Helu 55 ʻO 500 mau kamepiula hana kiʻekiʻe loa ma ka honua. Ua hoʻokiʻekiʻe ʻia ka helu o Iune e kahi alakaʻi hou - ka hui ʻo Iapana , kaulana no kona hoʻohana ʻana i nā mea hana ARM.
Pūʻulu Fugaku ma ka Institute of Physical and Chemical Research a hāʻawi i ka hana o 415.5 petaflops, ʻo ia ka 2.8 ʻoi aku ma mua o ke alakaʻi o ka helu mua, i hoʻokuʻu ʻia i kahi lua. Loaʻa i ka pūʻulu he 158976 SoC-based nodes , lako me ka 48-core CPU Armv8.2-A SVE (512 bit SIMD) me ke alapine o ka uaki o 2.2GHz. ʻO ka huina, ʻoi aku ka nui o ka hui ma mua o 7 miliona mau cores processor (ʻekolu manawa ʻoi aku ma mua o ke alakaʻi o ka helu mua), kokoke i 5 PB o RAM a me 150 PB o ka waiho ʻana ma muli o ka Lustre FS. Hoʻohana ʻia ʻo Red Hat Enterprise Linux e like me ka ʻōnaehana hana.
Ua hoʻopaʻa ʻia nā wahi mai ka lua a hiki i ka ʻelima e nā alakaʻi ma mua, ka mea i noho i ka mua a i ka ʻehā ma ka pae o Nowemapa:
- Kahi lua - cluster na IBM ma Oak Ridge National Laboratory (USA). Holo ka pūʻulu i ka Red Hat Enterprise Linux a loaʻa iā 2.4 miliona mau cores processor (me ka hoʻohana ʻana i 22-core IBM Power9 22C 3.07GHz CPUs a me NVIDIA Tesla V100 accelerators), e hāʻawi ana i ka hana o 148 petaflops.
- Wahi ʻekolu - ʻAmelika Hui Pū ʻIa , i hoʻokomoʻia ma ka Livermore National Laboratory e IBM ma ke kumu o kahi papahana e like me Summit a hōʻike i ka hana ma 94 petaflops (e pili ana i 1.5 miliona cores).
- Wahi ʻehā - Hui Pākē , e hana ana ma ka National Supercomputing Center o Kina, me ka ʻoi aku o 10 miliona mau cores computing a hōʻike i ka hana o 93 petaflops. Me nā hōʻailona hana like, hoʻopau ka hui Sierra i ka hapalua o ka ikehu e like me ka Sunway TaihuLight.
- Wahi ʻelima - Hui Pākē , kahi kokoke i 5 miliona cores a hōʻike i ka hana o 61 petaflops.
Ua loaʻa i nā hui hou nā wahi ʻeono a me ka ʻehiku (Italia, Dell EMC, 35 petaflops, 669 tausani cores) a (USA, 27 petaflops, 277 tausani cores), nāna i pani i ka hui ʻAmelika (Dell EMC, 23 petaflops, 448 tausani cores) i ka wahi ʻewalu. Ua loaʻa i ka pūʻulu Italia hou ka wahi ʻeiwa (IBM, 21.6 petaflops, 347 tausani cores), a ʻo ka ʻumi he hui Swiss. (Cray/HPE, 21.2 petaflops, 387 tausani cores), i lawe i ka 6th wahi ma ka pae mua.
ʻO nā hiʻohiʻona hoihoi loa:
- Huipuʻu hale (6.6 petaflops, Ubuntu 18.04.01/2/8168, i kūkulu ʻia e Sberbank ma ka NVIDIA DGX-24 platform, hoʻohana i ka Xeon Platinum 2.7 99600C 6GHz CPU a loaʻa iā 29 cores computing) neʻe mai ka 36th a 2th wahi i ka pae ma 107 mau mahina. ʻO kekahi hui kūloko, ʻo Lomonosov 131, neʻe mai XNUMX a XNUMX wahi. Hui i loko , ka mea i koho ʻia i ka 465th, ua hoʻokuke ʻia i waho o ka pae. No laila, ua emi mai ka helu o nā puʻupuʻu kūloko i ka pae ma luna o ʻeono mahina mai 3 a 2 (ma 2017 aia 5 i ka pae , a ma 2012 - 12);
- Ka mahele ʻana ma ka helu o nā supercomputers ma nā ʻāina like ʻole:
- Kina: 226 (228 ʻeono mahina aku nei). ʻO ka huina, hoʻopuka nā hui Kina i ka 45.2 o ka huina huahana (ʻeono mahina i hala - 31.9%);
- USA: 114 (117). Kuhi ʻia ka nui o ka huahana ma 22.8% (ʻeono mahina i hala - 37.8%);
- Iapana: 29 (29);
- Palani: 19 (18);
- Kelemānia: 16 (16);
- Netherlands: 15 (15);
- ʻIlelani: 14 (14);
- Kanaka 12(9);
- UK: 10(11);
- Italia: 7 (5);
- Palakila: 4 (3);
- Singapore 4 (4);
- South Korea, Saudi Arabia, Norewai: 3;
- Rusia, India, Australia, UAE, Switzerland, Suedena, Finland, Taiwan: 2.
