It's the high-performance computing shot heard 'round the world: China now has the fastest supercomputer. It was revealed today at the HPC China 2010 conference at the Beijing Computing Center that the Tianhe-1A supercomputer has set a new performance record of 2.507 petaflops as measured by the LINPACK benchmark, which may or may not be that impressive to you depending upon how much linear algebra makes a difference in your research. While the fact that Tianhe-1A is the brainchild of China’s National University of Defense Technology (NUDT) might the have the folks at the Pentagon and the National Security Agency feeling a little queasy this morning, the new supercomputer is a great example of heterogeneous computing with its 7,168 GPUs and 14,336 multi-core CPUs, and with a low-power draw of 4.04 megawatts, also serves as a solid case study for GPU-power efficiency. Despite its origins in defense, Tianhe will be an open-access system for all sorts of large-scale scientific computing projects.
Tianhe-1A is housed at National Supercomputer Center in Tianjin and is already fully operational:
“What is scary about this is that the U.S. dominance in high-performance computing is at risk… One could argue that this hits the foundation of our economic future,” says Wu-chun Feng, a supercomputing expert and professor at Virginia Polytechnic Institute and State University, in a New York Times piece about Tianhe-1A. But it’s not like the Chinese have built the Death Star of supercomputers that threatens Mom, apple pie, or even our computing-superpower status - the machine is after all comprised of Intel and Nvidia chips (both are U.S.-based companies). What they have developed however is apparently an interconnect technology that links Tiahne's processors together with a level of speed and efficiency far beyond that of the standard InfiniBand interconnect architecture used by almost all supercomputers today. Hopefully Tiahne's minders will some day soon be so kind as to let us dip our Freedom Fries in their secret sauce.