One of two HPE Cray EX supercomputers to exceed an exaflop, Aurora is the second-fastest supercomputer in the world.
Hewlett Packard Enterprise has announced it has delivered the world’s second exascale supercomputer, Aurora, in collaboration with Intel for the United States Department of Energy’s Argonne National Laboratory.
Aurora has reached 1.012 exaflops on 87% of the system, making it the world’s second-fastest supercomputer as verified by the TOP500 list of the most powerful supercomputers. HPE is one of the world leaders in supercomputing and Aurora is not only the company’s second exascale system, but also one of the largest AI-capable systems in the world, taking the top spot on the HPL Mixed Precision (MxP) Benchmark, achieving 10.6 exaflops on 89% of the system.
“We are honoured to celebrate another significant milestone in exascale with Aurora, which delivers massive compute capabilities to make breakthrough scientific discoveries and help solve the world’s toughest problems,” said Trish Damkroger, Senior Vice President and General Manager, HPC & AI Infrastructure Solutions, HPE. “We are proud of the strong partnership with the US Department of Energy, Argonne National Laboratory, and Intel to realise a system of this scale and magnitude that was made possible through our joint innovative engineering, multiple teams and most importantly, shared value of delivering state-of-the-art technology to fuel science and benefit humankind.
“We are proud of the strong partnership with the US Department of Energy, Argonne National Laboratory, and Intel to realise a system of this scale and magnitude that was made possible through our joint innovative engineering, multiple teams and most importantly, shared value of delivering state-of-the-art technology to fuel science and benefit humankind,” Damkroger added.
An exascale computing system can process one quintillion operations per second.
Computational power at this scale makes it possible to address some of humanity’s most complex problems. Aurora is built with the HPE Cray EX supercomputer, which is purpose-built to support the magnitude and scale of exascale. The system is also the largest deployment of open, Ethernet-based supercomputing interconnect – HPE Slingshot – on a single system.
The fabric connects Aurora’s 75,000 compute node endpoints and 2,400 storage and service network endpoints along with 5,600 switches to boost performance and achieve AI at scale by enabling high-speed networking across Aurora’s 10,624 compute blades, 21,248 Intel Xeon CPU Max Series processors and 63,744 Intel Data Center GPU Max units, making it one of the world’s largest GPU clusters.
Planned as an AI-capable system from inception, researchers will be able to use Generative AI models on Aurora to accelerate scientific discovery. Early AI-driven research that scientists have run on Aurora include brain mapping to better understand the human brain’s 80 billion neurons, high energy particle physics enhanced by Deep Learning, and Machine-Learning accelerated drug design and discovery, among others.
Computational power at this scale makes it possible to address some of humanity’s most complex problems.
“Aurora is a first-of-its-kind supercomputer and we expect it to be a gamechanger for researchers,” said Rick Stevens, Associate Laboratory Director and Distinguished Fellow at Argonne National Laboratory. “Reaching this milestone with a second exascale system in the US is an incredibly significant achievement that will advance open science initiatives globally.”
The Aurora exascale supercomputer is the result of a strong private-public partnership between HPE, Intel, the US Department of Energy and Argonne National Laboratory, which requires co-investment and co-development to create the breakthrough engineering that is necessary to advancing science.
Partnerships between the private and public sectors are key to achieving scientific progress as demonstrated by work done through the Aurora Early Science Program. As part of the process of optimising and stress-testing the system, researchers have already successfully run a diverse range of programming models, languages and applications on the system.
“The Aurora supercomputer was designed to support the research and science communities within the HPC and AI space,” said Ogi Brkic, Intel Vice President and General Manager, Data Center AI Solutions. “Our ongoing collaboration with Argonne National Laboratory and HPE has resulted in promising early science success stories. And we’re excited to see what’s to come as we continue to optimise system performance to accelerate the science and march toward what is next.”
Aurora has achieved exascale on a partial run of the system, tapping 9,234 of the total nodes. Aurora is an open science system housed at the Argonne Leadership Computing Facility (ALCF), a part of the US Department of Energy’s Office of Science user facility.
