GPU supercomputers: the largest public machines by GPU count
32 clusters. 26 are running or partly running. Installed GPUs on record add up to 238,448, with a further 122,496 planned across 4 clusters. The largest is Argonne Solstice (100,000 GPUs planned).
National labs, research centres and academic computing consortia publish exact node and GPU counts, so this group has the best-documented numbers on the list.
Last verified
32 clusters
Full sortable table| Cluster | Owner / operator | GPU model | GPUs installed | GPUs planned | Location | Status |
|---|---|---|---|---|---|---|
| Aurora | Argonne Leadership Computing Facility | Intel Data Center GPU Max Series | 63,744 | — | Lemont, Illinois, United States | operational |
| El Capitan | Lawrence Livermore National Laboratory | AMD Instinct MI300A APU | 46,080 | — | Livermore, California, United States | operational |
| Frontier | Oak Ridge Leadership Computing Facility | AMD Instinct MI250X | 39,424 | — | Oak Ridge, Tennessee, United States | operational |
| Leonardo Booster | CINECA | NVIDIA A100-64 custom Ampere, 64 GiB | 13,824 | — | Bologna, Italy | operational |
| LUMI-G | LUMI consortium | AMD Instinct MI250X | 11,912 | — | Kajaani, Finland | operational |
| Alps Grace Hopper partition | CSCS | NVIDIA GH200 (H100 GPU) | 10,752 | — | Lugano and Lausanne, Switzerland | operational |
| Perlmutter GPU partition | NERSC | NVIDIA A100 40GB and 80GB | 7,168 | — | Berkeley, California, United States | operational |
| ABCI 3.0 | AIST Solutions (ABCI) | NVIDIA H200 SXM5, 141 GB HBM3e | 6,128 | — | Kashiwa, Japan | operational |
| Isambard-AI | University of Bristol | NVIDIA GH200 Grace Hopper | 5,448 | — | Bristol, United Kingdom | operational |
| Tuolumne | Lawrence Livermore National Laboratory | AMD Instinct MI300A APU | 4,608 | — | Livermore, California, United States | operational |
| MareNostrum 5 ACC | Barcelona Supercomputing Center | NVIDIA Hopper, 64 GB HBM | 4,480 | — | Barcelona, Spain | operational |
| Horizon (TACC) | Texas Advanced Computing Center (UT Austin, NSF) | NVIDIA GB200 (NVL4 nodes) | 4,000 | — | Round Rock, Texas, United States | partly operational |
| JUWELS Booster | Jülich Supercomputing Centre | NVIDIA A100 40 GB | 3,744 | — | Jülich, Germany | operational |
| Shaheen III GPU partition | KAUST | NVIDIA GH200 Grace Hopper | 2,800 | — | Thuwal, Saudi Arabia | operational |
| Venado | Los Alamos National Laboratory | NVIDIA GH200 Grace Hopper | 2,560 | — | Los Alamos, New Mexico, United States | operational |
| Polaris | Argonne National Laboratory | NVIDIA A100 | 2,240 | — | Lemont, Illinois, United States | operational |
| Gefion | Danish Centre for AI Innovation (DCAI) | NVIDIA H100 | 1,528 | — | — | operational |
| Jean Zay H100 extension | IDRIS / CNRS (GENCI) | NVIDIA H100 80 GB | 1,456 | — | Saclay, France | operational |
| Adastra GPU partition | CINES | AMD Instinct MI250X | 1,352 | — | Montpellier, France | operational |
| MareNostrum 5 AI training partition | Barcelona Supercomputing Center | NVIDIA GB200 | 1,344 | — | Barcelona, Spain | operational |
| TSUBAME4.0 | Institute of Science Tokyo | NVIDIA H100 SXM5 | 960 | — | Tokyo, Japan | operational |
| MeluXina accelerator module | LuxProvide | NVIDIA A100 40GB | 800 | — | Bissen (LuxConnect DC2), Luxembourg | operational |
| Setonix AMD GPU partition | Pawsey Supercomputing Centre | AMD Instinct MI250X | 768 | — | Kensington, Australia | operational |
| Berzelius Ampere | NAISS | NVIDIA A100, 40GB or 80GB | 752 | — | Linköping, Sweden | operational |
| MareNostrum 5 AI inference partition | Barcelona Supercomputing Center | NVIDIA GB300 | 576 | — | Barcelona, Spain | operational |
| Argonne Solstice | Argonne National Laboratory (DOE) | NVIDIA Blackwell | — | 100,000 | Lemont, Illinois, United States | announced |
| Argonne Equinox | Argonne National Laboratory (DOE) | NVIDIA Blackwell | — | 10,000 | Lemont, Illinois, United States | announced |
| KISTI National Supercomputer No. 6 | KISTI (Korea Institute of Science and Technology Information) | NVIDIA GH200 | — | 8,496 | — | under construction |
| Lux (ORNL) | Oak Ridge National Laboratory (DOE) | AMD Instinct MI355X, 288 GB HBM3E | — | 4,000 | Oak Ridge, Tennessee, United States | under construction |
| Alice Recoque | Jules Verne consortium (GENCI, CEA, SURF, GRNET) for EuroHPC JU | AMD Instinct MI430X, 432 GB HBM4 | — | — | TGCC, CEA DAM Île-de-France, Bruyères-le-Châtel, France | under construction |
| Doudna (NERSC-10) | NERSC, Lawrence Berkeley National Laboratory (DOE) | NVIDIA Vera Rubin | — | — | Berkeley, California, United States | announced |
| JUPITER Booster | Jülich Supercomputing Centre (EuroHPC JU) | NVIDIA GH200 | — | — | Jülich, Germany | operational |