ProHoster > Блог > labaran intanet > Matsayin manyan kwamfutoci masu ƙwaƙƙwara an cika shi ta gungu dangane da ARM CPUs

Matsayin manyan kwamfutoci masu ƙwaƙƙwara an cika shi ta gungu dangane da ARM CPUs

aka buga Mas'ala ta 55 kimantawa 500 mafi kyawun kwamfutoci a duniya. Wani sabon shugaba ne ya yi kima a watan Yuni - gungu na Japan Fugaku, sananne don amfani da masu sarrafa ARM.

Fugaku Cluster aka buga a Cibiyar Nazarin Jiki da Kimiyya RIKEN kuma yana ba da aikin 415.5 petaflops, wanda shine 2.8 fiye da jagoran ƙimar da ta gabata, wanda aka tura zuwa matsayi na biyu. Tarin ya ƙunshi nodes na tushen SoC 158976 Fujitsu A64FX, sanye take da 48-core CPU Armv8.2-A SVE (512 bit SIMD) tare da mitar agogo na 2.2GHz. Gabaɗaya, gungu yana da fiye da nau'ikan sarrafawa sama da miliyan 7 (sau uku fiye da jagoran ƙimar da ta gabata), kusan 5 PB na RAM da 150 PB na ajiyar ajiya da aka raba dangane da Luster FS. Red Hat Enterprise Linux ana amfani dashi azaman tsarin aiki.

Matsayin manyan kwamfutoci masu ƙwaƙƙwara an cika shi ta gungu dangane da ARM CPUs

Wurare daga na biyu zuwa na biyar shugabannin da suka gabata ne suka tabbatar da su, wadanda suka mamaye na daya zuwa na hudu a cikin kimar watan Nuwamba:

  • Wuri na biyu - tari taron tura ta IBM a Oak Ridge National Laboratory (Amurka). Tarin yana gudanar da Red Hat Enterprise Linux kuma ya ƙunshi nau'ikan sarrafawa miliyan 2.4 (ta amfani da 22-core IBM Power9 22C 3.07GHz CPUs da NVIDIA Tesla V100 accelerators), waɗanda ke ba da aikin 148 petaflops.
  • Wuri na uku - gungu na Amurka Sierra, wanda aka shigar a Cibiyar Nazarin Ƙasa ta Livermore ta IBM bisa tsarin dandamali mai kama da taron koli da kuma nuna aikin a 94 petaflops (kimanin 1.5 miliyan cores).
  • Wuri na hudu - gungu na kasar Sin Sunway TaihuLight, aiki a Cibiyar Supercomputing na kasar Sin, ciki har da fiye da miliyan 10 na'urorin kwamfuta da kuma nuna aikin petaflops 93. Duk da makamantan alamun aiki, gungu na Saliyo yana cinye rabin makamashi kamar Sunway TaihuLight.
  • Wuri na biyar - gungu na kasar Sin Tianhe-2A, wanda ya haɗa da kusan nau'i-nau'i miliyan 5 kuma yana nuna aikin 61 petaflops.

Sabbin gungu sun ɗauki matsayi na shida da na bakwai HPC5 (Italiya, Dell EMC, 35 petaflops, 669 dubu cores) da Selene (Amurka, 27 petaflops, 277 dubu cores), wanda ya maye gurbin gungun Amurka. Frontera (Dell EMC, 23 petaflops, 448 dubu cores) zuwa matsayi na takwas. Sabon gungu na Italiya ya ɗauki matsayi na tara Marconi-100 (IBM, petaflops 21.6, cores dubu 347), kuma na goma shine gungu na Swiss Pizz Daint (Cray/HPE, 21.2 petaflops, 387 dubu cores), wanda ya ɗauki matsayi na 6 a cikin matsayi na baya.

Abubuwan da suka fi ban sha'awa:

  • Tarin gida SberCloud (6.6 petaflops, Ubuntu 18.04.01/2/8168, wanda Sberbank ya gina akan dandalin NVIDIA DGX-24, yana amfani da Xeon Platinum 2.7 99600C 6GHz CPU kuma yana da 29 computing cores) ya koma daga 36th zuwa 2th wuri a cikin matsayi a cikin watanni 107. Wani gungu na gida, Lomonosov 131, ya tashi daga XNUMX zuwa XNUMX wuri. Tari a ciki Roshydromet, wanda ke matsayi na 465, an fitar da shi daga matsayin. Don haka, adadin gungu na cikin gida a cikin martaba sama da watanni shida ya ragu daga 3 zuwa 2 (a cikin 2017 akwai 5 a cikin martaba. tsarin gida, kuma a cikin 2012 - 12);

