El Capitan (supercomputer)
| Company type | Supercomputer |
|---|---|
| Industry | Computing / High Performance Computing |
| Founded | 2024 |
| Headquarters | Livermore , United States |
Area served | United States (Lawrence Livermore National Laboratory) |
Key people | Antonio Neri (HPE CEO), Lisa Su (AMD CEO) |
| Products | Exascale supercomputer |
| Owner | Lawrence Livermore National Laboratory, Hewlett Packard Enterprise, AMD |
Hewlett Packard Enterprise El Capitan is an exascale supercomputer located at the Lawrence Livermore National Laboratory in Livermore, California, having been operational since 2024. Utilizing the Cray EX Shasta framework, El Capitan overtook Frontier as the fastest supercomputer worldwide in November 2024. It represents the third exascale machine within the United States and primarily supports the stockpile stewardship efforts of the US National Nuclear Security Administration.
Design
The architecture of El Capitan includes a total of 11,039,616 CPU and GPU cores, comprising 43,808 AMD fourth-generation EPYC 24-core "Genoa" processors (amounting to 1,051,392 cores) alongside 43,808 AMD Instinct MI300A GPUs (providing 9,988,224 compute units), which together deliver 639 million stream processors. The MI300A combines 24 Zen4 CPU cores with a CDNA3 GPU, supported by 128GB of HBM3 memory.
Connectivity among blades is achieved through an HPE Slingshot 64-port switch that offers 12.8 Tbps bandwidth arranged in a dragonfly network topology, permitting up to three hops between nodes. The cabling infrastructure is optimized for minimal length and utilizes both optical and copper cables.
El Capitan adopts an APU design strategy where CPU and GPU components share a coherent on-chip interconnect.
The entire system occupies approximately , equivalent to the area of two tennis courts. It comprises at least 87 compute racks, incorporating "Rabbit" NVM-Express high-speed storage arrays alongside compute nodes. According to The Next Platform, there are 11,136 liquid-cooled Cray EX nodes, each equipped with four MI300A processors (totaling 44,544 devices). Each processor contains 128 GB of HBM3 memory operating at 5.2 GHz with a shared bandwidth of 5.3 TB/sec between its CPU and GPU chiplets.
History
The development of El Capitan was initiated under the Department of Energy’s CORAL-2 program as the successor to Sierra, an IBM/Nvidia supercomputer deployed in 2018. The project was a collaboration between Lawrence Livermore National Laboratory (LLNL), HPE Cray, and AMD.
In June 2023, three El Capitan prototypes—rzVernal, Tioga, and Tenaya—were listed on the TOP500 rankings; rzVernal achieved performance levels of up to 4.1 petaflops. Installation at LLNL commenced in early July 2023 with full operational status anticipated by mid-2024.
By November 18, 2024, El Capitan was fully functional and confirmed as the fastest supercomputer globally, reaching a peak performance of 1.742 exaFLOPs. The official launch took place at LLNL in California in February 2025 with a total investment of $600 million. This system is dedicated to highly sensitive and classified operations supporting the security of the U.S. nuclear weapons stockpile.
Dedication
The formal dedication ceremony for El Capitan occurred on January 9, 2025, attended by Hewlett Packard Enterprise (HPE) CEO Antonio Neri and Advanced Micro Devices (AMD) CEO Lisa Su. Both executives underscored El Capitan’s critical role in advancing their companies’ artificial intelligence agendas. Neri highlighted the "full synergy" between El Capitan and AI training platforms, while Su described the technology as adaptable—"the same building blocks... arranged differently"—to drive AI progress effectively.