{"id":52852,"date":"2019-11-18T16:51:10","date_gmt":"2019-11-18T15:51:10","guid":{"rendered":"https:\/\/www.planet3dnow.de\/cms\/?p=52852"},"modified":"2019-11-18T16:51:10","modified_gmt":"2019-11-18T15:51:10","slug":"china-extends-lead-in-number-of-top500-supercomputers-us-holds-on-to-performance-advantage","status":"publish","type":"post","link":"https:\/\/www.planet3dnow.de\/cms\/52852-china-extends-lead-in-number-of-top500-supercomputers-us-holds-on-to-performance-advantage\/","title":{"rendered":"China Extends Lead in Number of <span class=\"caps\">TOP500<\/span> Supercomputers, <span class=\"caps\">US<\/span> Holds on to Performance Advantage"},"content":{"rendered":"<p><span class=\"caps\">BERKELEY<\/span>, Calif.; <span class=\"caps\">FRANKFURT<\/span>, Ger\u00adma\u00adny; and <span class=\"caps\">KNOXVILLE<\/span>, Tenn.\u2014 The 54th edi\u00adti\u00adon of the <span class=\"caps\">TOP500<\/span> saw Chi\u00adna and the <span class=\"caps\">US<\/span> main\u00adtai\u00adning their domi\u00adnan\u00adce of the list, albeit in dif\u00adfe\u00adrent cate\u00adgo\u00adries. Mean\u00adwhile, the aggre\u00adga\u00adte per\u00adfor\u00admance of the 500 sys\u00adtems, based on the High Per\u00adfor\u00admance Lin\u00adpack (<span class=\"caps\">HPL<\/span>) bench\u00admark, con\u00adti\u00adnues to rise and now sits at 1.65 exa\u00adflops. The ent\u00adry level to the list has risen to 1.14 peta\u00adflops, up from 1.02 peta\u00adflops in the pre\u00advious list in June&nbsp;2019.<\/p>\n<p><strong>Top-ran\u00adked systems<\/strong><\/p>\n<p>The top of the list remains lar\u00adge\u00adly unch\u00adan\u00adged. In fact, the top 10 sys\u00adtems are unch\u00adan\u00adged from the pre\u00advious&nbsp;list.<\/p>\n<p>Sum\u00admit and Sier\u00adra remain in the top two spots. Both are IBM-built super\u00adcom\u00adpu\u00adters employ\u00ading Power9 CPUs and <span class=\"caps\">NVIDIA<\/span> Tes\u00adla <span class=\"caps\">V100<\/span> GPUs. Oak Ridge Natio\u00adnal Laboratory\u2019s Sum\u00admit sys\u00adtem holds top honors with an <span class=\"caps\">HPL<\/span> result of 148.6 peta\u00adflops. The second-ran\u00adked Sier\u00adra sys\u00adtem at Law\u00adrence Liver\u00admo\u00adre Natio\u00adnal Labo\u00adra\u00adto\u00adry comes in at 94.6 petaflops.<\/p>\n<p>Clo\u00adse behind at num\u00adber three is the Sun\u00adway Tai\u00adhu\u00adLight super\u00adcom\u00adpu\u00adter, with an <span class=\"caps\">HPL<\/span> mark of 93.0 peta\u00adflops. Tai\u00adhu\u00adLight was deve\u00adlo\u00adped by China\u2019s Natio\u00adnal Rese\u00adarch Cen\u00adter of Par\u00adal\u00adlel Com\u00adpu\u00adter Engi\u00adnee\u00adring <span class=\"amp\">&amp;<\/span> Tech\u00adno\u00adlo\u00adgy (<span class=\"caps\">NRCPC<\/span>) and is instal\u00adled at the Natio\u00adnal Super\u00adcom\u00adpu\u00adting Cen\u00adter in Wuxi. It is powered exclu\u00adsi\u00adve\u00adly by Sunway\u2019s <span class=\"caps\">SW26010<\/span> processors.<\/p>\n<p>Tian\u00adhe-2A (Mil\u00adky Way-2A), a sys\u00adtem deve\u00adlo\u00adped by China\u2019s Natio\u00adnal Uni\u00adver\u00adsi\u00adty of Defen\u00adse Tech\u00adno\u00adlo\u00adgy (<span class=\"caps\">NUDT<\/span>) and deploy\u00aded at the Natio\u00adnal Super\u00adcom\u00adpu\u00adter Cen\u00adter in Guang\u00adzhou, Chi\u00adna, holds the num\u00adber four spot with 61.4 peta\u00adflops. It is powered by Intel Xeon CPUs and Matrix-2000 accelerators.