New Exascale Supercomputer by HPE Optimized for AI

Aurora, the world's second-fastest supercomputer, has reached an impressive milestone of 1.012 exaflops on 87% of its system, as confirmed by the prestigious TOP500 list of supercomputers. Developed by HPE, a global leader in supercomputing, Aurora stands out not only as the company's second exascale system but also as the largest AI-capable system worldwide. It has claimed the top spot on the HPL Mixed Precision Benchmark, achieving a remarkable 10.6 exaflops on 89% of its system.

An exascale computing system, capable of processing one quintillion operations per second, opens up new possibilities for tackling some of humanity's most complex challenges. Aurora is powered by the HPE Cray EX supercomputer, specifically designed to handle the immense scale of exascale computing. The system boasts the largest deployment of the open, Ethernet-based supercomputing interconnect known as HPE Slingshot, connecting 75,000 compute node endpoints, 2,400 storage and service network endpoints, and 5,600 switches to enhance performance across its 10,624 compute blades, 21,248 Intel Xeon CPU Max Series processors, and 63,744 Intel Data Center GPU Max units, forming one of the world's largest GPU clusters.

From its inception, Aurora was planned as an AI-capable system, enabling researchers to leverage generative AI models for accelerating scientific discoveries. Early experiments on Aurora have delved into diverse fields such as brain mapping to unravel the complexities of the human brain's 80 billion neurons, high-energy particle physics enhanced by deep learning, and machine-learning-driven drug design and discovery processes.

"Aurora is a groundbreaking supercomputer that is poised to revolutionize research," remarked Rick Stevens, associate laboratory director and distinguished fellow at Argonne National Laboratory. "Achieving this milestone with a second exascale system in the U.S. is a significant accomplishment that will drive global open science initiatives forward."

The realization of the Aurora exascale supercomputer is the outcome of a robust public-private partnership involving HPE, Intel, the U.S. Department of Energy, and Argonne National Laboratory. This collaborative effort, requiring co-investment and co-development, has paved the way for breakthrough engineering that propels scientific advancements. The Aurora Early Science Program exemplifies the importance of partnerships between the private and public sectors in fostering scientific progress, as evidenced by the successful testing of various programming models, languages, and applications on the system.

"Aurora was purposefully designed to support the research and scientific communities in the realms of HPC and AI," stated Ogi Brkic, Intel's vice president and general manager of Data Center AI Solutions. "Our ongoing collaboration with Argonne National Laboratory and HPE has yielded promising outcomes in early scientific endeavors."

Having achieved exascale performance on a partial run of the system utilizing 9,234 nodes, Aurora represents an open science platform hosted at the Argonne Leadership Computing Facility (ALCF), an integral part of the U.S. Department of Energy's Office of Science user facility.