There is no question that GPUs have taken the HPC industry by storm over the past several years, speeding up all sorts of highly parallel scientific workloads by huge amounts. I thought this use of GPUs was particularly cool. Every day some 30,000-plus aircraft take flight over the US alone, each with an extremely detailed flight plan that keeps the flight safe, minimizes fuel consumption, and gets the plane and crew between airports with just enough time to turn around for another flight. If weather or maintenance delays just one flight in a busy airport the dominoes begin to fall as delays spread like a virus from airport to airport.
Part of the problem stems from the fact that rerunning the computer model for some 35,000 flights over the course of a normal day takes around 10 minutes, so with conditions that change constantly, there’s no way the computer model can keep up. Some researchers in California have ported this model to run on Nvidia GPUs and sped the simulation time up some 250X, now down to just 2.5s per simulation. This means, for all intents and purposes, the model can now be run real-time and keep up with the constantly changing variables inherent in having so many planes in the air at the same time. This will allow the model to make better flight plans and support the ever increasing number of daily flights. Here is HPC making many peoples’ lives better.
(Source: USING GPUS FOR AIR TRAFFIC MANAGEMENT)
