{"id":3063,"date":"2023-07-05T17:52:52","date_gmt":"2023-07-05T17:52:52","guid":{"rendered":"https:\/\/salarydistribution.com\/machine-learning\/2023\/07\/05\/a-change-in-the-weather-ai-accelerated-computing-promise-faster-more-efficient-predictions\/"},"modified":"2023-07-05T17:52:52","modified_gmt":"2023-07-05T17:52:52","slug":"a-change-in-the-weather-ai-accelerated-computing-promise-faster-more-efficient-predictions","status":"publish","type":"post","link":"https:\/\/salarydistribution.com\/machine-learning\/2023\/07\/05\/a-change-in-the-weather-ai-accelerated-computing-promise-faster-more-efficient-predictions\/","title":{"rendered":"A Change in the Weather: AI, Accelerated Computing Promise Faster, More Efficient Predictions\u00a0\u00a0"},"content":{"rendered":"
\n

The increased frequency and severity of extreme weather and climate events could take a million lives and cost $1.7 trillion annually by 2050, according to the Munich Reinsurance Company<\/a>.<\/p>\n

This underscores a critical need for accurate weather forecasting, especially with the rise in severe weather occurrences such as blizzards, hurricanes and heatwaves. AI and accelerated computing<\/a> are poised to help.<\/p>\n

More than 180 weather modeling centers employ robust high performance computing (HPC)<\/a>\u00a0infrastructure to crunch traditional numerical weather prediction (NWP) models. These include the European Center for Medium-Range Weather Forecasts (ECMWF), which operates on 983,040 CPU cores, and the U.K. Met Office\u2019s supercomputer, which uses more than 1.5 million CPU cores and consumes 2.7 megawatts of power.<\/p>\n

Rethinking HPC Design<\/strong><\/h2>\n

The global push toward energy efficiency is urging a rethink of HPC system design. Accelerated computing, harnessing the power of GPUs, offers a promising, energy-efficient alternative that speeds up computations.<\/p>\n

<\/p>\n

\"Chart<\/p>\n

<\/a>

On the left, results based on ECMWF Integrated Forecast System 51-member ensembles on Intel Broadwell CPUs, FourCastNet 1,000-member ensembles on 4x NVIDIA A100 Tensor Core GPUs; assuming 10 modeling centers running the same forecast workload. On the right,\u00a0 results based on measured performance of the ICON model. CPU: 2x AMD Milan. GPU: 4x NVIDIA H100 Tensor Core PCIe.<\/figcaption><\/figure>\n

NVIDIA GPUs have made a significant impact on globally adopted weather models, including those from ECMWF, the Max Planck Institute for Meteorology, the German Meteorological Service and the National Center for Atmospheric Research.<\/p>\n

GPUs enhance performance up to 24x, improve energy efficiency, and reduce costs and space requirements.<\/p>\n

\u201cTo make reliable weather predictions and climate projections a reality within power budget limits, we rely on algorithmic improvements and hardware where NVIDIA GPUs are an alternative to CPUs,\u201d said Oliver Fuhrer, head of numerical prediction at MeteoSwiss, the Swiss national office of meteorology and climatology.<\/p>\n

AI Model Boosts Speed, Efficiency<\/strong><\/h2>\n

NVIDIA\u2019s AI-based weather-prediction model FourCastNet<\/a> offers competitive accuracy with orders of magnitude greater speed and energy efficiency compared with traditional methods. FourCastNet rapidly produces week-long forecasts and allows for the generation of large ensembles \u2014 or groups of models with slight variations in starting conditions \u2014 for high-confidence, extreme weather predictions.<\/p>\n

For example, based on historical data, FourCastNet accurately predicted the temperatures on July 5, 2018, in Ouargla, Algeria \u2014 Africa\u2019s hottest recorded day.<\/p>\n

<\/p>\n

\"An<\/p>\n

<\/a>

A visualization of the ground-truth weather across Africa in July 2018 (center), surrounded by globes displaying heat domes that represent accurate predictions produced by FourCastNet (ensemble members).<\/figcaption><\/figure>\n

Using NVIDIA GPUs, FourCastNet quickly and accurately generated 1,000 ensemble members, outpacing traditional models. A dozen of the members accurately predicted the high temperatures in Algeria based on data from three weeks before it occurred.<\/p>\n

This marked the first time the FourCastNet team predicted a high-impact event weeks in advance, demonstrating AI\u2019s potential for reliable weather forecasting with lower energy consumption than traditional weather models.<\/p>\n

FourCastNet uses the latest AI advances, such as transformer models<\/a>, to bridge AI and physics for groundbreaking results. It\u2019s about 45,000x faster than traditional NWP models. And when trained, FourCastNet consumes 12,000x less energy to produce a forecast than the Europe-based Integrated Forecast System, a gold-standard NWP model.<\/p>\n

\u201cNVIDIA FourCastNet opens the door to the use of AI for a wide variety of applications that will change the shape of the NWP enterprise,\u201d said Bjorn Stevens, director of the Max Planck Institute for Meteorology.<\/p>\n

Expanding What\u2019s Possible<\/strong><\/h2>\n

In an NVIDIA GTC session<\/a>, Stevens described what\u2019s possible now with the ICON<\/a> climate research tool. The Levante supercomputer<\/a>, using 3,200 CPUs, can simulate 10 days of weather in 24 hours, Stevens said. In contrast, the JUWELS Booster supercomputer<\/a>, using 1,200 NVIDIA A100 Tensor Core GPUs<\/a>, can run 50 simulated days in the same amount of time.<\/p>\n

Scientists are looking to study climate effects 300 years into the future, which means systems need to be 20x faster, Stevens added. Embracing faster technology like NVIDIA H100 Tensor Core GPUs<\/a> and simpler code could get us there, he said.<\/p>\n

Researchers now face the challenge of striking the optimal balance between physical modeling and machine learning to produce faster, more accurate climate forecasts. A ECMWF blog<\/a> published last month describes this hybrid approach, which relies on machine learning for initial predictions and physical models for data generation, verification and system refinement.<\/p>\n

Such an integration \u2014 delivered with accelerated computing \u2014 could lead to significant advancements in weather forecasting and climate science, ushering in a new era of efficient, reliable and energy-conscious predictions.<\/p>\n

Learn more about how accelerated computing and AI boost climate science through these resources:<\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

https:\/\/blogs.nvidia.com\/blog\/2023\/07\/05\/ai-efficient-weather-predictions\/<\/p>\n","protected":false},"author":0,"featured_media":3064,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[3],"tags":[],"_links":{"self":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/posts\/3063"}],"collection":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/comments?post=3063"}],"version-history":[{"count":0,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/posts\/3063\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/media\/3064"}],"wp:attachment":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/media?parent=3063"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/categories?post=3063"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/tags?post=3063"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}