A somewhat rainy Hamburg, Germany, is home to NVIDIA’s Tesla high performance computing (HPC) team this week as we spend a few days at the International Supercomputing Conference ‘11 (ISC’11), where the Top500 organization this morning announced the new Top500 List of World’s Supercomputers for June 2011.
This year, two GPU-enabled HPC systems from Europe’s leading academic and research institutions landed on the updated list of the world’s leading supercomputers. The Italy-based CINECA took the #54 spot, and Moscow State University in Russia reached #13 on the list with its recently upgraded Lomonosov Supercomputer.
Barcelona Supercomputing Center will also deplopy GPUs in their MareNostrum supercomputer this summer, allowing it to double its peak performance while consuming seven-times less power. More performance for less power.
We’re also hearing that GENCI in France will finish their GPU-based expansion of their CURIE HPC system by the end of 2011. The CURIE system will use NVIDIA’s Tesla M2090 GPUs, and is expected to deliver around 184 teraflops peak performance.
Tackling today’s scientific problems helps keep the Tesla team energized. We love HPC breakthroughs like the world’s largest molecular simulation recently achieved by the Chinese Academy of Sciences. We’re also hearing a great deal of discussion, here at ISC 2011 this week, about the next evolution of computing – reaching the exascale computing milestone and solving the performance and power problems associated with that level of performance.
Thomas Lippert, director of the Jülich Supercomputing Center, in Germany, believes that GPUs will put us on the path to exascale supercomputers.
“Only with massive technological advances in many-core processor technologies can the peak performance of supercomputers, currently set at a few petaflops, be increased to exascale (1,000 times faster) by the end of this decade – at affordable energy consumption”, he said.
The New Top500 list also revealed a new #1. Japan’s K Computer took the top spot from China’s Tianhe-1A, with published Linpack performance of 8.1 petaflops. This is the end result of a five-plus year project, and we congratulate them on hitting this performance milestone. Tokyo Institute of Technology’s GPU-enabled Tsubame 2.0 is still ruling the roost in the Top 5 with regards to performance per watt though, delivering 1.1 petaflops for 1.3 megawatts, besting the K Computer’s nearly 10 megawatts of total power consumption.
We’re sure that exascale performance and the use of GPUs in scientific supercomputing will be central to tomorrow’s discussion, featuring NVIDIA’s former chief scientist and fellow, David Kirk, and Louisiana State University’s Thomas Sterling. If you’re here in Hamburg, be sure to check it out.