missiles, which America’s top general likened to a Sputnik moment.
A month after the Oceanlite report, the U.S. placed Phytium on a blacklist that prohibits the company from the use of American technology in April last year.
There’s also been a global race to develop the world’s first quantum computers, which promise to solve problems exceeding the ability of exascale machines.
The HPL–AI benchmark used to rank the Oak Ridge Frontier system highlights the convergence of HPC and AI workloads. While traditional HPC focused on simulation models for physics, chemistry, and biology, such models typically require 64–bit accuracy.
More recently, the machine–learning methods powering AI achieve desired results at 32–bit and even lower floating–point precision formats. This reduced demand for accuracy fueled interest in hardware platforms that deliver new performance levels and energy savings.
The road to exascale computing hasn’t been easy. In 2013, the U.S. DOE planned to have an exascale system by 2020, designed by Cray using Intel processors. In 2017, Intel announced that Aurora would be delayed to 2021. In October 2020, the DOE said that Aurora would be delayed again.
HPC in the cloud accelerates key research applications and business workloads. The latest Google Cloud C2D virtual machines (VMs) use Epyc processors. Electric supercar innovator Rimac recently deployed VMs powered by Epyc processors with AMD 3D V–Cache technology to increase performance and scalability for electric supercar simulations.
In Europe, CSC’s LUMI supercomputer is third on the Top500 list with 152 petaflops of performance and third on the Green500 list with 51.63 gigaflops/watt power–efficiency. The Adastra system at GENCI–CINES is tenth on the Top500 list and fourth on the Green500 list.
“LUMI, one of the EuroHPC world–class supercomputers and leading platforms for AI, is built with leadership environmental sustainability and performance in mind,” said Pekka Manninen, director of the LUMI Leadership Computing Facility, CSC. “AMD Epyc processors and AMD Instinct accelerators enable us to reach our most ambitious scientific research goals while at the same time meeting and exceeding the EU’s most stringent climate targets.”
AMD is helping government organizations and private customers deploy computing clusters of various sizes across research areas including manufacturing, life sciences, financial services, climate research, and more.
The Ohio Supercomputer Center recently announced Ascend, a new HPC cluster comprised of Dell Technologies PowerEdge servers powered by Epyc processors.