El Capitan has officially claimed the title of the world’s fastest supercomputer, delivering an unprecedented 1,742 PetaFLOPS on the High-Performance Linpack benchmark. This performance places it 400 PetaFLOPS ahead of its closest competitor and gives it the potential to exceed 2,700 PetaFLOPS at peak capacity. Developed to aid in nuclear weapons research and high-energy-density physics, El Capitan is set to redefine the landscape of supercomputing.
Built at Lawrence Livermore National Laboratory (LLNL) in California, El Capitan is a $600 million system constructed under the US Department of Energy’s Coral-2 program. Work began in May 2023, and the system went live in November 2024. Now, with its capabilities fully realized, it surpasses all previous exaFLOP-scale supercomputers, including Aurora and Frontier.
At the heart of El Capitan’s power are 44,544 AMD MI300A APUs, integrating Epyc Genoa CPUs and CDNA3 graphics cores. Each processor is equipped with 128GB of high-bandwidth memory (HBM), the same ultra-fast but costly memory once used in high-end graphics cards. Developed by Hewlett Packard Enterprise for the National Nuclear Security Administration (NNSA), El Capitan replaces the aging Sierra supercomputer, delivering nearly 15 times its computational power.
El Capitan’s primary role is to enhance the security and reliability of the US nuclear stockpile. While LLNL has not explicitly stated its involvement in designing new nuclear weapons, officials confirm that the system will be used in “the absence of underground testing.” Additionally, the supercomputer will support research in high-energy-density physics, materials science, and other critical national security applications. With its unprecedented speed and processing power, El Capitan is poised to shape the future of scientific discovery and defense technology.
