How Frontier Became the First Exascale Supercomputer and What It Means for Science

The Frontier supercomputer from Oak Ridge National Laboratory (ORNL) has claimed the top spot on the TOP500 list, overtaking Japan's Fugaku and becoming the first machine to exceed one exaflop (1.1 × 10¹⁸ flops) in the Linpack benchmark.

Built on the HPE Cray EX235a architecture, Frontier combines AMD third‑generation EPYC 64‑core processors with Instinct MI250X accelerators, totaling 8,730,112 cores.

In addition to the performance lead, Frontier also tops the GREEN500 and HPL‑AI rankings, delivering an energy efficiency of 52.23 gigaflops per watt.

The system comprises 74 purpose‑built cabinets, housing more than 9,400 AMD‑based nodes, over 37,000 GPUs, 700 PB of storage, a 90‑mile (≈145 km) network, and liquid‑cooling capable of removing about 6,000 gallons of water per minute. Its theoretical peak performance approaches two exaflops.

With Frontier’s debut, the United States now occupies five of the top ten positions in the TOP500 list (1st, 4th, 5th, 7th, and 8th), and five of the top ten machines use AMD processors.

China’s fastest listed system remains the Sunway TaihuLight at sixth place; however, China still leads in the total number of supercomputers worldwide, operating 186 machines compared with the United States' 123.

Frontier is still undergoing testing and validation and is expected to enter full scientific research operations in early next year.