New Intel Parallel Computing Center established at ETH Zurich

The increasing performance in today's processors almost exclusively stems from the growing number of compute cores. Every processor today is a multi-core chip and co-processors with dozens of cores are available. This growing parallelism creates an enormous opportunity to drastically increase the performance and at the same time creates challenges for the programmability of today's and upcoming architectures. The newly created Intel Parallel Computing Center at the Scalable Parallel Computing Lab ETH Zurich will investigate many-core computing on modern architectures such as the Intel Xeon Phi Coprocessor to prepare for the future. The goal of the center is to meet the prospective demands of scientific computing by developing new parallelization techniques for scientific applications that require processing of massive amounts of data.

"Intel(R) Parallel Computing Centers are collaborations to modernize key applications to unlock performance gains that come through parallelism, enabling the way for the next leap in discovery. We are delighted to be working with the ETH team in their endeavor to improve the prediction of weather and climate research," says Stephan Gillich, Intel’s director for technical computing in EMEA.

"The collaboration with Intel provides a great opportunity to improve the state of the art in parallel computing by designing model-driven techniques for parallel application development," ensures Torsten Hoefler, professor of computer science at ETH and the principal investigator of the center. The center's activities will focus on researching the utility of architectural models for optimization of parallel applications. The first steps will include the development of a detailed system model for many-core, an extension of the previous performance model for on-chip transfers in Intel's Xeon Phi Coprocessor. The developed models will enable algorithm developers and programmers to design close-to-optimal algorithms for computational problems executing on many-core architectures.

The parallel computing center will demonstrate the applicability of the developed modeling techniques in the context of the real-world regional climate and weather model COSMO. MeteoSwiss - the Swiss national weather service - has been working together with the Scalable Parallel Computing Lab for nearly two years. COSMO is the main working horse for operational weather prediction at ten national weather services and is intensively used on supercomputers worldwide for climate research. Ensuring high throughput, efficiency and scalability of COSMO on the current and emerging hardware architectures is of key importance and will have a high impact on the daily lives of researchers. Oliver Fuhrer, senior scientist at MeteoSwiss states: "Exciting results both in the efficient implementation of stencils on multi-core and many-core architectures as well as in achieving bit-reproduciblity without paying a price in performance across different hardware architectures have already been achieved." Fuhrer is convinced that the collaboration between the Scalable Parallel Computing Lab and MeteoSwiss will be significantly expanded through this center.

Scalable codes will eventually run on hundreds of nodes equipped with many-core co-processors. The new parallel computing center will work with the Swiss National Supercomputing Centre (Centro Svizzero di Calcolo Scientifico, CSCS) to ensure the applicability of the results at large scale." This newly created parallel computing laboratory will further strengthen computational science in Switzerland. The approach to focus on real-world computational problems from the beginning is promising and will certainly support CSCS and its users to prepare for the next generation of Exascale supercomputers," says Prof. Thomas Schulthess, director of CSCS.

The center will also benefit students and researchers in that practical many-core computing will be added to the curriculum.