The ten most powerful supercomputers on Earth, ranked by their measured LINPACK performance on the November 2025 TOP500 list.
Quick answer: 1. El Capitan, 2. Frontier, 3. Aurora, 4. JUPITER Booster, 5. Eagle, 6. HPC6, 7. Supercomputer Fugaku, 8. Alps, 9. LUMI, 10. Leonardo.
Twice a year, the TOP500 project ranks the world's most powerful publicly known supercomputers by how fast they can solve a giant system of linear equations — the standard yardstick for raw number-crunching muscle. This list captures the top ten from the November 2025 edition, an era defined by "exascale" machines capable of more than a quintillion calculations per second. At the top sits El Capitan, the U.S. Department of Energy system that first broke 1.8 EFlop/s, trailed by fellow American giants Frontier and Aurora and Europe's first exascale machine, JUPITER Booster. Below them, the ranking spans national labs, a cloud provider, an energy company, and research centers across the U.S., Europe, and Japan — a snapshot of who commands the most computing power on the planet.
The world's fastest supercomputer since 2024 and the first system ever to exceed 1.8 EFlop/s Rmax. This HPE Cray EX255a at Lawrence Livermore pairs AMD 4th Gen EPYC CPUs with Instinct MI300A accelerators.
The first-ever exascale system, which broke the barrier in 2022, now holds second place. It is an HPE Cray EX235a at Oak Ridge built on AMD 3rd Gen EPYC CPUs and Instinct MI250X accelerators.
The third machine to reach exascale, Argonne's Aurora is an HPE Cray EX built on Intel Xeon CPU Max and Intel Data Center GPU Max, networked over Slingshot-11.
The fourth exascale system ever built and the first in Europe. This BullSequana XH3000 from Eviden, hosted at Germany's Jülich Supercomputing Centre, runs on NVIDIA GH200 Grace Hopper Superchips.
The highest-ranked cloud-based system on the list, Microsoft's Eagle runs on Azure NDv5 instances powered by NVIDIA H100 GPUs — proof that on-demand cloud can compete with dedicated national labs.
The highest-ranked private-sector system, operated by energy company Eni. It is an HPE Cray EX235a using AMD Instinct MI250X accelerators, mirroring the architecture of Frontier.
The world's No. 1 from 2020 to 2022, RIKEN's Fujitsu-built Fugaku is unusual for the top ranks: it is CPU-only, relying on ARM A64FX processors with no separate accelerators.
An HPE Cray EX254n run by the Swiss National Supercomputing Centre, Alps is built on NVIDIA GH200 Grace Hopper Superchips.
Europe's flagship system before JUPITER, LUMI is an HPE Cray EX235a hosted at CSC in Finland and powered by AMD Instinct MI250X accelerators.
Rounding out the top ten, Leonardo is an Atos BullSequana XH2000 hosted at CINECA in Italy, built on NVIDIA A100 SXM4 64GB accelerators.
| # | Name | Location | Rpeak (PFlop/s) | Cores | Accelerator |
|---|---|---|---|---|---|
| 1 | El Capitan | DOE/NNSA/LLNL, United States | 2821.1 | 11340000 | AMD Instinct MI300A |
| 2 | Frontier | DOE/SC/Oak Ridge NL, United States | 2055.72 | 9066176 | AMD Instinct MI250X |
| 3 | Aurora | DOE/SC/Argonne NL, United States | 1980.01 | 9264128 | Intel Data Center GPU Max |
| 4 | JUPITER Booster | EuroHPC/FZJ, Germany | 1226.28 | 4801344 | NVIDIA GH200 Grace Hopper Superchip |
| 5 | Eagle | Microsoft Azure, United States | 846.84 | 2073600 | NVIDIA H100 |
| 6 | HPC6 | Eni S.p.A., Italy | 606.97 | 3143520 | AMD Instinct MI250X |
| 7 | Supercomputer Fugaku | RIKEN Center, Japan | 537.21 | 7630848 | None (ARM A64FX CPU-only) |
| 8 | Alps | Swiss NSC (CSCS), Switzerland | 574.84 | 2121600 | NVIDIA GH200 Grace Hopper Superchip |
| 9 | LUMI | EuroHPC/CSC, Finland | 531.51 | 2752704 | AMD Instinct MI250X |
| 10 | Leonardo | EuroHPC/CINECA, Italy | 306.31 | 1824768 | NVIDIA A100 SXM4 64GB |
Systems are ranked by maximal achieved LINPACK performance (Rmax) on the High-Performance Linpack (HPL) benchmark, as recorded on the TOP500 list published in November 2025. Rmax — measured in PFlop/s (petaflops, or quadrillions of floating-point operations per second) — is the exact metric TOP500 uses to order its rankings, so a higher Rmax means a higher rank. Inclusion is limited to systems appearing on the November 2025 TOP500 list; the ten highest Rmax values become ranks 1 through 10. Rpeak (theoretical peak performance), core count, and accelerator hardware are reported alongside each system for context but do not affect the order. All figures are drawn directly from the TOP500 record and confirming primary sources.
As of the November 2025 TOP500 list, the fastest supercomputer is El Capitan at the Lawrence Livermore National Laboratory in the United States, with a measured LINPACK performance (Rmax) of 1,809 PFlop/s — the first system to exceed 1.8 EFlop/s.
Rmax is the maximal performance a system actually achieves on the High-Performance Linpack (HPL) benchmark, measured in PFlop/s (quadrillions of floating-point operations per second). TOP500 ranks supercomputers strictly by Rmax, so the machine with the highest measured score takes the top spot. Rpeak, listed alongside, is the theoretical maximum a system could reach and is always higher than Rmax.
An exascale supercomputer can perform at least one exaflop — a quintillion (10^18) calculations per second, or 1,000 PFlop/s. On the November 2025 TOP500 list, four systems reached exascale on the LINPACK benchmark: El Capitan, Frontier, Aurora, and JUPITER Booster, the last of which is the first exascale machine in Europe.
The top three systems — El Capitan, Frontier, and Aurora — are all U.S. Department of Energy machines. The rest of the top ten span Germany, Italy, Japan, Switzerland, and Finland, reflecting a mix of national labs, a cloud provider, an energy company, and European research centers.