Europe's First Million-Trillion-FLOPS Supercomputer JUPITER Ranks 4th, Opening New Horizons in Scientific Computing}

Editor | Baicai Ye

If you want to describe the airflow on Earth's surface per square meter, you need to map nearly 30 trillion square kilometers of land, ocean, and space. To do this, you need a supercomputer of immense scale.

For this reason, computational scientist Ioan Hadade, who works on weather forecasting and climate models, is excited about this machine just an hour's drive from his lab in Bonn, Germany. Europe's first exascale supercomputer—JUPITER—has nearly been fully operational. It can run scientific programs on its powerful processors.

JUPITER made its debut on the TOP500 list of the world's most powerful supercomputers in June 2025, ranking fourth.

Located at the Jülich Supercomputing Center between Cologne and Aachen in Germany, it operates on an enhanced module with 5,900 acceleration nodes. About 24,000 Nvidia Grace-Hopper chips power JUPITER; it also includes a general cluster with 1,300 nodes using Rhea processors and an InfiniBand NDR network for high-speed interconnects.

The semiannual TOP500 ranking assesses the performance of each component of these machines. Benchmark tests demonstrate their capability for highly complex operations. Thomas Lippert, director of the Jülich Supercomputing Center, said, “Now, it’s best to conduct scientific research on these machines.”

Large-Scale Scientific Computing

By mid-June, research institutions have begun using JUPITER for scientific computing. “You need a very large machine to run these simulations,” said Hadade, referring to their participation in the European Center for Medium-Range Weather Forecasts’ “Destination Earth” digital twin project.

Destination Earth: https://destination-earth.eu/

The digital twin of Earth, used for monitoring, simulating, and predicting natural phenomena and human activities, has a resolution of over 9 kilometers based on physics-based observations of atmospheric conditions and weather. The team is using JUPITER to reveal processes that trigger deep convection and turbulence, capable of visualizing physical phenomena at 700 meters or finer resolution.

At Ilmenau University of Technology, physicist Jörg Schumacher visualizes complex thermal plumes and turbulence, which can be found in clouds or on the Sun’s surface during intense activity. Convection, driven by temperature differences, becomes highly turbulent when the temperature gradient is large.

“Everything should be very chaotic, irregular, and random—yet it’s not,” Schumacher said. “How does nature create these beautiful patterns amid turbulence?”

Schumacher’s team visualized these phenomena on supercomputers, revealing astonishing details. This work was done on the JUWELS module at Jülich, which can no longer handle larger tasks. They plan to use JUPITER for further exploration.

Hadade emphasized that the ability to access these machines at any time is more important than the ranking. “This shows Europe’s serious commitment to high-performance computing,” he said.

Energy Issues in Supercomputing

The plan for JUPITER’s exascale computing began in 2018 with the European High-Performance Computing Joint Undertaking (EuroHPC JU). The project has advanced during the pandemic: in 2024, the JEDI prototype and JETI transition system were put into use, and JUPITER was fully installed by late April this year. Europe is building a second exascale system, the Alice Recoque, costing €544 million, located in Bure, southern Paris.

These ambitious projects are not only about scale and speed but also focus on energy consumption and cooling efficiency. JUPITER’s cooling system recycles water from the nearby Ruhr River and uses excess heat in winter to warm campus buildings.

JUPITER’s prototype JEDI ranks 21st on the GREEN500 list, which measures energy efficiency. Lippert noted that the difference between JEDI and JUPITER is scale: moving to larger machines reduces parallelization efficiency. As JUPITER is optimized, its efficiency is expected to improve; it already ranks among the top five fastest systems with the highest energy efficiency, performing over 60 billion floating-point operations per watt.

Despite this, JUPITER consumes about 17 MW of power at peak load. As more large data centers are built to meet AI and tech demands, uncontrolled energy consumption becomes a pressing issue.

Researchers like Schumacher and Hadade see enough convection and temperature anomalies in their visualizations. “Any measures to reduce environmental impact are welcome,” Hadade said. “We need to be careful not to build nuclear power plants nearby to power these devices.”

Related article: https://spectrum.ieee.org/jupiter-exascale-supercomputer-europe