AMD и Cray Frontier

2024: Supercomputer creates largest model of universe in history

At the end of November 2024, researchers from the Argonne National Laboratory power engineering specialists USA of the Ministry announced the creation of the largest model of the universe in history. The project, called ExaSky, is expected to help scientists better understand the evolution and physics of the universe and study the nature of dark matter. More. here

2022: Frontier supercomputer start-up

On May 30, 2022, Hewlett Packard Enterprise announced that the new Frontier supercomputer, built for the U.S. Department of Energy's Oak Ridge National Laboratory (ORNL), achieved a performance of 1.1 exaflops, becoming the world's first supercomputer to break the exaflop speed barrier and the fastest supercomputer in the world, according to a list of the world's Top500 most powerful supercomputers.

Frontier also came first in the so-called mixed-precision computing category, which rates performance in formats commonly used for artificial intelligence, with a performance of 6.88 exaflops. In addition, the new supercomputer was ranked number one on the Green500 list as the most energy efficient supercomputer in the world at 52.23 gigaflops per watt, making it 32% more energy efficient compared to the previous list leader at 39.38 gigaflops per watt.

Earned the world's first supercomputer with a capacity of more than 1 exaflops

This supercomputer is faster than the 7 other most powerful machines combined from the May 2022 Top500 list. The system will allow scientists to conduct simulations and simulations at the exaflops level to solve problems that are 8 times more difficult, up to 10 times faster. The Frontier is also expected to reach an even higher speed level with a theoretical peak performance of 2 exaflops.

In addition to modeling and simulating complex, higher-resolution biological, physical, and chemical science research, Frontier will also allow for a significant breakthrough in AI. At exaflop speed, Frontier specialists will be able to develop AI models that will work 4.5 times faster and process 8 times more data, which will increase the degree of predictability and speed up the time to discovery.

Frontier is built around HPE Cray EX supercomputers using the following state-of-the-art technologies implemented on other HPE Cray EX model supercomputers to offer significantly higher performance for modeling and simulation at a new level, and to target new AI and machine learning applications to improve accuracy faster and more efficiently:

• 74 HPE Cray EX cases weighing more than 3.62 tons each. All nodes contain one optimized AMD EPYC™ 3rd generation processor and four AMD Instinct MI250x accelerators, totaling 9,408 CPUs and 37,632 GPUs throughout the system.

• Cray Clusterstor E1000 storage system, which enables Frontier Orion storage to provide 700 petabytes of storage, peak write speeds of more than 35 TB/s, and more than 15 billion read-free IOPS.

• Complex liquid cooling technologies, which also contribute to lower noise levels in the data center compared to noisy air-cooled systems, allow efficient heat removal from high-performance devices such as processors, GPUs and switches using an auxiliary cooling distribution unit (CDU).

Today's debut of the Frontier exaflops supercomputer provides a breakthrough in speed and performance and will give us a chance to answer questions we never knew we had, "said Justin Hotard, executive vice president and general manager of HPC&AI at HPE. Frontier is the first of its kind system that was conceived by technologists, scientists and researchers to unleash a new level of opportunity to provide open science, AI and other breakthroughs that will benefit humanity. We are proud of this moment that continues to hold the U.S. lead in supercomputing, now exaflops, made possible by the ongoing public-private partnership between the U.S. Department of Energy, Oak Ridge National Laboratory, HPE and AMD.

When researchers get access to the fully operational Frontier system later this year, it will be the culmination of work that began more than three years ago involving hundreds of talented people from the Department of Energy and our industrial partners from HPE and AMD, said Jeff Nichols, Associate Director of the Lab for Computing and Computing Sciences at Oak Ridge National Laboratory. Scientists and engineers from around the world will use these extraordinary computational speeds to solve some of the most difficult issues of our era, and many of them will begin their research on the first day.[1]

2019: Frontier supercomputer announcement

On May 7, 2019, it became known that AMD, together with the US Department of Energy (DOE), Oak Ridge National Laboratory (ORNL) and Cray Inc. announced the creation of an exascale supercomputer, which is expected to be one of the fastest in the world and will be commissioned into ORNL in 2021.

To provide the expected computational performance of more than 1.5 exaflops, the Frontier system will use custom AMD EPYC processors optimized for high performance computing (HPC) and artificial intelligence and specialized Radeon Instinct GPUs. Researchers at ORNL will use the unprecedented computing power of the Frontier system and next-generation artificial intelligence to simulate, simulate and explore the interactions underlying weather science, subatomic structures, genomics, physics and other important scientific fields.

"Frontier will consist of AMD's own central and GPUs and represents the latest achievement in a long list of technological innovations that have been implemented by AMD in exascale programs for the Department of Energy." Forrest Norrod, Senior Vice President and General Manager, Data Center Solutions and Integrated Solutions Business Group

AMD's innovations in the Frontier project include custom AMD EPYC processors optimized for high performance computing (HPC) and artificial intelligence, and specialized Radeon Instinct GPUs in conjunction with high bandwidth memory (HBM) and the ability to optimally work with several types of data at once, which is necessary in machine learning; Infinity Fabric dedicated high-speed interconnect with low latency to attach four AMD Radeon Instinct GPUs to one AMD EPYC CPU per node; an improved version of the ROCm open source programming environment developed in collaboration with Cray to share AMD CPUs and GPUs.

"Cray Shasta supercomputers are designed to support advanced processor technology and high-performance storage closely linked to the Cray Slingshot network. The combination of Cray and AMD technologies in the Frontier system will significantly increase performance at scale for AI, analytics and modeling, allowing the Department of Energy to further expand the boundaries of scientific discovery. "

As noted in, CNews the Frontier system uses many years of technology investment from the Department of Energy. The contract award includes financing technology development, the establishment of a professional development center, several early delivery systems, the main Frontier system and many years of support for the systems.

"Frontier represents the current state of high-performance computing. Designing and installing a supercomputer of this scale requires close collaboration with industry and a partnership that is making a breakthrough in science. We are excited to work with AMD to integrate the CPU and GPU technologies that underpin this extremely powerful accelerated node architecture[2]

Notes