Researchers at Lawrence Livermore National Laboratory (LLNL), the University of Texas at Austin's (UT) Oden Institute and Scripps Institution of Oceanography at the University of California San Diego (UCSD) on Nov. 20 were awarded the prestigious 2025 Association for Computing Machinery (ACM) Gordon Bell Prize for developing a real-time tsunami early-warning framework powered by the world's fastest supercomputer, El Capitan.
Widely viewed as the highest recognition in high-performance computing (HPC), the Gordon Bell Prize recognizes innovations that push the limits of computational performance, scalability and scientific impact on pressing real-world problems. The Prize was announced at the International Conference for High Performance Computing, Networking, Storage, and Analysis (SC25) in St. Louis. In their winning paper, the team demonstrated for the first time that true real-time, physics-based tsunami forecasting is computationally achievable at exascale, opening a path toward faster and more reliable early-warning systems for coastal communities.
"For the first time, we can combine real-time sensor data with full-physics modeling and uncertainty quantification - fast enough to make decisions before a tsunami reaches the shore," said Omar Ghattas, professor of mechanical engineering and principal faculty at the Oden Institute at UT and senior member of the team. "It is a tremendous honor for our team to receive this highest recognition in the field of supercomputing."
The team approached the challenge by devising innovative supercomputing algorithms for digital twins that infer seafloor motion from real-time pressure sensor data. The seafloor motion is then used to forecast the destructive tsunami waves. The entire framework executes in a fraction of a second - 10 billion times faster than conventional algorithms. A key to the building of the high-fidelity digital twin is the precomputation of a set of full-physics wave simulations.
LLNL's exascale El Capitan, which retained its title at SC25 as the most powerful supercomputer ever benchmarked, according to the Top500 organization, was instrumental in resolving the wave simulations to the full margin of the Cascadia subduction zone in the Pacific Northwest.
For co-author and LLNL computational mathematician Tzanio Kolev, the project represents the culmination of years of research in numerical methods, high-order modeling and exascale computing by the MFEM project at LLNL.
"Our goal was to show that physics-based simulation can deliver near-instant results for critical applications like tsunami warnings," Kolev said. "By combining real-time sensor data with exascale modeling, we demonstrated a practical path toward next-generation warning systems. Working with our amazing partners at UT Austin and Scripps, we showed how fast and accurate these forecasts can be when you have the right algorithms and the right machine."
Tsunamis remain among the world's most devastating natural hazards, and warning centers often have only minutes to determine whether an offshore earthquake will generate a life-threatening wave. Existing systems rely on simplified approximations, sparse instrumentation and historical patterns that may not capture the complex physics of tsunami generation and propagation. These limitations can lead to false alarms, underestimated risks or delays that reduce the time available for evacuation.
El Capitan's computational power supported the largest-ever unstructured-mesh finite-element simulation to date, resolving 55.5 trillion degrees of freedom with LLNL's open source MFEM finite element library, developed by Kolev and his Lab team. The researchers leveraged this capability to reconstruct seafloor motion, infer earthquake characteristics and forecast tsunami arrival times and wave heights in a matter of seconds, including quantified confidence estimates for emergency managers.
UT's Oden Institute led the project with major innovations in inverse algorithms that enabled the real-time inference of seafloor deformation from offshore pressure sensor data. Their work ensured that the physics-based reconstructions remained stable, accurate and consistent with real-time signals.
"The biggest technical challenge in creating the Cascadia digital twin was the enormous problem size combined with the need for providing forecasts in real time," said Stefan Henneking, research associate at the Oden Institute. "Overcoming this challenge required novel parallel inversion algorithms, advanced numerical discretization libraries and exploitation of leading-edge GPU supercomputers, including El Capitan and the National Energy Research Scientific Computing Center's (NERSC's) Perlmutter."
The Scripps Institution of Oceanography (Scripps) provided essential expertise in tsunami science, ocean-floor dynamics and the interpretation of deep-ocean sensor data. This work ensured that the modeling pipeline aligned with what real-world warning centers require in an actual emergency, creating the first true tsunami digital twin linking the full physics of megathrust dynamic rupture and tsunami generation with real-time seafloor observations.
"Our work shows that physics-based models of earthquakes and tsunami generation are now fast enough to guide real-time response, which not only improves early warning for the Pacific Northwest, but also points the way toward global, physics-based earthquake and tsunami early-warning systems," said Alice-Agnes Gabriel, an earthquake seismologist and associate professor at the Institute of Geophysics and Planetary Physics at Scripps.
The team's approach is grounded in physics rather than purely statistical approximations. When an offshore earthquake occurs, real-time pressure sensors deployed across the ocean floor provide immediate readings of wave propagation. By rapidly solving the inverse problem, the system can infer the seafloor displacement scenario that most closely matches pressure sensor data and generate a physics-consistent tsunami forecast in seconds. The result is not only faster but significantly more reliable, providing emergency agencies with actionable information and clear estimates of uncertainty at a moment when every second matters.
The achievement was made possible by El Capitan's architecture, which integrates AMD Instinct MI300A Accelerated Processing Units at unprecedented scale. With more than 46,000 APUs and more than 11 million CPU cores, El Capitan provides the compute power required to carry out the massive wave simulations underlying the digital twin. The machine's memory bandwidth, energy efficiency and heterogeneous design all contributed to the project's breakthrough performance.
The ACM Gordon Bell Prize has recognized some of the most influential achievements in computational science since 1987. By joining this lineage, the collaboration has demonstrated that exascale computing is not only a scientific milestone but also an engine for solving urgent global challenges. The team's work is a demonstration of how high-fidelity modeling, advanced numerical methods and state-of-the-art hardware can converge to produce technology with immediate societal benefit.
The researchers plan to continue refining their method, exploring how it could be integrated with operational tsunami warning centers and expanded to cover a broader range of geophysical hazards. They also intend to extend the modeling approach to additional regions, improve the fidelity of the digital twin and explore machine-learning accelerations that remain fully grounded in physical principles.
"Winning the Gordon Bell Prize is an extraordinary recognition of what this team has achieved," said LLNL Weapon Simulation and Computing Associate Director Rob Neely. "This project highlights the versatility of El Capitan as a tool for real-world impact and shows what happens when computational innovation meets real-world urgency. It is a glimpse of what exascale computing can achieve as we continue pushing the boundaries of what is possible."
Other co-authors included Veselin Dobrev and John Camier of LLNL, and Sreeram Venkat and Milinda Fernando of the UT Oden Institute.
