21 November 2025
A research team involving the Jülich Supercomputing Centre (JSC) has won the prestigious Gordon Bell Prize in the category of climate modelling. The Gordon Bell Prizes are among the most important awards in the field of high-performance computing. The jury was convinced by the team's novel approach, which enables the simulation of the Earth system with an unprecedented level of spatial detail. To achieve this, the team used Europe's first exascale supercomputer, JUPITER at Forschungszentrum Jülich, as well as the Swiss supercomputer Alps-both of which rank among the fastest systems in the world.
At the heart of the work lies an enhanced version of the ICON climate model that incorporates the atmosphere, oceans, land surfaces and sea ice, and operates at a horizontal resolution of approximately 1.25 kilometres. For comparison, climate simulations with the ICON model have typically operated at a resolution of around 100 kilometres until now. This new level of detail enables key processes, such as convection, extreme precipitation and ocean eddies, to be represented far more realistically.
This was made possible by algorithmic innovations and the targeted use of JUPITER's powerful NVIDIA GH200 hardware, which was used as part of the JUPITER Early Access Program (JUREAP). The programme gives researchers access to the system during its early testing phase.
The team, comprising members from the Max Planck Institute for Meteorology (MPI-M), the German Climate Computing Centre (DKRZ), the Jülich Supercomputing Centre (JSC), the Swiss National Supercomputing Centre (CSCS), ETH Zurich and NVIDIA, succeeded in fully exploiting the capabilities of NVIDIA's GH200 superchips. These combine a Grace CPU and a Hopper GPU into a single, large superchip.
This simulation is one of the first model runs to use the Grace CPU and Hopper GPU simultaneously for different, yet closely linked, model components. This marks a crucial step toward making efficient use of JUPITER's heterogeneous exascale architecture and enabling ever more complex and accurate climate simulations.
For the calculations, ICON was scaled to 5,120 of JUPITER's compute nodes. According to current knowledge, these runs are among the largest and most technically demanding simulations conducted on the system to date. From JUPITER's first development module, JEDI, through to the full system, the JSC configured the supercomputer for the ICON code and prepared JUPITER for the simulations from the outset. The JSC also supported the optimisation of ICON for the hybrid Grace-Hopper architecture by conducting tests and evaluating performance, and played a key role in scaling the simulations to the exascale level.
The award ceremony took place at the Supercomputing Conference (SC25) in St. Louis. The team also received an HPCwire Award in recognition of their work demonstrating how the next generation of supercomputers can improve our understanding of global climate dynamics.
Funding
JUPITER is funded jointly, with half of its funding being provided by the European High Performance Computing Joint Undertaking (EuroHPC JU), a quarter coming from the Federal Ministry of Research, Technology and Space (BMFTR, formerly BMBF), and a quarter from the Ministry of Culture and Science of the State of North Rhine-Westphalia (MKW NRW).