<\/p>\n<p>Fron\u00adte\u00adra, a Dell <span class=\"caps\">C6420<\/span> sys\u00adtem, keeps it num\u00adber five ran\u00adking, with an <span class=\"caps\">HPL<\/span> result of 23.5 peta\u00adflops. Powered sole\u00adly by Xeon Pla\u00adti\u00adnum pro\u00adces\u00adsors, it was instal\u00adled at the Texas Advan\u00adced Com\u00adpu\u00adting Cen\u00adter of the Uni\u00adver\u00adsi\u00adty of Texas in&nbsp;2018.<\/p>\n<p>At num\u00adber six is Piz Daint, a Cray <span class=\"caps\">XC50<\/span> super\u00adcom\u00adpu\u00adter instal\u00adled at the Swiss Natio\u00adnal Super\u00adcom\u00adpu\u00adting Cent\u00adre (<span class=\"caps\">CSCS<\/span>) in Luga\u00adno, Switz\u00ader\u00adland. At 21.2 peta\u00adflops, it remains the most powerful sys\u00adtem in Europe.<\/p>\n<p>Tri\u00adni\u00adty, a Cray <span class=\"caps\">XC40<\/span> sys\u00adtem ope\u00adra\u00adted by Los Ala\u00admos Natio\u00adnal Labo\u00adra\u00adto\u00adry and San\u00addia Natio\u00adnal Labo\u00adra\u00adto\u00adries is ran\u00adked num\u00adber seven, with an <span class=\"caps\">HPL<\/span> per\u00adfor\u00admance of 20.2 peta\u00adflops.&nbsp; It is powered by Intel Xeon and Xeon Phi processors.<\/p>\n<p>Lis\u00adted at num\u00adber eight is the <span class=\"caps\">AI<\/span> Bridging Cloud Infra\u00adstruc\u00adtu\u00adre (<span class=\"caps\">ABCI<\/span>), instal\u00adled at Japan\u2019s Natio\u00adnal Insti\u00adtu\u00adte of Advan\u00adced Indus\u00adtri\u00adal Sci\u00adence and Tech\u00adno\u00adlo\u00adgy (<span class=\"caps\">AIST<\/span>). The Fuji\u00adtsu-built sys\u00adtem is equip\u00adped with Intel Xeon Gold pro\u00adces\u00adsors and <span class=\"caps\">NVIDIA<\/span> Tes\u00adla <span class=\"caps\">V100<\/span> GPUs, achie\u00adving an <span class=\"caps\">HPL<\/span> result of 19.9 petaflops.<\/p>\n<p>SuperMUC-NG is in the num\u00adber nine posi\u00adti\u00adon with 19.5 peta\u00adflops. It\u2019s instal\u00adled at the Leib\u00adniz-Rechen\u00adzen\u00adtrum (Leib\u00adniz Super\u00adcom\u00adpu\u00adting Cent\u00adre) in Gar\u00adching, near Munich. The Leno\u00advo-built machi\u00adne is powered by Intel Pla\u00adti\u00adnum Xeon processors.<\/p>\n<p>The Las\u00adsen super\u00adcom\u00adpu\u00adter is ran\u00adked num\u00adber 10, deli\u00adve\u00adring 18.2 peta\u00adflops. Instal\u00adled at Law\u00adrence Liver\u00admo\u00adre Natio\u00adnal Labo\u00adra\u00adto\u00adry, Las\u00adsen is the unclas\u00adsi\u00adfied coun\u00adter\u00adpart to the clas\u00adsi\u00adfied Sier\u00adra sys\u00adtem and shares the same <span class=\"caps\">IBM<\/span> Power9\/<span class=\"caps\">NVIDIA<\/span> <span class=\"caps\">V100<\/span> <span class=\"caps\">GPU<\/span> architecture.<\/p>\n<p>The most powerful new super\u00adcom\u00adpu\u00adter on the list is AiMOS, which shows up at posi\u00adti\u00adon 25, tur\u00adning in an <span class=\"caps\">HPL<\/span> result of 8.0 peta\u00adflops. The IBM-built sys\u00adtem is instal\u00adled at the Rens\u00adse\u00adlaer Poly\u00adtech\u00adnic Insti\u00adtu\u00adte Cen\u00adter for Com\u00adpu\u00adta\u00adtio\u00adnal Inno\u00adva\u00adtions (<span class=\"caps\">CCI<\/span>), and like Sum\u00admit, Sier\u00adra, and Las\u00adsen, it is equip\u00adped with Power9 CPUs and <span class=\"caps\">NVIDIA<\/span> <span class=\"caps\">V100<\/span>&nbsp;GPUs.