Traditionally in November updated list of most powerful supercomputers in the world. The sixth time in a row was the leader of the Chinese Tianhe-2 monster, built-in 2013, the Defense Science and Technology University of China in Guangzhou. The total number of members of the list of Chinese supercomputer has increased over the six months to three times (from 37 to 109). The share of the United States (199/500) is now a record low in the history of reference ranking since 1993.
The top ten included only two new supercomputers – Trinity (sixth place) and Hazel-Hen (eighth place). This system Cray XS40 processors Xeon E5-26xx and without accelerators floating-point calculations as their productivity is 8.1 and 5.6 respectively Pflops.
Supercomputer Cray Trinity was built together with the specialists of the Department of Energy Los Alamos National Laboratory and Sandia National Laboratories. Hazel-Hen also established Cray and installed in Stuttgart, Germany. He made his debut in the summer of supercomputer Shaheen II, set in the King Abdullah University of Science and Technology (Saudi Arabia), has moved from seventh to ninth place.
In addition to these three high-performances all the others were created / upgraded to the II quarter of 2015. The oldest of them Korean TachyonII (Sun Blade x6048) and the Canadian GPC (xSeries iDataPlex) operating since 2009. However, they still have not moved to the end of the list, holding 372 and 392 seats. Eleven supercomputers configuration has not changed since 2010, thirty-one – since 2011, sixty-three – in 2012, seventy-seven – in 2013 three hundred supercomputer first appeared in the ranking last year.
Chinese supercomputer “Tianhe-2” is still ahead of the best American (Titan) almost doubled. It consists of 260 thousand. With processors Intel Xeon E5-2692 (IvyBridge) and 48 thousand. Accelerators Intel Xeon Phi 31S1P. The total volume of its memory is 1.4 petabytes.
Tianhe-2 is running a specialized OS Kylin Linux and demonstrates HPL Linpack benchmark peak performance level Pflops 33.86 (33.86 quadrillion operations per second). The developers emphasize that continue to optimize load management algorithms. Due to pure software solutions productivity can be increased by another 50% – 60%.
The only advantage of the “Titan” remains its best energy efficiency. Using the system Cray XK7 accelerators Nvidia K20x allows you to limit the total power consumption of up to 8.21 MW (17.8 MW against the “Tianhe-2”) and achieved a 2143 Gflops / watt (compared to 1.9 Gflops / W at the Tianhe-2).
Six months ago, 90 systems on the list use math coprocessors. Now their number has grown to 104. Sixty-six of them are using GPU Nvidia, three – AMD / ATI Radeon, and twenty-seven – Intel Xeon Phi. In four supercomputers used hybrid Nvidia Kxx architecture and Intel Xeon Phi.
Intel continues to hold the lead: its processors are used in 89% of the ranking of supercomputers TOP500, while the share of AMD Opteron is only 4.2%. Interest in the IBM Power processors also reduced in spite of their architectural advantages. Now they continue to use only 26 systems, and six months ago there were 38. Cray XC Series is now the most popular in the TOP 10.
Among the vendors leading role in the segment of high-performance systems plays a Hewlett-Packard. By HP created nearly a third of the top supercomputers in the world (155 or 31.2%). Second place belongs to Cray (69 systems), and the third – IBM (45 supercomputers). For the first time IBM has overtaken the Chinese vendor Sugon. It owns 49 systems of the updated rankings TOP500.
However, the overall performance of supercomputers is now growing much slower than in previous years. From 1994 until 2008 the growth was 90% per year. Since 2008 – 55% or less. Although investments in high-performance computing in general are growing, their main stream is now aimed at the clusters of average power.