ICON Simulations Nominated for Prestigious Gordon Bell Climate Modeling Prize

The WarmWorld project is proud to announce that groundbreaking global Earth system simulations, powered by developments within our consortium, have been nominated for the ACM Gordon Bell Prize for Climate Modelling. This recognition is a major leap forward in climate science: the first-ever simulation of the full Earth system at a horizontal grid spacing of just 1.25 km — a endeavor long considered out of reach.

This achievement was made possible by the close collaboration of the Max Planck Institute for Meteorology (MPI-M), the German Climate Computing Center (DKRZ), and the Jülich Supercomputing Centre (JSC, Forschungszentrum Jülich) — bringing together the partnerships built within the BMFTR projects WarmWorld and IFCES2-Scalexa, as well as the international collaborations with ETH Zurich, CSCS, and NVIDIA. The work pushes the limits of both climate modeling and high-performance computing, harnessing more than 8,000 NVIDIA Grace Hopper GH200 superchips on the Alps supercomputer at CSCS.

A New Era in Climate Simulation

At the heart of the effort is ICON, the atmosphere–ocean–land–biogeochemistry modeling framework co-developed by MPI-M, DKRZ, DWD, KIT and a Swiss consortium lead by MeteoSwiss and ETH Zürich. For the first time, ICON has been configured to couple the cycles of water, energy, and carbon globally at kilometer scale, enabling explicit representation of small-scale phenomena — from afternoon thunderstorms to ocean eddies — and their interactions with long-term carbon storage and climate change.

Achieving this required innovations in how the model’s diverse components are mapped onto heterogeneous computing architectures. By running ocean and biogeochemistry processes on the Grace CPUs while the atmosphere and land run on the Hopper GPUs, the team ensured all parts of the model could run concurrently, squeezing maximum performance from the system. Additional breakthroughs included using CUDA Graphs to accelerate small land-surface kernels, and applying the DaCe data-centric optimization framework to improve performance without compromising code maintainability.

Performance and Impact

The production configuration comprises nearly a trillion degrees of freedom — over 7.8×10¹¹ physical-spatial variables — and reached an unprecedented temporal compression of 82.5 simulated days per day on Alps. This performance allows scientists to conduct long, fully coupled Earth system experiments that were previously impractical, such as multi-decadal simulations of climate change impacts at local scales.

Beyond raw speed, the scientific implications are profound. For the first time, global models can directly resolve how small-scale processes in the atmosphere, ocean, and biosphere influence — and are influenced by — carbon fluxes across the Earth system. This opens the door to new insights into feedbacks, tipping points, and regional climate change impacts, providing critical information for adaptation and mitigation strategies.

A WarmWorld Milestone

These simulations build on the experience gained in WarmWorld to enable kilometer-scale Earth system modeling on the next generation of supercomputers. The nomination for the Gordon Bell Prize for Climate Modeling is a recognition not just of a technical tour de force, but of the collaborative framework that made it possible.

The award will be announced at the SC25 supercomputing conference in November. Regardless of the outcome, the work already stands as a landmark — demonstrating that with the right mix of science, computing, and collaboration, climate modeling can enter a new era of detail, realism, and societal relevance.