Peter DeSantis, VP of AWS Global Infrastructure in his keynote in AWS re:Invent mentioned that there are few workloads that are more demanding than weather forecasting such that it needs these extreme high performance super-computers.<p>Why does it need such high performance hardware?
Edit. Even better: <a href="https://en.wikipedia.org/wiki/Numerical_weather_prediction" rel="nofollow">https://en.wikipedia.org/wiki/Numerical_weather_prediction</a><p>Very broadly, I'm not an expert in this.<p>It's simulation.<p>The models for wheather split the atmosphere in 3D blocks, on to which observed data parameters (temp, pressure, wind direction & speed, humidity, more perhaps?) is "projected".<p>Then calculations are run to forecast the behaviour of each and every block and its neighbors, on the n next time steps.<p>The smaller the blocks, the more precise you can forecast.
The more the blocks, the larger the area you can cover.
Each and every parameter can impact others.
Each and every cell impacts its neighbours.<p>All of this combines into quite a huge computational challenge.
It's physics simulations on a gigantic scale, performed continuously on deadlines, and can easily use any available gain in computer performance to improve precision.