Faster Predictions: To Introduces NVIDIA Earth-2 NIM Microservices to Deliver Higher-Resolution Simulations 500x Faster. Weather technology firms can now create and implement AI models for snow, ice, and hail predictions with to new NVIDIA NIM microservices.
Two new NVIDIA NIM microservices that can 500x the speed of climate change modeling simulation results in NVIDIA Earth-2 were unveiled by NVIDIA today at SC24.
NVIDIA Earth-2 NIM microservices
High-resolution, AI-enhanced, accelerated climate and weather models with interactive visualization.
Climate Digital Twin Cloud Platform
NVIDIA Earth-2 simulates and visualizes weather and climate predictions at a global scale with previously unheard-of speed and accuracy by combining the capabilities of artificial intelligence (AI), GPU acceleration, physical models, and computer graphics. The platform is made up of reference implementations and microservices for simulation, visualization, and artificial intelligence.
Users may employ AI-accelerated models to optimize and simulate real-world climate and weather outcomes with NVIDIA NIM microservices for Earth-2.
The Development Platform for Climate Science
GPU-Optimized and Accelerated Climate Simulation
To increase simulated days per day (SDPD), the Earth-2 development platform is tuned for GPU-accelerated numerical climate simulations at the km-scale.
Data Federation and Interactive Weather Visualization
Extremely large-scale, high-fidelity, interactive projections of global weather conditions are made possible by NVIDIA Omniverse. A data federation engine included into Omniverse Nucleus provides transparent data access across external databases and real-time feeds.
A digital twin platform called Earth-2 is used to model and visualize climate and weather phenomena. To help with forecasting extreme weather occurrences, the new NIM microservices give climate technology application developers cutting-edge generative AI-driven capabilities.
- While maintaining data security, NVIDIA NIM microservices aid in the quick deployment of foundation models.
- The frequency of extreme weather events is rising, which raises questions about readiness and safety for disasters as well as potential financial effects.
- Nearly $62 billion in natural disaster insurance losses occurred in the first half of this year. Bloomberg estimates that is 70% greater than the 10-year average.
The CorrDiff NIM and FourCastNet NIM microservices are being made available by NVIDIA to assist weather technology firms in producing more accurate and high-resolution forecasts more rapidly. When compared to conventional systems, the NIM microservices also provide the highest energy efficiency.
New CorrDiff NIM Microservices for Higher-Resolution Modeling
NVIDIA a generative AI model for super resolution at the kilometer scale is called CorrDiff. At GTC 2024, it demonstrated its potential to super-resolve typhoons over Taiwan. In order to produce weather patterns at a 12x better resolution, CorrDiff was trained using numerical simulations from the Weather Research and Forecasting (WRF) model.
Meteorologists and companies depend on high-resolution forecasts that can be shown within a few kilometers. In order to evaluate risk profiles, the insurance and reinsurance sectors depend on comprehensive meteorological data. However, it is frequently too expensive and time-consuming to be feasible to achieve this level of precision using conventional numerical weather forecast models like WRF or High-Resolution Rapid Refresh.
Compared to conventional high-resolution numerical weather prediction utilizing CPUs, the CorrDiff NIM microservice is 10,000 times more energy-efficient and 500 times quicker. Additionally, CorrDiff is currently functioning at a 300x greater scale. In addition to forecasting precipitation events, such as snow, ice, and hail, with visibility in kilometers, it is super-resolving, or enhancing the quality of lower-resolution photos or videos, for the whole United States.
Enabling Large Sets of Forecasts With New FourCastNet NIM Microservice
High-resolution predictions are not necessary for all use cases. Larger forecast sets with coarser resolution are more advantageous for some applications. Due to computational limitations, state-of-the-art numerical models like as IFS and GFS can only provide 50 and 20 sets of predictions, respectively.
Global, medium-range coarse predictions are provided by the FourCastNet NIM microservice, which is now accessible. Providers may provide predictions over the following two weeks 5,000 times faster than with conventional numerical weather models by using the initial assimilated state from operational weather centers like the National Oceanic and Atmospheric Administration or the European Centre for Medium-Range Weather predictions.
By estimating hazards associated with extreme weather at a different scale, climate tech providers may now anticipate the chance of low-probability occurrences that are missed by present computational processes.
