Ushicilelo lwe-60 lwe-500 ephezulu sele ishicilelwe

Emva kweenyanga ezi-6 zokupapashwa kwenombolo yangaphambili kunye nokuthobela ikhalenda yopapasho, i uhlelo olutsha lwama-60 loluhlu lweekhompyuter ezingama-500 eziqhuba kakhulu kwihlabathi.

Kuhlelo olutsha, kukho utshintsho olunye kuphela kwabalishumi abaphezulu: iqela likaLeonardo, elikwiziko lophando lase-Italiya I-CINECA, ibekwe kwindawo yesine. Iqela libandakanya phantse i-1,5 yezigidi zeprosesa (i-Xeon Platinum 8358 32C 2,6 GHz CPU) kwaye ibonelela ngokusebenza kwe-255,75 petaflops.

Iindawo ezilishumi eziphezulu ziquka:

1. I-Frontier, ekwiSebe laseMelika le-Oak Ridge National Laboratory, eli qela linee-processor cores eziphantse zibe zizigidi ezisi-9 (64GHz AMD EPYC 2C CPU, AMD Instinct MI250X accelerator) kwaye ibonelela nge-1.102 exaflops yokusebenza, ephantse yaphinda kathathu ngaphezulu kweyesibini. indawo yeqela.
2. I-Fugaku, ehlala kwi-RIKEN Institute for Physical and Chemical Research (eJapan). Iqela lakhiwe ngeeprosesa ze-ARM (i-158976 nodes esekwe kwi-Fujitsu A64FX SoC, exhotyiswe nge-8.2-core 48GHz Armv2,2-A SVE CPU) enika i-442 petaflops yokusebenza.
3. I-LUMI ibanjwe kwi-European Supercomputing Centre (EuroHPC) eFinland kwaye inikezela nge-151 petaflops yokusebenza. Iqela lisekelwe kwiqonga elifanayo le-HPE Cray EX235a njengenkokeli kwinqanaba, kodwa libandakanya i-processor cores ye-1,1 yezigidi (i-AMD EPYC 64C 2GHz, i-AMD Instinct MI250X i-accelerator, inethiwekhi ye-Slingshot-11).
4. ULeonardo ubambe i-EuroHPC eyahlukileyo eCINECA, eItali. Yinkqubo ye-Atos BullSequana XH2000 ene-Xeon Platinum 8358 32C 2.6GHz njengeeprosesa eziphambili, i-NVIDIA A100 SXM4 40 GB njenge-accelerators kunye ne-Quad-rail NVIDIA HDR100 Infiniband njenge-interconnect. Iphumelele ukusebenza kwe-Linpack ye-174,7 Pflop / s.
5. Ingqungquthela, eyakhiwe yi-IBM kwaye igcinwe kwi-Oak Ridge National Laboratory (ORNL) eTennessee, eU.SA, ngoku ibekwe kwi-#5 kunye nokusebenza kwe-148,8 Pflop / s kwi-benchmark ye-HPL, esetyenziselwa ukubeka uluhlu lwe-TOP500.
6. ISierra, ebanjwe kwiLabhoratri yeSizwe yeLawrence Livermore, CA, USA, i-architecture yayo ifana kakhulu nenkqubo yeNgqungquthela #5. Yakhiwe ngeenodi ze-4320 ezine-POWER9 CPU ezimbini kunye ne-NVIDIA Tesla V100 GPUs ezine. ISierra yenza i-94,6 Pflop/s.
7. I-Sunway TaihuLight, inkqubo ephuhliswe yi-National Parallel Computer Engineering and Technology Research Centre (NRCPC) kwaye ifakwe kwi-National Supercomputing Centre e-Wuxi, iphondo lase-China le-Jiangsu, inqanaba le-7 kunye ne-93 Pflop / s.
8. I-Perlmutter kwi-#8 isekwe kwiqonga le-HPE Cray "Shasta" kunye nenkqubo engafaniyo ne-AMD EPYC-based nodes kunye ne-1536 NVIDIA A100 iinodi ezikhawulezayo. UPerlmutter ubethe i64,6 Pflop/s
9. I-Selene ngoku kwiNombolo ye-9 yi-NVIDIA DGX A100 SuperPOD efakwe endlwini kwi-NVIDIA e-US. Inkqubo isekelwe kwiprosesa ye-AMD EPYC ene-NVIDIA A100 yokukhawulezisa kunye ne-Mellanox HDR InfiniBand njengenethiwekhi kwaye iphumelele i-63,4 Pflop / s.
10. I-Tianhe-2A (i-Milky Way-2A), inkqubo ephuhliswe yi-China National University of Defense Technology (NUDT) kwaye iphunyezwe kwi-National Supercomputer Centre e-Guangzhou, e-China, ngoku ibhalwe njengenkqubo ye-10 kunye ne-61,4, i-XNUMX Pflop / s. .

