Thanks for visiting Scientific Computing World.

You're trying to access an editorial feature that is only available to logged in, registered users of Scientific Computing World. Registering is completely free, so why not sign up with us?

By registering, as well as being able to browse all content on the site without further interruption, you'll also have the option to receive our magazine (multiple times a year) and our email newsletters.

Swiss to simulate weather using GPUs

Share this on social media:

The Swiss Federal Office of Meteorology and Climatology (MeteoSwiss) has announced that it has taken delivery of the first GPU-accelerated supercomputer used to power the numerical weather forecasts.

Located in Lugano, Switzerland, the Swiss national Supercomputer Centre (CSCS) provides key supercomputing capabilities required to solve important problems in science and society, including regular weather forecasts. The purchase of this new supercomputer by the CSCS is the first time a GPU-based supercomputer has been used to run production numerical weather models for a major national weather service.

Based on Cray’s CS-Storm system, the new supercomputer is comprised of two cabinets, which each contain 12 hybrid computing nodes with 96 Nvidia Tesla K80 GPU accelerators and 24 Intel Haswell CPUs. This gives the Cray CS-Storm a peak performance of up to 15 teraflops per node.

‘High-quality weather forecasts always depend upon processing power,’ said CSCS director Thomas Schulthess. ‘The GPUs and the re-engineered code allow us to increase the simulation performance 40 times but remain within the same energy and facility footprint of the supercomputer that we installed for MeteoSwiss in 2012. Moreover, thanks to the GPU technology we spend three times less on power compared to a system built entirely with conventional processors.’

MeteoSwiss worked closely with CSCS, the Center for Climate Systems Modelling (C2SM), the development partner Supercomputing Systems, the computer manufacturer Cray and accelerated computing specialist Nvidia to develop the powerful, yet cost-effective, and energy-efficient supercomputer.

The new system architecture and completely revised application software based on the COSMO model means that MeteoSwiss can now provide state-of-the-art weather forecasts at a resolution of up to three times higher than was previously possible.

‘To guarantee more detailed weather forecasts, the simulations will be based on a grid spacing of 1.1 kilometres, which runs recurrently every three hours,’ said Peter Binder, director general of MeteoSwiss. ‘This grid spacing makes it possible to predict with more detail the precipitation distribution and the risk of storms or valley wind systems in the Swiss mountains.’

Barry Bolding Cray’s senior vice president and chief strategy officer, stressed the importance, not only of the first GPU accelerated supercomputer used for national weather prediction but also the performance increases delivered using this architecture.

Bolding said: ‘Today’s weather and climate models are ingrained with massive amounts of data and science that have increasingly demanding compute requirements. With an eight-to-two ratio of GPU accelerators to CPUs, the Cray CS-Storm system at MeteoSwiss is a powerful tool for running production models at a much higher resolution and granularity. This ground-breaking use of the Cray CS-Storm system to run data-intensive, operational weather forecasts is a milestone in supercomputing.’

The Cray CS-Storm system is one of the most powerful single-node cluster architectures available today, designed to support highly scalable applications in areas such as energy, manufacturing, earth sciences, life sciences, financial services, machine learning and geospatial intelligence. The system uses the supercomputing architecture developed originally for the Cray CS400 system, and includes the Cray Advanced Cluster Engine cluster management software and the complete Cray Programming Environment.

After a transitional period, the new system is expected to replace the current one by spring/summer 2016. Until then both systems will run in parallel for weather prediction in Switzerland.