Simulation, AI, data analysis Supercomputer Aurora is now available for all researchers

Related Vendors

Argonne National Laboratory

FAULHABER Drive Systems

Wuxi InfiMotion Technology Co., Ltd.

Aurora, the exascale supercomputer at Argonne National Laboratory, is now available to researchers worldwide, as announced by the system's operators from the U.S. Department of Energy on January 28, 2025. One of the goals for Aurora is to train large language models for science.

According to official reports, among the world's fastest supercomputers, there are currently only three systems that reach at least one exaflop. An exaflop is a quintillion (10¹⁸) calculations per second—that's like a regular calculator computing continuously for 31 billion years, but completing everything in just a single second. Or, to put it briefly: exaflop supercomputers are incredibly fast.

The fastest among the swift three is El Capitan at the Lawrence Livermore National Laboratory with 1.742 exaflops per second under the HPL benchmark (High-Performance Linpack, a standardized test for measuring the computing power of supercomputers). It is followed by Frontier with 1.353 exaflops/s at the Oak Ridge National Laboratory. The trio is completed by Aurora with 1.012 exaflops/s. Incidentally, all three laboratories belong to the U.S. Department of Energy (DOE).

"We are honored to host one of the most powerful supercomputers ever built," said Paul Kearns, Director of Argonne. "The development of the DOE's exascale systems is an important step in advancing fundamental research and strengthening the U.S. leadership in high-performance computing." And these computing capacities are available to the global research community, as announced on January 28, 2025. Until now, only selected scientists had access to Aurora's computing power.

Supercomputer + AI = Super Innovations?

One of the major goals for Aurora, according to Rick Stevens, Associate Laboratory Director for Computing, Environment and Life Sciences at Argonne, is to train Aurora Large Language Models (LLM) for use in the scientific community. "With the AuroraGPT project, we are building a science-oriented foundation model that can distill knowledge from many fields, from biology to chemistry. One of Aurora's goals is to enable researchers to develop new AI tools that help them advance as quickly as they can think—not just as quickly as they can calculate," explains Stevens.

Aurora is already being used to develop highly accurate models of complex systems such as the human circulatory system, nuclear reactors, and supernovae to gain new insights. Additionally, the supercomputer's computing power enables the analysis of large data sets, such as those from research facilities like the U.S. Department of Energy's Advanced Photon Source (APS) and the Large Hadron Collider (LHC) at CERN.

Two basketball courts full of technology

Aurora was developed in collaboration with Intel and Hewlett Packard Enterprise (HPE), just like El Capitan and Frontier. Aurora features 63,744 GPUs, 84,992 network endpoints, and has been connected with over 300 miles of network cable. The facility covers almost 10,000 square meters (about the size of two basketball courts), weighs 600 tons (1,322,773 pounds), and consists of eight rows of computing units. (sb)