El Capitan by Lawrence Livermore National Laboratory: status and timeline

El Capitan

World's 2nd-fastest supercomputer at 1.809 exaflops. 11M cores, 44,000+ AMD MI300A GPUs. 100% fanless liquid cooling, 58.9 GFlop/W.

DeployStage 4 of 5

Operational November 2024. Ranked 2nd on TOP500 June 2026.

Updated Jun 2025Checked 10 Oct0 updates this week

Milestones

Next · Stockpile stewardship and research campaigns

El Capitan procurement announcedAug 2019Complete.
El Capitan becomes third U.S. exascale supercomputerNov 2024Complete.
El Capitan becomes world's fastest supercomputerNov 2024Complete.
El Capitan dominates TOP500 benchmarksJun 2025Complete.
Stockpile stewardship and research campaignsTarget 2026Current milestone.

Most important updates

Upcoming

2026Stockpile stewardship and research campaigns (next)

Current obstacles

Code porting and optimizationLegacy scientific codes must be refactored for GPU acceleration. Porting Fortran and C code requires domain expertise.
Memory bottlenecks for AI workloadsMoving massive datasets through memory hierarchy remains challenging for machine learning tasks.

Physics limits

Power density and cooling constraintsA 30 MW system requires significant electrical infrastructure and cooling capacity. Future scaling will hit physical limits in power delivery and heat dissipation at the facility.
GPU memory bandwidth versus compute ratioModern GPUs can perform vast amounts of floating-point arithmetic faster than data can be fed to them. Memory bandwidth becomes the bottleneck for bandwidth-limited algorithms.

How it works

3 parts

CPU and GPU hybrid design

AMD EPYC and Instinct partnership

EPYC Genoa CPUs paired with MI300A accelerators. CPUs handle serial work and memory tasks; GPUs accelerate floating-point simulations.

Fanless cooling

Direct liquid cooling system

100% fanless HPE direct liquid cooling eliminates mechanical failures and improves efficiency.

Data and bandwidth

High-speed interconnect

High-speed interconnects move data between CPU nodes, GPU accelerators, and memory optimized for computational capacity.

Spec sheet

Spec	El Capitan
HPL performance	1.809 exaflops^R (reported)
Total cores	11,039,616^R (reported)
AMD Instinct MI300A accelerators	44,000+^R (reported)
Memory capacity	5.4375 PB^R (reported)
Power consumption	30 MW^R (reported)
Energy efficiency	58.9 gigaflops per watt^R (reported)
TOP500 rank (June 2026)	2nd^R (reported)

^R reported by the company

Update log

2 updates

Jun 2025

Major: El Capitan dominates TOP500 benchmarks, leading on HPL, HPCG, and HPL-MxP metricsPressTest

Nov 2024

Major: El Capitan at Lawrence Livermore becomes world's fastest supercomputer, breaking 1.8 exaflopsPressHardware

About Lawrence Livermore National Laboratory

The team behind El Capitan

Lawrence Livermore National Laboratory

El Capitan exascale supercomputer, NNSA, nuclear security

A U.S. National Laboratory operated for the Department of Energy's NNSA. LLNL operates El Capitan, ranked 2nd on TOP500, supporting nuclear stockpile stewardship and scientific research.

Operational November 2024 with 1.809 exaflops. Ranked 2nd on June 2026 TOP500.
Built by HPE with AMD EPYC CPUs and Instinct MI300A accelerators; 100% fanless direct liquid cooling.
Primary mission: Stockpile Stewardship Program ensuring U.S. nuclear arsenal safety and effectiveness.

llnl.gov (opens llnl.gov)Company profile →

Founded: 195274 yrs
Headquarters: United States
Status: Government
Valuation: government
Works in: ComputingAI factories & supercomputers
Coverage: 2 programs · 4 updateslatest 1 Jun 2025checked 10 Oct
Customers: National Nuclear Security Administration (opens energy.gov)
Partners: Los Alamos National Laboratory (opens lanl.gov)Sandia National Laboratories (opens sandia.gov)
Suppliers: Hewlett Packard Enterprise (opens hpe.com)Advanced Micro Devices (opens amd.com)
People: Lawrence Livermore National Security, LLC (opens llnl.gov)Operating contractor

Spec

El Capitan

HPL performance

1.809 exaflops^R (reported)

Total cores

11,039,616^R (reported)

AMD Instinct MI300A accelerators

44,000+^R (reported)

Memory capacity

5.4375 PB^R (reported)

Power consumption

30 MW^R (reported)

Energy efficiency

58.9 gigaflops per watt^R (reported)

TOP500 rank (June 2026)

2nd^R (reported)