ACM Gordon Bell Prize Awarded for Tsunami Prediction Simulation (www.acm.org)

An eight-member US team won the 2025 ACM Gordon Bell Prize for building a real-time “digital twin” for tsunami early warning that couples full-physics Bayesian inversion with extreme-scale HPC. Their framework ingests sensor data to perform PDE-constrained Bayesian inference of earthquake rupture and tsunami generation, enabling data-driven forecasts that adapt to complex source behavior and address limits of seismic-only methods that often leave under ten minutes for coastal warning. Technically, the team solved the fastest time-to-solution reported for a PDE-based Bayesian inverse problem: 1 billion parameters in 0.2 seconds — a ~10-billion× speedup over prior state-of-the-art. They ran the largest-to-date unstructured mesh finite-element simulation (55.5 trillion degrees of freedom) on 43,520 GPUs of the El Capitan system, demonstrating 92% weak and 79% strong parallel efficiency across a 128× GPU scaling. The test case modeled the Cascadia Subduction Zone, a 1,000 km megathrust that threatens the US Pacific Northwest. Beyond tsunami warning, this result showcases that real‑time, physics‑aware digital twins and extreme-scale Bayesian inference are now tractable for urgent geophysical hazards and other time-sensitive scientific applications.