The world of supercomputers has entered the era of exascale computing. The June 2025 TOP500 ranking recorded three American systems at the top, a sensational debut from Europe, and a notable presence of cloud and industrial installations in the top ten. High-performance computing now concerns not only national laboratories, but also commercial clouds and industrial centers.
Supercomputer performance is measured in floating-point operations per second (FLOPS). In the TOP500 ranking, systems are ranked according to the LINPACK (HPL) test, where the key indicator Rmax reflects the sustainable speed in solving a large system of linear equations. Today, the leaders demonstrate a stable level in exaflops in this rigorous test.
The top three from the USA
El Capitan (USA, Lawrence Livermore National Laboratory, Rmax 1,742 exaflops). The world’s most powerful supercomputer. Built on the HPE Cray EX255a architecture with 4th-generation AMD EPYC processors and Instinct MI300A accelerators connected via a Slingshot-11 network. It leads not only the LINPACK test, but also the more complex HPCG test, demonstrating balanced performance for real-world scientific tasks. Since the end of 2024, El Capitan has held the top spot, cementing exaflops as the standard for American labs.
Frontier (USA, Oak Ridge, Rmax 1,353 exaflops). The first exaflop system in history, which remains the benchmark. It uses HPE Cray EX235a cabinets with AMD EPYC processors and Instinct MI250X graphics accelerators, combined via Slingshot-11.From 2022 to 2024, it occupied the top spot, but has now moved to second. This shift in position does not reflect a decline, but the pace of progress in the industry. Frontier continues to push the limits of its capabilities in energy, materials, biology, and astrophysics research.
Aurora (USA, Argonne National Laboratory, Rmax 1,012 exaflops). The third exaflop-class system in the US. Built on the HPE Cray EX platform with Intel Xeon CPU Max processors and Intel Data Center GPU Max accelerators, connected via Slingshot-11. After many years of gradual introduction, Aurora has secured third place. Its aim is to combine modeling with the use of artificial intelligence in science: from research in thermonuclear fusion and climate to large-scale experiments with language models.
Italy in sixth and tenth place
HPC6 (Italy, Eni Green Data Center, Rmax 477.9 petaflops). Leading-edge industrial system. Built on the HPE Cray EX235a with AMD EPYC processors and Instinct MI250X GPUs, Slingshot-11 network. Its primary objectives are seismic exploration, field modeling, and low-carbon energy research. At the end of 2024, the system briefly became the European leader, but in 2025 it lost out to JUPITER Booster, rising to sixth place.
Leonardo (Italy, EuroHPC / CINECA, Rmax 241.2 petaflops). BullSequana XH2000 modular system with Intel Xeon Platinum 8358 processors and NVIDIA A100 GPUs, HDR100 InfiniBand network. Equipped with efficient liquid cooling. In 2025, it remains in the top ten, despite competition from new Grace-Hopper and MI300A-based systems. This indicates that a successful architecture has remained competitive for several consecutive years.
The rest of the top 10
JUPITER Booster (Germany, EuroHPC / Juelich, Rmax 793.4 petaflops). Europe’s new pride. Built on a BullSequana XH3000 with NVIDIA GH200 Grace-Hopper hybrid processors and a quad-channel InfiniBand NDR200 network, this accelerator module is part of the modular JUPITER architecture. The system immediately took first place in Europe and established itself among the global elite, underscoring the rapid progress of the EuroHPC program.
Fugaku (Japan, RIKEN R-CCS, Rmax 442.0 petaflops). A Fujitsu ARM system with A64FX processors and a Tofu-D network. A 2020-2021 leader, it is currently proving strong in the HPCG test, making it particularly effective for tasks with high memory and communication loads. Despite the decline in rankings, the scientific return remains enormous.
Alps (Switzerland, CSCS, Rmax 434.9 petaflops). One of the most versatile next-generation European systems. Based on HPE Cray EX254n with NVIDIA Grace processors and GH200 GPUs, combined via Slingshot-11. This platform reflects the modern trend: combining AI training and physics models on the same hardware.
LUMI (Finland, EuroHPC / CSC, Rmax 379.7 petaflops). European system on HPE Cray EX235a with AMD EPYC and Instinct MI250X GPU. It runs on renewable energy and uses heat recovery. From 2022 to 2024, it was the most important project in Europe, but with the advent of Alps and JUPITER Booster, it dropped to ninth place. It remains a key tool for climatology, materials science, and big data analysis.
The next frontier in the race is the zettaflop level, around a thousand exaflops. Japan is already working on the Fugaku Next project, scheduled for launch in 2030. It could change the global balance of power. However, the full picture is still unclear: China no longer publishes data for the TOP500, and the true scope of its systems is unknown. This adds to the intrigue: who will be the first to reach the zettaflop level remains an open question.
Redazione The editorial team of Red Hot Cyber consists of a group of individuals and anonymous sources who actively collaborate to provide early information and news on cybersecurity and computing in general.
Redazione