China now has nearly as many top-ranking supercomputers as Britain, France and Germany combined, and its Tianhe-2 leads the TOP500 list of the world’s fastest supercomputers — for the third consecutive time.
A detailed analysis of the latest TOP500 list is being presented today (23 June) at the International Supercomputing Conference (ISC’14) in Leipzig, Germany. The full list is published at TOP500.org.
The Tianhe-2 (Milky Way 2), located at the National Supercomputer Centre in Guangzhou, was developed by China’s National University of Defence Technology. It has a performance of 33.86 Pflop/s on the Linpack benchmark, almost double the next fastest machine, the Cray XK7 Titan machine at the Oak Ridge National Laboratory in the USA, which had a performance of 17.59 Pflop/s.
However, the performance comes at a price. The Chinese machine has 3,120,000 cores which consume a huge amount of power – 17,808kW – whereas the Titan has 560,640 cores and consumes 8,209kW.
Tianhe-2 and the seventh-ranked system, Stampede, use Intel Xeon Phi processors to speed up their computational rate. The second fastest system, Titan, and the sixth, Piz Daint, use NVIDIA GPUs to accelerate computation. A total of 62 systems on the list are using accelerator/co-processor technology, up from 53 from November 2013. Forty-four of these use NVIDIA chips, two use ATI Radeon, and there are now 17 systems with Intel Xeon Phi.
Among the world’s top 10 supercomputers, there was little change in the ranking in this list compared to last November’s (the list is published twice a year). The only new entry within the top 10 was actually at number 10 — another Cray machine, this time an XC30 installed at an undisclosed US government site and achieving a performance of 3.14 Pflop/s.
The United States still has the highest number of systems – 233 – although this is down from 265 in the previous list. The number of Chinese systems on the list rose from 63 to 76. In comparison, the UK has 30; France, 27; and Germany, 23; combined. Japan also increased its showing, up to 30 from 28 on the previous list.
Although total combined performance is still growing, there has been a noticeable slowdown compared to the previous long-term trend, according to the list’s authors. The total combined performance of all 500 systems is now 274 Pflop/s.
Intel continues to provide the processors for the largest share (85.4 per cent) of TOP500 systems. The share of IBM Power processors remains at 8 per cent, while the AMD Opteron family is used in 6 per cent of the systems, down from 9 per cent on the previous list.
In terms of numbers of systems, HP has the lead with 197 (39 per cent) compared to IBM with 164 systems (33 per cent). Cray remains third with 10 per cent (48 systems).