&nbsp;<\/p>\n<p><strong>Geo\u00adgra\u00adphi\u00adcal breakdown<\/strong><\/p>\n<p>The num\u00adber of <span class=\"caps\">TOP500<\/span> instal\u00adla\u00adti\u00adons in Chi\u00adna con\u00adti\u00adnues to rise and now sits at 227, up from 219 six months ago. Mean\u00adwhile, the share of US-based sys\u00adtem remains near its all-time low at 118. Howe\u00adver, sys\u00adtems in the <span class=\"caps\">US<\/span> are on avera\u00adge signi\u00adfi\u00adcant\u00adly lar\u00adger, which trans\u00adla\u00adted to a 37.8 per\u00adcent share of the list\u2019s aggre\u00adga\u00adte per\u00adfor\u00admance. Chi\u00adna is clo\u00adse behind with a 31.9 per\u00adcent per\u00adfor\u00admance share. Howe\u00adver, com\u00adpared to six months ago, this per\u00adfor\u00admance gap has shrunk. The June 2019 list had the <span class=\"caps\">US<\/span> with a 38.4 per\u00adcent of the list\u2019s aggre\u00adga\u00adte per\u00adfor\u00admance and Chi\u00adna with 29.9 percent.<\/p>\n<p>Japan remains in third place in the num\u00adber of <span class=\"caps\">TOP500<\/span> sys\u00adtems, with 29, fol\u00adlo\u00adwed by France with 18, Ger\u00adma\u00adny with 16, the Net\u00adher\u00adlands with 15, Ire\u00adland with 14, and the United King\u00addom with 11. All other count\u00adries were in the sin\u00adgle digits.<\/p>\n<p><strong>Ven\u00addor shares<\/strong><\/p>\n<p>&nbsp;As a reflec\u00adtion of China\u2019s domi\u00adnan\u00adce in sheer num\u00adbers, the top three sys\u00adtem ven\u00addors with regard to the num\u00adber of instal\u00adla\u00adti\u00adons are Leno\u00advo, (174), Sugon (71), and Inspur (65). Cray is num\u00adber four, with 36 sys\u00adtems, and <span class=\"caps\">HPE<\/span> is num\u00adber five, with 35. Note that Cray is now part of <span class=\"caps\">HPE<\/span>, so taken tog\u00ade\u00adther they would effec\u00adtively tie Sugon with 71 systems.<\/p>\n<p>At the chip level, Intel con\u00adti\u00adnues its domi\u00adnan\u00adce. Its pro\u00adces\u00adsors are pre\u00adsent in 470 of the 500 sys\u00adtems, split bet\u00adween mul\u00adti\u00adple gene\u00adra\u00adti\u00adons of Xeon and Xeon Phi hard\u00adware. <span class=\"caps\">IBM<\/span> is second with 14 sys\u00adtems \u2013 10 with Power CPUs and four with Blue Gene\/PowerPC CPUs. <span class=\"caps\">AMD<\/span> claims just three sys\u00adtems on the cur\u00adrent&nbsp;list.<\/p>\n<p>The\u00adre are now two Arm-based super\u00adcom\u00adpu\u00adters on the list: the Astra sys\u00adtem deploy\u00aded at San\u00addia Natio\u00adnal Labo\u00adra\u00adto\u00adries, which is equip\u00adped with Marvell\u2019s ThunderX2 pro\u00adces\u00adsors, and Fujitsu\u2019s <span class=\"caps\">A64FX<\/span> pro\u00adto\u00adty\u00adpe sys\u00adtem, a pre\u00adcur\u00adsor to the Fug\u00ada\u00adku (Post\u2011K) exas\u00adca\u00adle sys\u00adtem desti\u00adned for <span class=\"caps\">RIKEN<\/span> in 2021. The Fuji\u00adtsu machi\u00adne is new to the list and occu\u00adp\u00adies posi\u00adti\u00adon 159, with an <span class=\"caps\">HPL<\/span> per\u00adfor\u00admance of 2.0 petaflops.