  • Rarraba ta adadin manyan kwamfutoci a ƙasashe daban-daban:
    • China: 226 (228 watanni shida da suka wuce). A cikin duka, gungu na kasar Sin suna samar da kashi 45.2 na yawan yawan amfanin da aka samu (watanni shida da suka gabata - 31.9%);
    • Amurka: 114 (117). An kiyasta jimlar yawan aiki a 22.8% (watanni shida da suka gabata - 37.8%);
    • Japan: 29 (29);
    • Faransa: 19 (18);
    • Jamus: 16 (16);
    • Netherlands: 15 (15);
    • Ireland: 14 (14);
    • Kanada 12 (9);
    • Birtaniya: 10 (11);
    • Italiya: 7 (5);
    • Brazil: 4 (3);
    • Singapore 4 (4);
    • Koriya ta Kudu, Saudi Arabia, Norway: 3;
    • Rasha, Indiya, Australia, UAE, Switzerland, Sweden, Finland, Taiwan: 2.
  • A cikin martabar tsarin aiki da ake amfani da su a cikin manyan kwamfutoci, Linux kawai ya rage tsawon shekaru uku;
  • Rarraba ta rarrabawar Linux (a cikin baka - watanni 6 da suka gabata):
    • 54.4% (49.6%) ba su dalla-dalla rarrabawar,
    • 24.6% (26.4%) suna amfani da CentOS,
    • 6.8% (6.8%) - Cray Linux,
    • 6% (4.8%) - RHEL,
    • 2.6% (3%) - SUSE,
    • 2.2% (2%) - Ubuntu;
    • 0.2% (0.4%) - Linux na Kimiyya
  • Matsakaicin iyakar aiki don shigar da Top500 a cikin watanni 6 ya karu daga 1142 zuwa teraflops 1230 (a shekarar da ta gabata, gungu 272 kawai ya nuna aikin fiye da petaflop, shekaru biyu da suka gabata - 138, shekaru uku da suka gabata - 94). Don Top100, ƙofar shiga ya karu daga 2570 zuwa 2801 teraflops;
  • Jimlar ayyukan duk tsarin a cikin ƙimar ya karu a cikin shekara daga 1.65 zuwa 2.23 exaflops (shekaru uku da suka wuce 749 petaflops). Tsarin da ya rufe matsayi na yanzu ya kasance a matsayi na 449 a cikin fitowar karshe, kuma na 348 a cikin shekarar da ta gabata;
  • Babban rabon adadin manyan kwamfutoci a sassa daban-daban na duniya shine kamar haka.
    272 supercomputers suna cikin Asiya (274 - watanni shida da suka gabata),
    126 a Arewacin Amirka (129) da 96 a Turai (94), 4 a Kudancin Amirka da 2 a Oceania (3);

  • A matsayin tushen processor, Intel CPUs suna kan gaba - 93.8% (watanni shida da suka gabata ya kasance 94%), a matsayi na biyu shine IBM Power - 2.6% (daga 2.8%), a matsayi na uku AMD - 2.2% (0.6% ), a wuri na hudu shine ARM (Marvell ThunderX2 da Fujitsu A64FX) - 0.8%, akan SPARC64 na biyar - 0.2% (0.6%). A karon farko a cikin martaba, an gabatar da wani gungu bisa na'urori masu sarrafa ARM, wanda nan da nan ya fara aiki.
  • 37.4% (watanni 35.6 da suka gabata 20%) na duk na'urorin da aka yi amfani da su suna da nau'ikan 12.2, 13.8% (16%) - cores 10.6, 24% - XNUMX cores,
    10.4% (11%) - 18 cores, 9.8% (11.2%) - 12 cores, 7% (7.8%) - 14 cores;

  • 145 daga cikin 500 tsarin (watanni shida da suka gabata - 144) kuma suna amfani da accelerators ko coprocessors, yayin da tsarin 135 ke amfani da kwakwalwan kwamfuta na NVIDIA (watanni shida da suka gabata akwai 135), 6 - Intel Xeon Phi (akwai 5), 1 - PEZY (1) , 1 yana amfani da mafita na matasan (a baya 1), 1 yana amfani da Matrix-2000 (1), 1 GPU AMD Vega (1);
  • Daga cikin masana'antun gungu, Lenovo ya ɗauki matsayi na farko - 36% (watanni 34.8 da suka gabata XNUMX%), a wuri na biyu
    Sugon 13.6% (14.2%), a matsayi na uku Inspur - 12.8% (13.2%), wuri na hudu yana shagaltar da Hewlett-Packard - 7.6% (7%), Cray 7.2%, Atos - 5.2% (4.6%) , Fujitsu 2.6% (2.6%), IBM 2.4 (2.6%), Dell EMC 2% (2.2%), NVIDIA 1.4% (1.2%), Huawei 1.4% (2%),
    Kwamfuta na Penguin - 1.2% (2.2%). Shekaru uku da suka wuce, rarraba tsakanin masana'antun ya kasance kamar haka: Hewlett-Packard 28.6% (22.4%), a matsayi na biyu shine Lenovo 17% (18.4%), a matsayi na uku shine Cray 11.4% (11.2%), a matsayi na hudu. Sugon 9.2% (9.4%), a matsayi na biyar IBM 5.4% (6.6%).

  • Ana amfani da Ethernet don haɗa nodes a cikin 52.6% na gungu (watanni 52 da suka gabata 30.4%), InfiniBand - 28% (9.8%), Omnipath - 10% (52.6%). Idan muka yi la'akari da jimlar aikin, to, tsarin tushen Ethernet yana rufe 29% (500%) na jimlar Top30.4, InfiniBand - 40% (9.8%), Omnipath - XNUMX%.

A lokaci guda, akwai sabon sakin madadin ƙimar tsarin tari Siffar 500, mayar da hankali kan kimanta aikin dandamali na supercomputer da ke da alaƙa da daidaita tsarin tafiyar da jiki da ayyuka don sarrafa yawancin bayanai na yau da kullun don irin waɗannan tsarin. Rating Green500 daban more ba a sake shi ba kuma an haɗa shi da Top500, kamar yadda ingantaccen makamashi yake yanzu nuna a cikin babban darajar Top500 (ana la'akari da rabon LINPACK FLOPS zuwa amfani da wutar lantarki a watts).

source: budenet.ru

Add a comment