Europe’s Fastest Supercomputer Shatters Energy Efficiency Barrier

by George Millington

Back at our 2013 GPU Technology Conference in March, we announced plans for a new GPU-accelerated Swiss supercomputer – “Piz Daint.” Many shared our optimism that it would scale new heights in performance and energy efficiency.

Today, Piz Daint is exceeding those bold expectations.

The new Top500 list released just today of the world’s highest performance supercomputers, shows Piz Daint hit 6.2 petaflops of performance as measured by the LINPACK Benchmark — making it the highest performance supercomputer in Europe.

Not only that, but when it comes to energy efficiency, it’s the first petascale-class system to break the 3 gigaflop per watt barrier (3.11 gigaflops/watt).

That’s nearly 50 percent more efficient than next most efficient petascale system is at 2.17 gigaflops/watt.

Piz Daint’s success is due in part to our flagship Tesla K20X GPU accelerators, the fastest, most energy efficient GPUs ever built.

Named after a mountain in the Swiss Alps, the system was built by the Swiss National Supercomputing Center (CSCS) to provide scientists with extreme-scale computing resources. It has been designed to reach high efficiency in simulations for application areas ranging from materials science, life science, and physics, to climate, meteorology and geophysics.


European GPU Supercomputing on the Rise

Piz Daint isn’t the only new GPU-accelerated system debuting on the new Top500 list.

There are three other new GPU-accelerated entries, each reporting impressive performance numbers, and all designed to extend the boundaries of scientific research.

These include a new system at Rechenzentrum Garching in Germany, a joint computing center of the Max Planck Society and the Max Planck Institute for Plasma Physics, clocking in at 709 teraflops. Also new to the group is “Romeo” at France’s University of Reims and Cambridge University’s “Wilkes”  System.

These new systems show how the European research, supercomputing and high-performance computing communities are embracing accelerated computing to drive innovation in engineering, science, big data analytics and a range of other fields.

3X Faster, 7X More Efficient on Top Weather Forecasting Application

Among applications to run on Piz Daint is “COSMO,” an atmospheric model used by the German Meteorological Service, MeteoSwiss and other weather services for their daily weather forecasts.

In fact, COSMO is one of the top European regional weather and climate modeling applications used by national forecasting services in Germany, Switzerland, Italy, Poland, Greece, Romania and Russia.

It allows researchers to run sophisticated high-resolution models to monitor climate changes and track weather patterns. It’s maintained by the Consortium for Small-scale Modeling, a group of seven national weather services and research groups of about 50 universities worldwide.

The Center for Climate Systems Modeling (C2SM), CSCS, and MeteoSwiss have redesigned the dynamical core of COSMO and added support GPU accelerators, and decided to test the updated code on Piz Daint. On a given number of nodes, COSMO ran over three times faster on Piz Daint while consuming seven times less energy, compared with CSCS’s current flagship supercomputing system, “Monte Rosa.”

With this substantial increase in performance, researchers will be able to run many more sophisticated, ultra-high-resolution models, giving them new insights into how complex climate and weather systems work.

You can learn more about Piz Daint at the Supercomputing 2013 (SC13) in Denver this week.

Thomas Schulthess, director of CSCS, will present in our GPU Technology Theater, Booth #613, on Wednesday, Nov. 20 from 2:30 to 3:00 pm MT — “Piz Daint: A Productive, Energy Efficient Supercomputer with Hybrid CPU-GPU Nodes.”

Not going to SC13 this year? All of our booth presentations will be streamed live at our SC13 web page.