- I ka helu o nā ʻōnaehana hana i hoʻohana ʻia i nā supercomputers, ʻo Linux wale nō i koe no ʻekolu makahiki;
- Hāʻawi ʻia e nā māhele Linux (ma nā brackets - 6 mau mahina aku nei):
- 54.4% (49.6%) ʻaʻole kikoʻī i ka hāʻawi ʻana,
- 24.6% (26.4%) hoʻohana i CentOS,
- 6.8% (6.8%) - Cray Linux,
- 6% (4.8%) - RHEL,
- 2.6% (3%) - SUSE,
- 2.2% (2%) - Ubuntu;
- 0.2% (0.4%) - Linux ʻepekema
- ʻO ka paepae hana haʻahaʻa loa no ke komo ʻana i ka Top500 i loko o 6 mau mahina i hoʻonui ʻia mai 1142 a 1230 teraflops (i ka makahiki i hala, ʻo 272 mau puʻupuʻu wale nō i hōʻike i ka hana ʻoi aku ma mua o ka petaflop, ʻelua makahiki i hala - 138, ʻekolu makahiki i hala - 94). No Top100, ua hoʻonui ka paepae komo mai 2570 a i 2801 teraflops;
- Ua hoʻonui ka nui o ka hana o nā ʻōnaehana āpau i ka makahiki mai ka 1.65 a i ka 2.23 exaflops (ʻekolu makahiki i hala aku nei he 749 petaflops). ʻO ka ʻōnaehana e pani ana i ka pae o kēia manawa aia ma kahi 449th ma ka helu hope loa, a ʻo 348th i ka makahiki ma mua;
- ʻO ka mahele nui o ka helu o nā supercomputers ma nā wahi like ʻole o ka honua penei:
Aia 272 supercomputers ma Asia (274 - ʻeono mahina i hala),
126 ma ʻAmelika ʻĀkau (129) a me 96 ma ʻEulopa (94), 4 ma ʻAmelika Hema a me 2 ma Oceania (3); - Ma keʻano he papa hana, aia nā Intel CPU i ke alakaʻi - 93.8% (ʻeono mahina i hala aku nei he 94%), ma ka lua o ka IBM Power - 2.6% (mai 2.8%), ma ke kolu o ka AMD - 2.2% (0.6%). ), ma ka hā o ka ARM (Marvell ThunderX2 a me Fujitsu A64FX) - 0.8%, ma ka lima o SPARC64 - 0.2% (0.6%). No ka manawa mua i ka pae, ua hōʻike ʻia kahi pūʻulu e pili ana i nā kaʻina hana ARM, i lawe koke ʻia i kahi mua.
- 37.4% (35.6 mau mahina i hala aku nei 20%) o nā kaʻina hana i hoʻohana ʻia he 12.2 cores, 13.8% (16%) - 10.6 cores, 24% - XNUMX cores,
10.4% (11%) - 18 cores, 9.8% (11.2%) - 12 cores, 7% (7.8%) - 14 cores; - 145 mai loko o 500 ʻōnaehana (ʻeono mahina i hala aku nei - 144) hoʻohana hou i nā accelerators a i ʻole coprocessors, aʻo 135 mau ʻōnaehana hoʻohana i nā chips NVIDIA (ʻeono mahina i hala aku nei he 135), 6 - Intel Xeon Phi (aia he 5), 1 - PEZY (1) , 1 hoʻohana i nā hoʻonā hybrid (1 mua), 1 hoʻohana iā Matrix-2000 (1), 1 GPU AMD Vega (1);
- Ma waena o nā mea hana cluster, ua lawe ʻo Lenovo i kahi mua - 36% (34.8 mau mahina i hala aku nei XNUMX%), ma kahi lua.
ʻO Sugon 13.6% (14.2%), ma ke kolu o Inspur - 12.8% (13.2%), ʻehā wahi i noho ʻia e Hewlett-Packard - 7.6% (7%), ukali ʻia e 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%),
Penguin Computing - 1.2% (2.2%). ʻEkolu mau makahiki i hala aku nei, ʻo ka hāʻawi ʻana i waena o nā mea hana penei: Hewlett-Packard 28.6% (22.4%), ma ka lua o ka Lenovo 17% (18.4%), ma ke kolu o Cray 11.4% (11.2%), ma ka ʻehā o ka wahi. Sugon 9.2% (9.4%), ma ka ʻelima o ka IBM 5.4% (6.6%). - Hoʻohana ʻia ʻo Ethernet e hoʻopili i nā nodes ma 52.6% o nā pūʻulu (52 mahina i hala aku nei 30.4%), InfiniBand - 28% (9.8%), Omnipath - 10% (52.6%). Inā mākou e noʻonoʻo i ka nui o ka hana, a laila uhi ʻia nā ʻōnaehana pili Ethernet i ka 29% (500%) o ka nui o ka hana Top30.4, InfiniBand - 40% (9.8%), Omnipath - XNUMX%.
I ka manawa like, loaʻa kahi hoʻokuʻu hou o kahi helu ʻokoʻa o nā ʻōnaehana cluster , i ka nānā 'ana i ka hana o nā papahana supercomputer e pili ana me ka ho'ohālike i nā ka'ina hana kino a me nā hana no ka ho'oponopono 'ana i ka nui o ka 'ikepili ma'amau no ia mau pūnaewele. Ka helu kaawale hou aku a hui pū me Top500, e like me ka ikehu i kēia manawa i ka helu Top500 nui (ʻo ka ratio o LINPACK FLOPS i ka hoʻohana ʻana i ka mana i nā watts e mālama ʻia).
Source: opennet.ru