Five pioneering projects vying for the Gordon Bell Prize for outstanding high-performance computing (HPC) achievements are leveraging NVIDIA-powered supercomputers to advance breakthroughs in climate modeling, materials science, fluid dynamics, geophysics and electronic design.
Announced today at SC25, the finalists are pushing the boundaries of AI and HPC for science through physics-based simulation, high-precision computation and other cutting-edge techniques. Their work — spanning weather prediction, semiconductor design, space exploration and natural-hazard modeling — is openly available on ArXiv.
Their research runs on some of the world’s most advanced supercomputers:
- Alps — Hosted at the Swiss National Supercomputing Centre (CSCS), built with more than 10,000 NVIDIA GH200 Grace Hopper Superchips.
- Perlmutter — Hosted at NERSC and powered by NVIDIA accelerated computing.
- JUPITER — Europe’s first exascale system, hosted at JSC and built on the NVIDIA Grace Hopper platform with Quantum-X800 InfiniBand networking.
“At CSCS, we don’t just support open science — we accelerate it,” said Thomas Schulthess, director of CSCS. “The breakthroughs delivered by this year’s finalists… simply would not exist without Alps.”
ICON: Kilometer-Scale Earth System Modeling
Researchers from the Max Planck Institute for Meteorology, DKRZ, CSCS, JSC, ETH Zurich and NVIDIA developed a new configuration of the ICON Earth system model that simulates the entire planet at 1-kilometer resolution.
ICON captures interactions among the atmosphere, oceans and land with exceptional detail, enabling 146 simulated days every 24 hours — dramatically improving efficiency for long-term climate projections.
“Integrating all essential Earth-system components at 1-kilometer resolution gives researchers localized insights into future warming,” said Daniel Klocke of the Max Planck Institute for Meteorology.
ORBIT-2: Exascale AI Foundation Model for Weather & Climate
Developed by Oak Ridge National Laboratory, NVIDIA and partners, ORBIT-2 runs on the Alps supercomputer and represents a breakthrough AI foundation model for climate downscaling.
Using exascale compute and algorithmic innovations, ORBIT-2 produces hyper-resolution climate data to better capture localized extremes — from urban heat islands to subtle monsoon shifts.
“NVIDIA’s advanced supercomputing technologies enabled ORBIT-2 to achieve exceptional scalability and impact,” said Prasanna Balaprakash of Oak Ridge National Laboratory.
QuaTrEx: Transforming Nanoscale Transistor Design
A team at ETH Zurich developed QuaTrEx, an algorithm suite for next-generation transistor design. Running on Alps with NVIDIA GH200 Superchips, QuaTrEx simulates devices with 45,000+ atoms using FP64 precision and massive parallelism.
“Access to Alps allowed us to simulate devices we couldn’t have imagined handling months ago,” said ETH Zurich professor Mathieu Luisier.
MFC: Record-Scale Spacecraft Simulation
To design spacecraft with densely packed engines, researchers need highly detailed fluid-flow simulations. The MFC solver, created by Georgia Tech in collaboration with NVIDIA, offers:
- 4× faster performance
- 5× greater energy efficiency
- Same accuracy as previous world records
- Expected 10× scale-up on JUPITER
“Our new regularization method and the GH200’s unified memory drastically improved our ability to simulate rocket plumes at unprecedented scales,” said Georgia Tech’s Spencer Bryngelson.
Real-Time Tsunami Digital Twin
Researchers at UT Austin, Lawrence Livermore National Laboratory and UC San Diego have created the world’s first digital twin capable of issuing real-time probabilistic tsunami forecasts using full-physics modeling.
Applied to the Cascadia subduction zone, the system transformed what would normally require 50 years on 512 GPUs into 0.2 seconds on Alps and Perlmutter — a 10-billion-fold speedup.
“For the first time, real-time sensor data and full-physics modeling can be combined fast enough to act before disaster strikes,” said UT Austin’s Omar Ghattas.
Across the ICON, MFC and tsunami-twin projects, researchers relied heavily on NVIDIA CUDA-X and CUDA Graphs to maximize simulation speed, scalability and efficiency.
Source Link:https://blogs.nvidia.com/blog/gordon-bell-finalists-2025/