Ngokubhekiselele kwii-supercomputers zasekhaya, amaqela I-Chervonenkis, iGalushkin kunye neLyapunov eyenziwe nguYandex yawa ukusuka kwiindawo ze-22, 40 kunye ne-43 ukuya kwiindawo ze-25, 44 kunye ne-47. La maqela enzelwe ukusombulula iingxaki zokufunda koomatshini kwaye anikeze ukusebenza kwe-21,5, 16, kunye ne-12,8 petaflops, ngokulandelanayo.

I-Cluster ye-Christofari Neo esetyenziswe yi-Sberbank yawa ukusuka kwi-46 ukuya kwindawo ye-50. U-Christofari Neo uqhuba i-NVIDIA DGX OS 5 (Ubuntu Edition) kwaye unikezela nge-petaflops ye-11,9 yokusebenza. Iqela lingaphezulu kwe-98k cores esekwe kwi-AMD EPYC 7742 64C 2.25GHz CPU kwaye iza ne-NVIDIA A100 80GB GPU. Iqela lesibini le-Sberbank (Christofari) liye lahamba ukusuka kwi-80 ukuya kwindawo ye-87 kwindawo yokubeka kwiinyanga ezintandathu.

Inxalenye yeyona ndlela inomdla ngonikezelo lweLinux (kwizibiyeli – kwiinyanga ezi-6 ezidlulileyo):
I-47.8% (47.8%) ayinazo iinkcukacha zokusasazwa.
I-17,2% (18,2%) isebenzise i-CentOS
9,6% (8,8%) - RHEL
9% (8%) - CrayLinux
5,4% (5,2%) -Ubuntu
3,8% (3,8%) - ISUSE
0,8% (0,8%) - Alma Linux
0,8% (0,8%) - RockyLinux
0,2% (0,2%) - I-Linux yeNzululwazi.

Ubuncinci bomda wokusebenza ukungena kwiTop500 iinyanga ezi-6 ibiyi-1,73 petaflops (kwiinyanga ezintandathu ezidlulileyo, i-1,65 petaflops). Kwiminyaka emine eyadlulayo kuphela ngamaqela angama-272 abonisa ukusebenza kwiipetaflops, kwiminyaka emihlanu eyadlulayo 138, kwiminyaka emithandathu eyadlulayo 94). Kwi-Top 100, umda wokungena unyuke ukusuka kwi-5,39 ukuya kwi-9,22 petaflops.

Ukusebenza okupheleleyo kwazo zonke iinkqubo ezikuluhlu kunyuke ukusuka kwi-4,4 ukuya kwi-4,8 exaflops kwiinyanga ezi-6 (iminyaka emithathu edlulileyo yayiyi-1.650 exaflops kwaye kwiminyaka emihlanu edlulileyo yayiyi-749 petaflops). Inkqubo evala ireyithingi yangoku ibekwe kwi-458 kwinqaku lokugqibela.

Gqibela ukuba unomdla wokwazi okungakumbi ngayo, ungajonga iinkcukacha Kule khonkco ilandelayo.