Today at SC24, NVIDIA announced two new NVIDIA NIM microservices that enable 500x the speed of simulation results for climate change modeling in NVIDIA Earth-2.
Earth-2 NIM microservices from NVIDIA
AI-powered, high-resolution, expedited weather and climate models with interactive visualization.Cloud Platform for Climate Digital Twins
NVIDIA Earth-2 combines the power of artificial intelligence (AI), GPU acceleration, physical models, and computer graphics to simulate and depict weather and climate predictions at a global scale with unprecedented speed and accuracy. The platform consists of microservices and reference implementations for artificial intelligence, visualization, and simulation.With NVIDIA NIM microservices for Earth-2, users may use AI-accelerated models to optimize and simulate real-world climate and weather outcomes.
The Climate Science Development Platform
Accelerated and GPU-Optimized Climate Simulation
The Earth-2 development platform is optimized for GPU-accelerated numerical climate simulations at the km-scale in order to enhance the number of simulated days per day (SDPD).Interactive Weather Visualization and Data Federation
NVIDIA Omniverse enables extremely large-scale, high-fidelity, interactive weather forecasts all across the world. Omniverse Nucleus comes with a data federation engine that facilitates transparent data access across real-time streams and other databases.Climate and weather events are modeled and visualized using a digital twin platform called Earth-2. The new NIM microservices provide climate technology application developers with advanced generative AI-driven capabilities to aid in forecasting extreme weather events.
- NVIDIA NIM microservices facilitate the rapid deployment of foundation models while preserving data security.
- Extreme weather events are occurring more frequently, which raises concerns about preparedness and safety for disasters as well as the financial consequences.
- In the first half of this year, natural disaster insurance losses were around $62 billion. According to Bloomberg, that is 70% higher than the 10-year average.
Novel CorrDiff NIM Microservices for Modeling at Higher Resolution
CorrDiff is a generative AI model developed by NVIDIA for super resolution at the kilometer scale. It showed that it could super-resolve typhoons across Taiwan at GTC 2024. CorrDiff was trained using numerical simulations from the Weather Research and Forecasting (WRF) model to generate weather patterns with a 12x higher resolution.High-resolution forecasts that can be displayed within a few kilometers are essential to meteorologists and businesses. The insurance and reinsurance industries rely on detailed meteorological data to assess risk profiles. However, achieving this degree of precision using traditional numerical weather forecast models such as WRF or High-Resolution Rapid Refresh is often too costly and time-consuming to be practical.
CorrDiff NIM microservice is 500 times faster and 10,000 times more energy efficient than traditional high-resolution numerical weather forecast using CPUs. Furthermore, CorrDiff is currently operating at a scale that is 300 times larger. For the entire United States, it is super-resolving, or improving the quality of lower-resolution images or videos, in addition to forecasting precipitation events, such as snow, ice, and hail, with visibility in kilometers.
Using the New FourCastNet NIM Microservice to Enable Big Forecast Sets
For some application scenarios, high-resolution predictions are not required. Some applications benefit more from larger forecast sets with coarser resolution. The most advanced numerical models, like as IFS and GFS, are limited by computational constraints to producing just 50 and 20 sets of predictions, respectively.The FourCastNet NIM microservice is now available and offers global, medium-range coarse predictions. By using the initial assimilated state from operational weather centers such as the European Centre for Medium-Range Weather or the National Oceanic and Atmospheric Administration, providers can provide predictions over the next two weeks 5,000 times faster than with traditional numerical weather models.
Climate tech providers can now predict the likelihood of low-probability events that are overlooked by current computational techniques by assessing the hazards associated with extreme weather at a different scale.
0 Comments