The ten most energy-efficient supercomputers in the world, ranked by GFlops per watt on the November 2025 Green500 list, led by France's KAIROS at 73.28 GFlops/W.
KAIROS is the most energy-efficient supercomputer in the world, at 73.28 GFlops/W. ROMEO-2025 is second at 70.91 GFlops/W and the Levante GPU extension third at 69.43 GFlops/W. The ranking uses energy efficiency as published on the November 2025 Green500 list: HPL benchmark performance divided by the power measured during the run. All three kept their places in the June 2026 edition.
The Green500 re-ranks the systems on the TOP500 list not by raw speed but by how much computation they deliver per watt of electricity, measured in gigaflops per watt (GFlops/W). This list is about efficiency, not raw power: the fastest machine in this top ten runs at 24.50 PFlop/s, far below the leaders of the TOP500 speed ranking. This list shows the top ten from the November 2025 Green500. The top three are all Eviden BullSequana XH3000 systems built on NVIDIA GH200 Grace Hopper Superchips, led by KAIROS at CALMIP in Toulouse, France. The ten systems are in France, Germany, the United Kingdom, Poland, South Korea and the United States.
A BullSequana XH3000 system at CALMIP / University of Toulouse - CNRS, KAIROS was the new No. 1 on the November 2025 Green500 at 73.28 GFlops/W. It uses NVIDIA GH200 Grace Hopper Superchips and ran the HPL benchmark at 3.05 PFlop/s.
ROMEO-2025 uses the same BullSequana XH3000 Grace Hopper design as KAIROS but is about three times larger, at 9.86 PFlop/s on HPL, and slightly less efficient at 70.91 GFlops/W.
The GPU extension of Levante at DKRZ, Germany's climate computing centre, is the third BullSequana XH3000 Grace Hopper system in the top three, at 69.43 GFlops/W and 6.75 PFlop/s on HPL.
Isambard-AI phase 1 at the University of Bristol is an HPE Cray EX254n system with NVIDIA GH200 Grace Hopper Superchips and Slingshot-11 interconnect, at 68.84 GFlops/W and 7.42 PFlop/s on HPL.
The GPU partition of Otus at Paderborn University's PC2 centre is a Lenovo system with AMD EPYC 9655 processors and NVIDIA H100 GPUs, at 68.18 GFlops/W and 4.66 PFlop/s on HPL.
Capella at TU Dresden is a MEGWARE-integrated Lenovo system with AMD EPYC processors and NVIDIA H100 GPUs. At 24.06 PFlop/s on HPL it is the second-fastest machine in this top ten, and it reaches 68.05 GFlops/W.
The SSC-24 Energy Module at Samsung Electronics is an HPE Cray XD670 system with Intel Xeon Gold processors and NVIDIA H100 GPUs, at 67.25 GFlops/W and 3.82 PFlop/s on HPL.
Helios GPU at Cyfronet in Poland is an HPE Cray EX254n Grace Hopper system, at 66.95 GFlops/W and 19.14 PFlop/s on HPL.
AMD Ouranos, an Atos-sited BullSequana XH3000 with AMD Instinct MI300A accelerated processors, is the most efficient system on the list accelerated by AMD GPUs at 66.46 GFlops/W, with 2.99 PFlop/s on HPL.
Portage, an HPE Cray EX255a system with AMD Instinct MI300A at HPE in Chippewa Falls, Wisconsin, completes the top ten at 66.28 GFlops/W. At 24.50 PFlop/s on HPL it is the fastest machine in this top ten.
| # | Name | Site | Country | Rmax (PFlop/s) | System |
|---|---|---|---|---|---|
| 1 | KAIROS | CALMIP / University of Toulouse - CNRS | France | 3.05 | Eviden BullSequana XH3000, NVIDIA GH200 |
| 2 | ROMEO-2025 | ROMEO HPC Center, University of Reims Champagne-Ardenne | France | 9.86 | Eviden BullSequana XH3000, NVIDIA GH200 |
| 3 | Levante GPU extension | DKRZ (German Climate Computing Centre) | Germany | 6.75 | Eviden BullSequana XH3000, NVIDIA GH200 |
| 4 | Isambard-AI phase 1 | University of Bristol | United Kingdom | 7.42 | HPE Cray EX254n, NVIDIA GH200 |
| 5 | Otus (GPU only) | Paderborn University (PC2) | Germany | 4.66 | Lenovo ThinkSystem SD665-N V3, NVIDIA H100 |
| 6 | Capella | TU Dresden (ZIH) | Germany | 24.06 | Lenovo ThinkSystem SD650 V3 (MEGWARE), NVIDIA H100 |
| 7 | SSC-24 Energy Module | Samsung Electronics | South Korea | 3.82 | HPE Cray XD670, NVIDIA H100 |
| 8 | Helios GPU | Cyfronet | Poland | 19.14 | HPE Cray EX254n, NVIDIA GH200 |
| 9 | AMD Ouranos | Atos | France | 2.99 | Eviden BullSequana XH3000, AMD Instinct MI300A |
| 10 | Portage | Hewlett Packard Enterprise, Chippewa Falls, Wisconsin | United States | 24.5 | HPE Cray EX255a, AMD Instinct MI300A |
Systems are ranked by energy efficiency in GFlops/W exactly as published on the November 2025 Green500 list: the HPL (LINPACK) Rmax result divided by the power measured during that run. One edition only; Rmax figures are from the same list. Rank 1 is the most energy-efficient. A June 2026 Green500 edition has since been published; KAIROS, ROMEO-2025 and Levante GPU extension kept the top three places there. Power and core counts are not shown because they could not be checked against the list.
The Green500 re-ranks TOP500 supercomputers by energy efficiency, in GFlops per watt, instead of by raw LINPACK performance.
Gigaflops per watt: billions of floating-point operations per second for each watt of electrical power drawn. Higher means more computation per unit of electricity.
KAIROS, ROMEO-2025 and the Levante GPU extension all use the same Eviden BullSequana XH3000 design with NVIDIA GH200 Grace Hopper Superchips. Two are in France and one, Levante, is in Germany. KAIROS, the smallest of the three, is the most efficient.
More like this: all Technology rankings · browse every topic