Ushicilelo lwe-58 lwenqanaba leekhompyuter ze-500 ezisebenza kakhulu kwihlabathi lishicilelwe. Ekukhutshweni okutsha, ishumi eliphezulu alizange litshintshe, kodwa amaqela ama-4 amatsha aseRashiya afakwe kwinqanaba.
I-19, i-36 kunye ne-40 indawo kwinqanaba lithathwe ngamaqela aseRashiya aseChervonenkis, eGalushkin naseLyapunov, adalwe nguYandex ukuxazulula iingxaki zokufunda umatshini kunye nokubonelela ngokusebenza kwe-21.5, 16 kunye ne-12.8 petaflops, ngokulandelanayo. Amaqela aqhuba Ubuntu 16.04 kwaye axhotyiswe nge-AMD EPYC 7xxx processors kunye neNVIDIA A100 GPUs: iCluster yaseChervonenkis inamaqhuqhuva angama-199 (193 amawaka e-AMD EPYC 7702 64C 2GH cores kunye ne-1592 NVIDIAGkin A100 80 I-AMD136), i-NVIDIAGkin A134 7702 64 GPY2, i-AMD 1088, i-AMD 100C cores 80 137C 130GH kunye 7662 GPU NVIDIA A64 2G), Lyapunov - 1096 nodes (100 amawaka cores AMD EPYC 40 XNUMXC XNUMXGHz kunye XNUMX GPU NVIDIA AXNUMX XNUMXG).
Kwindawo ye-43 kwakukho iqela elitsha le-Sberbank, u-Christofari Neo, oqhuba i-NVIDIA DGX OS 5 (uhlelo lwe-Ubuntu) kunye nokubonisa ukusebenza kwe-petaflops ye-11.9. Iqela linama-cores angaphezu kwama-98 amawaka ekhompyuter esekwe kwi-AMD EPYC 7742 64C 2.25GHz CPU kwaye iza ne-NVIDIA A100 80GB GPU. Iqela le-Sberbank Christofari elaphunyezwa ngaphambili lisuka kwi-61 ukuya kwindawo ye-72 kwindawo yokubeka kwisiqingatha sonyaka.
Amaqela amabini asekhaya angaphezulu nawo ahlala kwirenki: iLomonosov 2 - isuka kwi-199 ukuya ku-241 indawo (ngo-2015, i-Lomonosov 2 cluster yathatha indawo ye-31, kunye ne-Lomonosov eyandulelayo ngo-2011 - indawo ye-13) kunye ne-MTS GROM - isuka kwi-240 ukuya ku-294 indawo . Ngaloo ndlela, inani lamaqela asekhaya kwinqanaba lenyuka ukusuka kwi-3 ukuya kwi-7 kwiinyanga ezintandathu (ukuthelekisa, ngo-2020 kwakukho iinkqubo ezi-2 zasekhaya kwinqanaba, ngo-2017 - 5, kwaye ngo-2012 - 12).
Ngokuphathelele ukulinganisa ngokubanzi, iqela laseJapan leFugaku, elakhiwe kusetyenziswa iiprosesa ze-ARM, lihlala lisendaweni yokuqala. Iqela le-Fugaku lifumaneka kwi-RIKEN Institute for Physical and Chemical Research kwaye libonelela ngokusebenza kwe-442 petaflops. Iqela libandakanya i-158976 nodes esekelwe kwi-Fujitsu A64FX SoC, exhotywe nge-48-core Armv8.2-A SVE CPU (512 bit SIMD) kunye ne-clock frequency ye-2.2GHz. Iyonke, i-cluster ine-cores ye-processor engaphezulu kwe-7.6 yezigidi (izihlandlo ezithathu ngaphezu kwenkokeli yangaphambili), i-5 PB ye-RAM kunye ne-150 PB yokugcinwa okwabelwanayo ngokusekelwe kwi-Luster FS. I-Red Hat Enterprise Linux isetyenziswa njengenkqubo yokusebenza. Ubude obupheleleyo beentambo ze-optical ezisetyenziselwa ukudibanisa ama-nodes malunga neekhilomitha ezingama-850.