<\/p>\n<p><span class=\"caps\">NVIDIA<\/span> is the domi\u00adnant ven\u00addor for acce\u00adle\u00adra\u00adtors. Its GPUs are pre\u00adsent in 136 of the 145 acce\u00adle\u00adra\u00adted sys\u00adtems. On the pre\u00advious list six months ago, the\u00adre were 134 acce\u00adle\u00adra\u00adted systems.<\/p>\n<p><strong>Inter\u00adcon\u00adnect breakdown<\/strong><\/p>\n<p>Ether\u00adnet is used in 52 per\u00adcent (258) of the <span class=\"caps\">TOP500<\/span> sys\u00adtems, while Infi\u00adni\u00adBand is the net\u00adwork-of-choice in 28 per\u00adcent (140) of sys\u00adtems. Howe\u00adver, from a per\u00adfor\u00admance per\u00adspec\u00adti\u00adve, tho\u00adse posi\u00adti\u00adons are rever\u00adsed, with Infi\u00adni\u00adBand-based machi\u00adnes repre\u00adsen\u00adting 40 per\u00adcent of the <span class=\"caps\">TOP500<\/span>\u2019s aggre\u00adga\u00adte per\u00adfor\u00admance and Ether\u00adnet-based machi\u00adnes with 29 per\u00adcent. Cus\u00adtom inter\u00adcon\u00adnects, with just 46 instal\u00adla\u00adti\u00adons, cla\u00adim 22 per\u00adcent of the list\u2019s instal\u00adled performance.<\/p>\n<p><strong>Green500 results<\/strong><\/p>\n<p>The rela\u00adted Green500 list, which mea\u00adsu\u00adres ener\u00adgy-effi\u00adci\u00aden\u00adcy of the world\u2019s top super\u00adcom\u00adpu\u00adters, has chan\u00adged con\u00adsider\u00ada\u00adb\u00adly com\u00adpared to six months ago. Num\u00adber one on the list is the afo\u00adre\u00admen\u00adtio\u00adned <span class=\"caps\">A64FX<\/span> pro\u00adto\u00adty\u00adpe super\u00adcom\u00adpu\u00adter, which deli\u00adver\u00aded 16.9 gigaflops\/watt. Just slight\u00adly less green at num\u00adber two sys\u00adtem is <span class=\"caps\">NA<\/span>\u20111, a Zet\u00adtas\u00adca\u00adler machi\u00adne that uses <span class=\"caps\">PEZY<\/span> Computing\u2019s <span class=\"caps\">PEZY-SC2<\/span> pro\u00adces\u00adsors and deli\u00advers 16.3 gigaflops\/watt. It is being rea\u00addi\u00aded for a future instal\u00adla\u00adti\u00adon at <span class=\"caps\">NA<\/span> Simu\u00adla\u00adti\u00adon in&nbsp;Japan.<\/p>\n<p>The third-ran\u00adked Green500 sys\u00adtem is the new AiMOS sys\u00adtem from <span class=\"caps\">IBM<\/span>, fol\u00adlo\u00adwed by two more <span class=\"caps\">IBM<\/span> sys\u00adtems based on the same Power9\/<span class=\"caps\">NVIDIA<\/span> <span class=\"caps\">V100<\/span> design: Sato\u00adri at 15.6 gigaflops\/watt and Sum\u00admit at 14.7 gigaflops\/watt. The remai\u00adning top 10 Green500 sys\u00adtems on the list \u2013 <span class=\"caps\">AI<\/span> Bridging Cloud Infra\u00adstruc\u00adtu\u00adre, MareN\u00ados\u00adtrum <span class=\"caps\">P9<\/span> <span class=\"caps\">CTE<\/span>, <span class=\"caps\">TSUBAME<\/span> 3.0, <span class=\"caps\">PANGEA<\/span> <span class=\"caps\">III<\/span>, and Sier\u00adra \u2013 also use <span class=\"caps\">NVIDIA<\/span> <span class=\"caps\">GPU<\/span> accelerators.