Monster Chinese HPC System Clocks 30+ PetaFLOPS

Intel-based supercomputer is expected to lead the next TOP500 list.

Details are emerging on a new 30+ petaFLOPS HPC system that will be operational by the end of this year in the Chinese National Supercomputing Center in Guangzhou, China. The new system is almost certainly claim the number one spot in the next version of the HPC TOP500 list, which will appear later in June 2013. Jack Dongarra, of the University of Tennessee, posted a detailed report on the computer after visiting the site in conjunction with the International HPC Forum in Changsha China. The system has a theoretical peak of 54.9 petaFLOPS with a measured HPL benchmark of 30.65 petaFLOPS.

This gigantic HPC system, which is called the Tianhe-2 (TH-2) or the Milky Way-2 supercomputer, comes with 32,000 Intel Ivy Bridge Xeon sockets and 48,000 Xeon Phi coprocessors for a total of 3,120,000 cores. (Each Ivy Bridge socket has 12 cores and each Xeon Phi socket has 57.) Each node consists of two Ivy Bridge sockets and three Intel Xeon Phi sockets; the system has a total of 16,000 nodes. Each node has 64GB of memory plus 8GB for each Xeon Phi for a total of 88GB per node, which works out to a total of 1.4PB of memory for the whole system. The water-cooled system uses a proprietary, custom-built interconnect.

The TOP500 computers get faster every year, but this system truly could be a game changer for the HPC industry. For purposes of comparison, the current leader of the TOP500 list the Titan Cray XK7 system in Oak Ridge, Tennessee, USA, which tops out at around 17.6 petaFLOPS, compared with the 30.65 petaFLOPS for the TH-2 system.