MANNHEIM, Ger­ma­ny; BERKELEY, Calif.; and KNOXVILLE, Tenn.—For the fourth con­se­cu­ti­ve time, Tianhe‑2, a super­com­pu­ter deve­lo­ped by China’s Natio­nal Uni­ver­si­ty of Defen­se Tech­no­lo­gy, has retai­ned its posi­ti­on as the world’s No. 1 sys­tem with a per­for­mance of 33.86 petaflop/s (qua­dril­li­ons of cal­cu­la­ti­ons per second) on the Lin­pack bench­mark, accor­ding to the 44th edi­ti­on of the twice-year­ly TOP500 list of the world’s most powerful supercomputers.

In fact, the­re was litt­le chan­ge among the ran­king of the world’s top 10 super­com­pu­ters in the latest edi­ti­on of the clo­se­ly wat­ched list. The only new ent­ry was at num­ber 10—a 3.57 petaflop/s Cray CS-Storm sys­tem instal­led at an undis­c­lo­sed U.S. govern­ment site.

A detail­ed ana­ly­sis of the latest TOP500 list will be pre­sen­ted Tues­day, Nov. 18, at the SC14 con­fe­rence in New Orleans. At that time, the full list will also be published at TOP500.org.

Alt­hough the United Sta­tes remains the top coun­try in terms of over­all sys­tems with 231, this num­ber is down from 233 in June 2014 and down from 265 on the Novem­ber 2013 list. The U.S. is nea­ring its his­to­ri­cal low num­ber on the list.

The num­ber of Euro­pean sys­tems rose to 130, up from 116 last June, while the num­ber of sys­tems across Asia drop­ped from 132 to 120. The num­ber of Chi­ne­se sys­tems on the list also drop­ped, now at 61, com­pared to 76 in June 2014. Over the same peri­od, Japan increased its num­ber of sys­tems from 30 to 32.

Slowing trend in performance growth

Sin­ce its incep­ti­on in June 1993, the TOP500 list has ser­ved as a con­sis­tent mea­su­re of the per­for­mance growth of super­com­pu­ters, sin­ce all sys­tems are ran­ked accor­ding to per­for­mance run­ning the same Lin­pack bench­mark appli­ca­ti­on. With the latest list, the over­all list-by-list growth rates of per­for­mance con­ti­nues to be at his­to­ri­cal­ly low values for the last two years.

This lag in the over­all avera­ge per­for­mance of all 500 sys­tems is noti­ce­ab­ly influen­ced by the very lar­ge sys­tems at the top of the list. Recent instal­la­ti­ons of very lar­ge sys­tems – up to June 2013 – have coun­ter­ac­ted the redu­ced growth rate at the bot­tom of the list, but with few new sys­tems at the top of the past few lists, the over­all growth rate is now slo­wing. This offers an indi­ca­ti­on that the mar­ket for the very lar­gest sys­tems might curr­ent­ly behave dif­fer­ent­ly from the mar­ket of mid-sized and smal­ler supercomputers.

This is sup­port­ed by the fact that the per­for­mance of the last sys­tem on the list (No. 500) has con­sis­t­ent­ly lag­ged behind his­to­ri­cal growth trends for the past five years, a tra­jec­to­ry that now increa­ses by 55 per­cent each year. Bet­ween 1994 and 2008, howe­ver, the annu­al growth rate for the No. 500 sys­tems’ per­for­mance was 90 percent.

On the latest edi­ti­on of the list, the No. 500 sys­tem recor­ded a per­for­mance of 153.6 tera­flops (tril­li­ons of cal­cu­la­ti­ons per second), com­pared with 133.7 teraflop/s six months ago. The last sys­tem on the newest list was lis­ted at posi­ti­on 421 in the pre­vious TOP500. This repres­ents the lowest tur­no­ver rate in the list in two decades.

Other highlights from the 44th list

• Total com­bi­ned per­for­mance of all 500 sys­tems has grown to 309 Pflop/s, com­pared to 274 Pflop/s in June and 250 Pflop/s one year ago. This increase in instal­led per­for­mance also exhi­bits a noti­ceable slow­down in growth com­pared to the pre­vious long-term trend.
• The­re are 50 sys­tems with per­for­mance grea­ter than 1 petaflop/s on the list, up from 37 six months ago.
• The No. 1 sys­tem, Tianhe‑2, and the No. 7 sys­tem, Stam­pe­de, use Intel Xeon Phi pro­ces­sors to speed up their com­pu­ta­tio­nal rate. The No. 2 sys­tem, Titan, and the No. 6 sys­tem, Piz Daint, use NVIDIA GPUs to acce­le­ra­te computation.
• A total of 75 sys­tems on the list are using acce­le­ra­tor/­co-pro­ces­sor tech­no­lo­gy, up from 62 from Novem­ber 2013. Fif­ty of the­se use NVIDIA chips, three use ATI Rade­on, and the­re are now 25 sys­tems with Intel MIC tech­no­lo­gy (Xeon Phi). Intel con­ti­nues to pro­vi­de the pro­ces­sors for the lar­gest share (85.8 per­cent) of TOP500 systems.
• Nine­ty-six per­cent of the sys­tems use pro­ces­sors with six or more cores and 85 per­cent use eight or more cores.
• HP has the lead in sys­tems with 179 (36 per­cent) com­pared to IBM with 153 sys­tems (30 per­cent). HP had 182 sys­tems (36.4 per­cent) six months ago, and IBM had 176 sys­tems (35.2 per­cent) six months ago. In the sys­tem cate­go­ry, Cray remains third with 62 sys­tems (12.4 percent).

About the TOP500 List

The first ver­si­on of what beca­me today’s TOP500 list star­ted as an exer­cise for a small con­fe­rence in Ger­ma­ny in June 1993. Out of curio­si­ty, the aut­hors deci­ded to revi­sit the list in Novem­ber 1993 to see how things had chan­ged. About that time they rea­li­zed they might be on to some­thing and deci­ded to con­ti­nue com­pi­ling the list, which is now a much-anti­ci­pa­ted, much-wat­ched and much-deba­ted twice-year­ly event.

The TOP500 list is com­pi­led by Erich Stroh­mai­er and Horst Simon of Law­rence Ber­ke­ley Natio­nal Labo­ra­to­ry; Jack Don­gar­ra of the Uni­ver­si­ty of Ten­nes­see, Knox­ville; and Mar­tin Meu­er of Pro­me­teus, Germany.