Kwindawo yesibini liqela leNgqungquthela, elasetyenziswa yi-IBM e-Oak Ridge National Laboratory (USA). Iqela liqhuba iRed Hat Enterprise Linux kwaye ibandakanya i-2.4 yezigidi zeeprosesa (22-core IBM Power9 22C 3.07GHz CPUs kunye neNVIDIA Tesla V100 accelerators zisetyenziswa), ezibonelela ngokusebenza kwe-148 petaflops, ephantse ibe kathathu ngaphantsi kwenkokeli umlinganiselo.
Indawo yesithathu ibanjwe yi-American Sierra cluster, efakwe kwi-Livermore National Laboratory yi-IBM ngesiseko seqonga elifana neNgqungquthela kunye nokubonisa ukusebenza kwe-94 petaflops (malunga ne-1.5 yezigidi ze-cores). I-Red Hat Enterprise Linux isetyenziswa njengenkqubo yokusebenza.
Kwindawo yesine yi-Chinese Sunway TaihuLight cluster, esebenza kwi-National Supercomputer Centre yaseChina, kubandakanywa ngaphezu kwe-10 yezigidi ze-computing cores kunye nokubonisa ukusebenza kwe-93 petaflops. Ngaphandle kwezalathi zokusebenza ezifanayo, iqela laseSierra lidla isiqingatha samandla amaninzi njengeSunway TaihuLight. Isixokelelwano sokusebenza sisibani seLinux usasazo RaiseOS.
Kwindawo yesihlanu yi-Perlmutter cluster, eveliswe yi-HPE kwaye ibekwe kwiziko leSizwe loPhando lwaMandla e-USA. Iqela libandakanya ama-cores angama-761 amawaka asekelwe kwi-AMD EPYC 7763 64C 2.45GHz CPU kwaye ibonelela ngokusebenza kweepetaflops ezingama-71. Inkqubo yokusebenza yiCray OS.
Eyona ndlela inomdla kakhulu:
- Ukuhanjiswa ngenani leekhompyuter ezinkulu kumazwe ahlukeneyo:
- I-China: 173 (188 - kwiinyanga ezintandathu ezidlulileyo). Iyonke, amaqela aseTshayina avelisa i-17.5% yayo yonke imveliso (iinyanga ezintandathu ezidlulileyo - 19.4%);
- USA: 149 (122). Iyonke imveliso iqikelelwa kwi-32.5% (kwiinyanga ezintandathu ezidlulileyo - 30.7%);
- EJapan: 32 (34);
- EJamani: 26 (23);
- EFransi: 19 (16);
- ENetherlands: 11 (16);
- E-UK: 11 (11);
- eKhanada ali-11 (11);
- iRashiya 7 (3);
- Mzantsi Korea 7 (5)
- EItali: 6 (6);
- iSaudi Arabia 6 (6);
- Brazil 5 (6);
- eSweden 4 (3);
- Poland 4 (4);
- EOstreliya, eIndiya, eSwitzerland, eFinland: 3.
- Kwinqanaba leenkqubo zokusebenza ezisetyenziswa kwii-supercomputers, i-Linux kuphela ihleli iminyaka emine enesiqingatha;
- Ukuhanjiswa kweLinux (kwizibiyeli - kwiminyaka emibini eyadlulayo):
- I-51.6% (49.6%) ayinazinkcukacha ngosasazo,
- I-18% (26.4%) isebenzisa i-CentOS,
- 7.6% (4.8%) - RHEL,
- 7% (6.8%) - Cray Linux,
- 5.4% (2%) - Ubuntu;
- 4% (3%) - SUSE,
- 0.2% (0.4%) - I-Linux yezeNzululwazi
- Ubuncinci bomgangatho womsebenzi wokungena kwi-Top500 kwiinyanga ezi-6 zonyuka ukusuka kwi-1511 ukuya kwi-teraflops ye-1649 (kwiminyaka emithathu edlulileyo, amaqela angama-272 kuphela abonisa ukusebenza okungaphezulu kwe-petaflop, kwiminyaka emine eyadlulayo - 138, kwiminyaka emihlanu eyadlulayo - 94). Kwi-Top100, umda wokungena unyuke ukusuka kwi-4124 ukuya kwi-4788 teraflops;
