AMD and Oak Ridge Lab Build ORBIT-2, a 0.9 Km Weather AI Model
ORBIT-2 moves weather forecasting from simulation-based methods to AI-based methods at an unprecedented global resolution, with enough speed for edge-device deployment and recognition from the HPC community's top awards.
Reporting from 1 sources: GIGAZINE.
AMD and Oak Ridge National Laboratory developed ORBIT-2, an AI foundation model for global weather downscaling at 0.9 km resolution. The model combines Vision Transformer architecture with scalable algorithms, aims to replace traditional simulation-based forecasting, and is a finalist for the Gordon Bell Prize and a Best Paper Award at Supercomputing 2025.
AMD and Oak Ridge National Laboratory have released ORBIT-2, an AI foundation model that performs global weather downscaling at 0.9 km resolution. The model uses a Vision Transformer image recognition architecture and scalable algorithms to generate physically consistent predictions of temperature and precipitation. AMD says ORBIT-2 generalizes across regions and variables, shifting weather forecasting from traditional simulation-based approaches to AI-based ones. The model has been selected as a finalist for the Gordon Bell Prize, often called the Nobel Prize of Supercomputing, and for the Best Paper Award at Supercomputing 2025.
AMD highlights three advantages: disaster preparedness through detailed spatial data, infrastructure planning for flood- and heatwave-prone areas, and near-real-time situational awareness. ORBIT-2 can run on edge devices with inference times in milliseconds, broadening access to high-resolution weather data for researchers, policymakers, and industries. Ralph Wittig, Corporate Fellow at AMD, said the project embodies the transformative potential of exascale AI for science.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.