The National Science Foundation has awarded an English professor at a U.S. university $107,600 to help solve the mystery of the Aurora supercomputer. Two-year grant recipients will document what the Aurora supercomputer does and how scientists use it.
Professor Sarah Reid of Portland State University (PSU), Professor Jordan Frith of Clemson University, and colleagues will investigate why the United States needs supercomputers and the large-scale infrastructure to run those systems. .
“This project aims to advance the public’s understanding of large-scale supercomputing infrastructure to answer these questions and generate new forms of knowledge,” NSF says on the grant page. There is.
The grant envisioned Aurora becoming the world’s fastest computer, but that hasn’t happened yet. The Frontier supercomputer maintained its top position on the Top500 list, with Argonne’s Aurora taking second place.
Argonne Laboratory’s Aurora supercomputer (Credit: Argonne/ALCF)
However, Argonne led the Top 500 in HPL-MxP, which is moving into AI computing. The makers of Aurora made a conscious decision not to include a processing unit that improves a key Top500 benchmark called LINPACK.
The British professors will receive full cooperation from the staff of the Aurora supercomputer at Argonne National Laboratory in Illinois.
Scientists and staff at Argonne University will be interviewed, which will also be part of the study. The final report is expected to explain the US government’s decisions regarding investments in public computing infrastructure.
A press release on PSU’s website explains that “one of the companies’ focuses will be user documentation to support scientists in preparing projects to run on supercomputers.”
There are many possible purposes for such a study. This could be a fundamental record for measuring the usefulness, practicality, and efficiency of America’s scientific computing equipment.
NREL’s September newsletter justified how the facility’s high-speed supercomputers will help speed up research. NREL’s newest supercomputer found that the U.S. could reach 100% renewable energy by 2035, while older supercomputers dating back to 2013 previously predicted the U.S. could reach 80% renewable energy by 2045. We predicted that we would reach renewable energy.
PSU’s Sarah Read was previously a visiting scholar at ALCF and author of numerous papers on U.S. supercomputing infrastructure. This includes scientific and technical computing research on these systems, operational efficiency, and comparative results of computing equipment.
“By talking to users about their experiences using and creating documents, we learn about the unknowns and uncertainties of using cutting-edge supercomputers and how those edges shape researchers’ science. ,” Reed said in a PSU release. .
In 2022, the two published a paper on “infrastructure research” that drives technological processes and large-scale technology projects.
“Underneath nearly every high-level practice are layers of social and material infrastructure that shape those practices,” the researchers write in their paper. states.
Aurora itself took eight years to build and cost about $500 million. Aurora is equipped with over 60,000 Intel GPUs and was until recently considered one of the largest GPU installations in the world.
Aurora has over 10,000 compute nodes, 166 racks, and 80,000 networking nodes. Aurora is currently training AuroraGPT, the largest scientific AI model with 1.2 million parameters. Data is based on scientific research papers.
The U.S. government invests billions of dollars in scientific research and discovery to establish competitive leadership.
