SANTA CLARA — At SC24, NVIDIA has unveiled two groundbreaking NIM microservices aimed at accelerating the development and deployment of AI models for weather prediction, specifically targeting snow, ice, and hail forecasts. These new microservices, CorrDiff and FourCastNet, are designed to significantly improve the speed and resolution of weather simulations, offering a dramatic leap in climate technology advancements.
Through the use of these microservices, NVIDIA is pushing the boundaries of climate change modeling, allowing for 500x faster simulations within the NVIDIA Earth-2 platform. Earth-2 serves as a digital twin, offering a comprehensive environment for simulating and visualizing both weather and climate patterns. These innovations come at a crucial time as extreme weather events become more frequent and impactful, posing both safety risks and financial burdens. According to a Bloomberg report, natural disasters accounted for $62 billion in insured losses during the first half of 2024, a 70% increase compared to the 10-year average.
NVIDIA’s CorrDiff NIM microservice leverages generative AI to super-resolve weather models at kilometer-scale resolution. A recent demonstration showed its capability to enhance typhoon simulations over Taiwan, offering a 12x higher resolution than traditional models like WRF. This level of detail is crucial for industries like insurance and reinsurance, which depend on precise weather data for risk assessments. What sets CorrDiff apart is its remarkable speed and energy efficiency—500x faster and 10,000x more energy-efficient than conventional high-resolution prediction models running on CPUs. The microservice is now operational on a larger scale, predicting precipitation events, including snow and hail, with unprecedented precision across the entire United States.
Meanwhile, the FourCastNet NIM microservice offers global, medium-range forecasts at a coarser resolution, but with the ability to generate thousands of forecasts at once. By incorporating initial data from global weather agencies like the European Centre for Medium-Range Weather Forecasts (ECMWF) and NOAA, FourCastNet produces forecasts up to two weeks in advance, delivering predictions 5,000x faster than traditional numerical models. This allows for a broader scope of forecasting, helping climate technology providers predict low-probability extreme events that might be missed by conventional methods.
These new advancements from NVIDIA are setting the stage for a major leap forward in climate technology, enabling more accurate and faster predictions of extreme weather events while providing enhanced energy efficiency. The release of these microservices offers significant potential for weather technology companies to better manage the growing challenges posed by climate change.
Learn more about CorrDiff and FourCastNet NIM microservices on ai.nvidia.com.
