2013 International Supercomputing Conference opens in historic, modern Leipzig

18 Jun 2013 Leipzig - The 2013 International Supercomputing Conference (ISC'13) opened its 28th year - in the city of Leipzig to about 2362 attendees, and many are still registering. The conference, held in the Congress Center Leipzig, will continue through Thursday, June 20. ISC'13 offers more than 350 technical presentations and Europe's largest exhibition for high performance computing. Over 150 of world's leading manufacturers, vendors and research organisations will be showcasing their products and accomplishments during the three-day exhibition.

During the ISC'13 opening session, Prof. Klaus Peter Fähnrich from the University of Leipzig and director of the university computing centre, informed attendees that Leipzig is a city of science and commerce - the university was founded 600 years ago to focus on science and one of the world's first stock exchanges was founded 500 years ago. At the end of the 19th century, Leipzig was one of the world's 10 richest cities, a distinction that suffered under the East German government, though the city is now rebounding. As Prof. Fähnrich put it, "In Chemnitz, the people work hard, in Leipzig they earn the money, and in Dresden, they spend it."

Also, during the opening session, China's Tianhe-2 supercomputer was named the no. 1 system in the world, according to the 41st edition of the TOP500 list of the world's top supercomputers. Tianhe-2, or Milky Way 2, has more than 3 million processor cores and achieved a performance of 33.86 petaflop/s running the Linpack benchmark application. Tianhe-2 is more than twice as fast and the next-nearest supercomputer, Titan at Oak Ridge National Laboratory in the U.S. This is the largest jump in performance since Japan's Earth Simulator stunned the HPC community by taking the top spot in June 2002. For more information, you can visit the TOP500 website.

The opening keynote address was given by Bill Dally, NVIDIA's Chief Scientist and Senior Vice President of Research and former head of Stanford University's Computer Science Department. Bill Dally spoke about the path to exascale that aspires to produce a 1000-fold increase over current levels of supercomputer performance. But he warned that focusing on counting flop/s, or floating point operations, misses the real issue - the need to significantly increase the energy efficiency of supercomputers. "We've been so fixated on counting flops that we think they matter in terms of power, but communication inside the system takes more energy than arithmetic", Bill Dally stated. "Power goes into moving data around. Power limits all computing and communication dominates power."

The solutions, Bill Dally said, need to come from two sides. First, the power issue can be addressed through conventional technology scaling (which gets only 2.2x) computer architecture innovation - where he believes we can get four times more efficient - and improved circuits - from which we can get another 3x. The second area is programming, an area in which the programming tools are not yet trustworthy enough, so programmers do a lot of things themselves. To reach exascale, there needs to be more teamwork between programmers, architects and tools developers, Bill Dally said, and that requires trust.

The PRACE ISC Award, presented by the Partnership for Advanced Computing in Europe, went to a paper on "591 Tflops multi-trillion particles simulation on SuperMUC", describing large-scale performance and scalability studies of an optimized version of the molecular dynamics simulation code ls1 Mardyn. Authors are Wolfgang Eckhardt and Alexander Heinecke of the Technical University of Munich; Reinhold Bader, Matthias Brehm, Nicolay Hammer, Herbert Huber and Hans-Georg Kleinhenz of the Leibniz Computing Center; Jadran Vrabec of the University of Paderborn; Hans Hasse and Martin Horsch of the Laboratory of Engineering Thermodynamics at the Technical University of Kaiserslautern; Martin Bernreuther, Colin W. Glass and Christoph Niethammer of the High Performance Computing Centre Stuttgart; and Arndt Bode and Hans-Joachim Bungartz of the Technical University of Munich and the Leibniz Computing Center.

The Gauss Award was presented for "TUE, a new energy-efficiency Metric Applied at ORNL's Jaguar". Authors are Michael K. Patterson, Intel; Stephen W. Poole, Chung-Hsing Hsu and Don Maxwell, all of Oak Ridge National Laboratory; William Tschudi and Henry Coles, both of Lawrence Berkeley National Laboratory; David J. Martinez of Sandia National Laboratories; and Natalie Bates of the Energy Efficient HPC Working Group. The Gauss Award is sponsored by the German Gauss Center for Supercomputing, a collaboration of the German national supercomputing centers at Garching, Jülich and Stuttgart.

The engaging Vendor Showdown which replaced the well-known Hot Seat Session hosted 18 leading HPC vendors in two different afternoon sessions. The task to evaluate the presentations given by the company CEOs and CTOs was left to the audience. The winners were DataDirect and NEC.

Source: ISC'13