NVIDIA announced two new AI-powered microservices for its Earth-2 platform at SC24. The NVIDIA NIM (NVIDIA Inference Microservices) tools aim to accelerate climate modeling simulations by 500 times while helping weather technology companies maintain data security.
Natural disasters have caused approximately $62 billion in insured losses during the first half of 2024, 70% above the 10-year average, according to Bloomberg. The new microservices are designed to assist weather technology companies in developing more precise predictions.
CorrDiff NIM Technical Specifications
CorrDiff NIM, demonstrated at GTC 2024 for typhoon prediction over Taiwan, generates weather patterns at 12 times higher resolution than current standards. The service offers:
- Processing 500x faster than traditional CPU-based weather prediction
- Energy efficiency improved by 10,000x
- Operational scale increased by 300x
- Coverage across the United States
- Precipitation forecasting for snow, ice, and hail
- Kilometer-scale visibility
Insurance and reinsurance industries can use this data for risk assessment, addressing previous computational limitations.
FourCastNet NIM Capabilities
FourCastNet NIM focuses on global, medium-range forecasting at coarser resolutions, offering:
- Processing speed 5,000x faster than conventional numerical weather models
- Two-week forecast range
- Integration with initial state data from ECMWF and NOAA
- Enhanced ability to predict low-probability weather events
Current models IFS and GFS are limited to 50 and 20 forecast sets respectively. FourCastNet NIM allows weather technology providers to process larger forecast sets.
Earth-2 Platform Integration
Both microservices are part of Earth-2, NVIDIA’s digital twin platform for weather and climate simulation. The platform combines CorrDiff NIM’s resolution capabilities with FourCastNet NIM’s broad-scale forecasting to provide weather technology companies with tools for developing forecasting applications.