<\/p>\n<p><strong><span class=\"caps\">HPCG<\/span> results<\/strong><\/p>\n<p>The two top-ran\u00adked Sum\u00admit and Sier\u00adra super\u00adcom\u00adpu\u00adters on the <span class=\"caps\">TOP500<\/span>, also remain in the top two spots on the list based on the High-Per\u00adfor\u00admance Con\u00adju\u00adga\u00adte Gra\u00addi\u00adent (<span class=\"caps\">HPCG<\/span>) bench\u00admark. Sum\u00admit achie\u00adved 2.93 HPCG-peta\u00adflops, with Sier\u00adra at 1.80 HPCG-peta\u00adflops. All the remai\u00adning top 10 <span class=\"caps\">HPCG<\/span> ent\u00adries, deli\u00adver\u00aded less than one HPCG-peta\u00adflops. With the excep\u00adti\u00adon of the now-decom\u00admis\u00adsio\u00adned K\u2011computer, all 10 of the\u00adse sys\u00adtems car\u00adri\u00aded over from the pre\u00advious list six months ago.<\/p>\n<p><strong>About the <span class=\"caps\">TOP500<\/span> List<\/strong><\/p>\n<p>The first ver\u00adsi\u00adon of what beca\u00adme today\u2019s <span class=\"caps\">TOP500<\/span> list star\u00adted as an exer\u00adcise for a small con\u00adfe\u00adrence in Ger\u00adma\u00adny in June 1993. Out of curio\u00adsi\u00adty, the aut\u00adhors deci\u00added to revi\u00adsit the list in Novem\u00adber 1993 to see how things had chan\u00adged. About that time, they rea\u00adli\u00adzed they might be onto some\u00adthing and deci\u00added to con\u00adti\u00adnue com\u00adpi\u00adling the list, which is now a much-anti\u00adci\u00adpa\u00adted, much-wat\u00adched, and much-deba\u00adted twice-year\u00adly&nbsp;event.<\/p>\n","protected":false},"excerpt":{"rendered":"<p><span class=\"caps\">BERKELEY<\/span>, Calif.; <span class=\"caps\">FRANKFURT<\/span>, Ger\u00adma\u00adny; and <span class=\"caps\">KNOXVILLE<\/span>, Tenn.\u2014 The 54th edi\u00adti\u00adon of the <span class=\"caps\">TOP500<\/span> saw Chi\u00adna and the <span class=\"caps\">US<\/span> main\u00adtai\u00adning their domi\u00adnan\u00adce of the list, albeit in dif\u00adfe\u00adrent cate\u00adgo\u00adries. Mean\u00adwhile, the aggre\u00adga\u00adte per\u00adfor\u00admance of the 500 sys\u00adtems, based on the High Per\u00adfor\u00admance Lin\u00adpack (<span class=\"caps\">HPL<\/span>) bench\u00admark, con\u00adti\u00adnues to rise and now sits at 1.65 exa\u00adflops. The ent\u00adry (\u2026) <a class=\"moretag\" href=\"https:\/\/www.planet3dnow.de\/cms\/52852-china-extends-lead-in-number-of-top500-supercomputers-us-holds-on-to-performance-advantage\/\">Wei\u00adter\u00adle\u00adsen&nbsp;\u00bb<\/a><\/p>\n","protected":false},"author":593,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"wp_typography_post_enhancements_disabled":false,"ngg_post_thumbnail":0,"footnotes":""},"categories":[22],"tags":[1598,759],"class_list":["post-52852","post","type-post","status-publish","format-standard","hentry","category-pressemitteilungen","tag-supercomputer","tag-top500","entry"],"share_on_mastodon":{"url":"","error":""},"_links":{"self":[{"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/posts\/52852","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/users\/593"}],"replies":[{"embeddable":true,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/comments?post=52852"}],"version-history":[{"count":1,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/posts\/52852\/revisions"}],"predecessor-version":[{"id":52853,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/posts\/52852\/revisions\/52853"}],"wp:attachment":[{"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/media?parent=52852"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/categories?post=52852"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/tags?post=52852"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}