The World’s Fastest Supercomputer Is About To Be Replaced By A 5x More Powerful Machine

The US Department of Energy (DOE) has revealed intentions to keep the US at the forefront of high-performance computing. Frontier, the world’s fastest supercomputer at the moment, will be replaced by a new supercomputer called Discover. The DOE released a request for proposals (RFP) earlier this month for Discover, which is expected to be five times more potent than Frontier.

As of May of this year, Frontier, which is located at Tennessee’s Oak Ridge National Laboratory (ORNL), has been ranked first among the Top 500 supercomputers in the world for five times running. Frontier, which was constructed using AMD Epyc processors on HPE Cray EX architecture, has a high-performance Linpack benchmark score of 1.206 exaflops and almost 8.7 million combined CPU and GPU cores.

While the DOE has not specified an exact performance target for Discover, it is expected to deliver three to five times more computational throughput than Frontier, potentially exceeding 8.5 exaflops. “This project is exciting because we will be building something more capable than Frontier, with technologies that will push the edge of what’s possible,” said Matt Sieger, ORNL’s project director for Discover. Discover aims to revolutionize scientific research across various fields, driving breakthroughs in climate change prediction, drug discovery, high-energy physics, and green energy solutions with its enhanced computational power. Georgia Tourassi, ORNL’s associate laboratory director of computing and computational sciences, emphasized that Discover would enable the scientific community to model real-world situations at unprecedented detail. “It will help us study challenging problems we can’t easily explore with experiment, observation, or theory alone.”

Beyond traditional scientific applications, Discover is designed to excel in advanced artificial intelligence and machine learning tasks, pushing the boundaries in fields like materials science and industrial product design. Additionally, Discover will play a crucial role in the DOE’s Integrated Research Infrastructure (IRI) initiative, aiming to combine various research tools and scientific facilities.

The DOE has allotted time until August 30, 2024, for interested vendors to submit proposals for Discover, with the target delivery date set for 2027 or early 2028. The DOE has outlined objectives for the next-generation system, including improved energy efficiency and full-system modeling and simulation, alongside advanced AI and machine learning capabilities.

Energy efficiency is a top priority for the OLCF, which has boosted its computational power 500x while only increasing energy consumption fourfold over the past decade. Once operational, Discover is expected to be accessible to researchers worldwide, competing for computing time to tackle scientific challenges.

“These discoveries will help shape our understanding of the universe, advance our capabilities for the predictability of the Earth system, bolster US economic competitiveness, and contribute to a better future,” emphasized Dr. Ceren Susut, associate director for science in the DOE’s Advanced Scientific Computing Research program.