Exascale Supercomputers are Diversifying the World’s Most Powerful Platforms for AI-Enabled Science

The U.S. Department of Energy’s (DOE) Aurora supercomputer has officially broken the exascale barrier. Today at the 2024 ISC High Performance conference in Hamburg, Germany, the 63^rd edition of the high performance computing Top500 list announced that DOE holds the #1 and #2 positions for most powerful supercomputers in the world. The Top500’s benchmark has long been the world’s measuring stick for large scale supercomputing performance. 

The Aurora supercomputer at Argonne National Laboratory, built in partnership with Intel and HPE, achieved 1.012 exaflops, joining Oak Ridge National Laboratory’s Frontier supercomputer (1.206 exaflops) as the only two official exascale systems in the world. An exaflop is equivalent to one quadrillion high precision calculations per second. 

“The Department of Energy is thrilled by the Aurora team’s achievement,” said Geraldine Richmond, DOE’s Under Secretary for Science and Innovation. “Aurora is expanding U.S. leadership in advanced scientific computing and will provide researchers with a unique instrument to meld simulation, artificial intelligence (AI), and data analysis. Frontier and Aurora will be pivotal tools in advancing trustworthy AI as we focus on harnessing AI’s potential while understanding and managing its risks.”

“This announcement heralds Aurora’s readiness to deliver science and join Frontier in extending the impact of our investments in the exascale computing ecosystem,” said Harriet Kung, Acting Director of the DOE’s Office of Science. “Congratulations to the integrated Argonne, Intel, and HPE team for achieving this major milestone.”Aurora is a massive first-of-its-kind system spanning 10,624 computing nodes, including 63,744 Intel graphics processing units (GPUs), the computational engines driving the AI revolution. Aurora’s thousands of processors are tied together by HPE’s highspeed Slingshot interconnect. The Aurora achievement utilized 87 percent of the system, with even higher performance expected as the system nears completion. 

The Oak Ridge team has steadily improved Frontier’s performance since its debut as #1 on the Top500 in 2022. The Argonne and Oak Ridge teams are close collaborators.

Frontier has already delivered significant science impact since entering full operations, highlighted by two prestigious Association for Computing Machinery Gordon Bell prizes. Last November, a team of eight scientists won the Gordon Bell Prize for a new approach to materials discovery, using Frontier to run the largest-ever simulation of an alloy to near quantum (i.e., individual atom) accuracy. A separate 19-member team won the 2023 Gordon Bell Special Prize for Climate Modeling for a major improvement in cloud modeling, punctuating years of effort under DOE’s Energy Exascale Earth System Model, or E3SM, to prepare for exascale. 

The Argonne Leadership Computing Facility, home to Aurora and other advanced scientific computing platforms open to the research community, has enabled teams participating in the Aurora Early Science Program and DOE’s Exascale Computing Project to trial Aurora, exercising the supercomputer and the software that powers scientific research. Projects ranging from design of new materials and pharmaceuticals to fundamental physics and cosmology have provided tantalizing glimpses of the raw power the full Aurora system will deliver for science. 

These two supercomputing achievements form a strong basis for DOE’s recently announced FASST initiative. The FASST (Frontiers in Artificial Intelligence for Science, Security and Technology) initiative will harness DOE’s scientific data for AI training, advance next generation AI platforms and infrastructure, build safe and trustworthy AI models and systems and apply them to the nation’s most pressing scientific, energy, and national security challenges.