- Ukusebenza okupheleleyo kwazo zonke iinkqubo zokulinganisa kunyuke ukusuka kwi-2.8 ukuya kwi-3 exaflops kunyaka (iminyaka emibini edlulileyo yayiyi-1.650 exaflops, kwaye kwiminyaka emihlanu eyadlulayo - i-566 petaflops). Inkqubo evala indawo yangoku yayikwindawo ye-433 kwinqaku lokugqibela, kwaye kwindawo ye-401 kunyaka ongaphambili;
- Ukusasazwa ngokubanzi kwenani lee-supercomputers kwiindawo ezahlukeneyo zehlabathi ngolu hlobo lulandelayo: ii-supercomputer ze-226 zifumaneka e-Asia (245 - kwiinyanga ezintandathu ezidlulileyo), i-160 eMntla Melika (133) kunye ne-105 eYurophu (113), i-5 eMzantsi. IMelika (6), i-3 e-Oceania (2) kunye ne-1 e-Afrika (1);
- Njengesiseko somprosesa, ii-Intel CPUs zihamba phambili - 81.6% (kwiminyaka emibini edlulileyo yayiyi-94%), i-AMD ikwindawo yesibini nge-14.6% (0.6% !!), kwaye i-IBM Power ikwindawo yesithathu nge-1.4% ( ibiyi 2.8%). Kukho ukukhula okusebenzayo kwamaqela asekwe kwiiprosesa ze-AMD; umzekelo, zonke iinkqubo ezintsha ezibandakanyiweyo kwi-Top15 zixhotyiswe ngee-AMD CPUs.
- I-26.6% (iminyaka emibini edlulileyo 35.6%) yazo zonke iiprosesa ezisetyenzisiweyo zine-20 cores, 17.6% - 24 cores, 11.2% - 64 cores, 8.6% (13.8%) - 16 cores, 8.2% (11%) - 18 cores, 5.8 I-% (11.2%) - i-12 cores.
- I-149 kwiinkqubo ze-500 (iminyaka emibini edlulileyo - i-144) iphinda isebenzise i-accelerators okanye i-coprocessors, kunye neenkqubo ze-143 zisebenzisa i-chips ze-NVIDIA, i-2 - i-Intel Xeon Phi (ukusuka kwi-5), i-1 - i-PEZY (1), kunye ne-1 AMD Vega GPU;
- Phakathi kwabavelisi beqela, iLenovo ibambe indawo yokuqala - 36.8% (kwiminyaka emibini edlulileyo 34.8%), i-Inspur ibambe indawo yesibini - 11.6% (13.2%), iHewlett-Packard Enterprise ithathe indawo yesithathu - 9% (7%), ilandelwa nguSugon 7.8 % (14.2%), Atos - 7.2% (4.6%), Cray 6.4% (7%), Dell EMC 3.2% (2.2%), Fujitsu 3% (2.6%), NVIDIA 2.4 (1.2%), NEC 2% , Huawei 1.4% (2%), IBM 1.4% (2.6%), iPenguin Computing - 1.4% (2.2%). Kwiminyaka eyisixhenxe edlulileyo, ukuhanjiswa phakathi kwabavelisi kwakumi ngolu hlobo lulandelayo: Hewlett-Packard 36%, IBM 35%, Cray 10.2% kunye ne-SGI 3.8%;
- I-Ethernet isetyenziselwa ukudibanisa ama-nodes kwi-49.4% (iminyaka emibini edlulileyo i-52%) yamaqela, i-InfiniBand isetyenziswe kwi-33.6% (28%) yamaqela, i-Omnipath - 8.4% (10%). Ukujonga ukusebenza ngokubanzi, iinkqubo ze-InfiniBand-based account ye-43.3% ye-Top500 yokusebenza ngokubanzi, ngelixa i-Ethernet i-akhawunti ye-21.3%.
Kwixesha elizayo elingekude, upapasho lohlelo olutsha lolunye uhlengahlengiso lweenkqubo zeqela iGrafu 500, egxile ekuvavanyeni ukusebenza kwamaqonga eekhompyutha ezinkulu anxulumene nokulinganisa iinkqubo zomzimba kunye nemisebenzi yokusetyenzwa kweedatha ezininzi eziqhelekileyo kwiinkqubo ezinjalo, kulindeleke. I-Green500, i-HPCG (i-High-Performance Conjugate Gradient) kunye ne-HPL-AI yokulinganisa idibaniswe ne-Top500 kwaye ibonakaliswe kwi-Top500 yokukala.
umthombo: opennet.ru