I-AI engajwayelekile igqashuka ngomjikelezo wembewu eyi-mega nendlela entsha yama-chips e-AI

Ukufika kuka I-AI engajwayelekile Kunyakazise isimo se-artificial intelligence hardware ngomjikelezo woxhaso osekuxoxwa ngawo kuyo yonke indilinga yezimboni. cishe izinyanga ezimbalwa ubudala, Inkampani Ikwazile ukubamba intshisekelo yezimali ezinamandla kakhulu emhlabeni wezobuchwepheshe.ukubheja ngombono othi, ephepheni, uthembise ukucabanga kabusha ukuthi izinsiza zekhompiyutha ze-AI zenziwa futhi zisetshenziswe kanjani.

Kunokuba igxile ekukhuleni kwamamodeli amakhudlwana nayingozi, inkampani ifuna ukuhlasela inkinga emsukeni wayo: ukusebenza kahle kwamandla kanye nezakhiwo ezibonakalayo zama-chipsIsiphakamiso sakhe sigqugquzelwe ngokusobala yibhayoloji nokusebenza kobuchopho, ne Umgomo uwukusondela kusistimu ekwazi ukunikeza amandla amakhulu ekhompuyutha kuyilapho kudla ingxenye yamandla adingekayo namuhla. izikhungo ezinkulu zedatha.

Umjikelezo wembewu wehadiwe we-AI omkhulu kakhulu wonyaka

I-AI engajwayelekile ivale umjikelezo wembewu oyizigidi ezingama- $ 475Isibalo, ngisho nasemakethe ejwayele amanani amakhulu, sigqama ngobukhulu baso ekuqaleni. Okwenziwayo kwazisa inkampani endaweni AmaRandi ayizigidigidi eziyi-4.500, okuyenza ibe ngelinye lamacala agqama kakhulu oxhaso lwembewu ku-AI ​​hardware ecosystem.

Lo mzuliswano ubuholwa yi-venture capital funds U-Andreessen Horowitz (a16z) y Ozakwethu beLightspeed VentureAbadlali ababili ababalulekile uma kuziwa ekutshalweni kwezimali kwesikhathi eside kubuchwepheshe obujulile. Bahlanganiswe nabanye abatshalizimali bezinga eliphezulu njenge I-Lux Capital, I-DCVC, Ama-Databricks futhi ngisho nomsunguli we-Amazon, UJeff BezosLokhu kuqinisa umuzwa wokuthi iphrojekthi ithathwa njengesinyathelo samasu wesikhathi eside.

Ngaphezu kwezimali zangaphandle, omunye wabasunguli abambisene naye usenqume ukufaka isandla ephaketheni lakhe. AmaRandi ayizigidigidi eziyi-10...ngemibandela efanayo nabanye abatshalizimali abakhulu. Lesi sinyathelo, esingaphezu kwenani, sithumela isignali ecacile yokuzibophezela kanye nokuzethemba kwangaphakathi kuthisisi yenkampani yezobuchwepheshe nebhizinisi.

Ngokwezingxoxo ezahlukahlukene, lesi sigaba sokuqala sezigidi ezingama-475 sizoba yisiqalo sohlelo lokuqongelela imali olungafinyelela AmaRandi ayizigidigidi eziyi-1.000 kulesi sigaba esifanayo. Ubukhulu benhloso bugqamisa uhlobo lwephrojekthi ababhekene nayo: izingxenyekazi zekhompuyutha eziyinkimbinkimbi, imijikelezo yentuthuko emide, kanye nokutshalwa kwezimali kokuqala okusindayo ku-R&D.

Uma kuqhathaniswa nokunye okwenziwe kwakamuva, inani lehle kancane 5.000 billion okwaxoxwa ngakho kumahemuhemu okuqala, kodwa isabeka i-AI Engajwayelekile ohlwini lwabaqalayo, okuthi, kungenamali engenayo noma umkhiqizo wokuhweba, asevele adlala emazingeni emali ebekelwe izinkampani ezivuthiwe kakhulu.

Umbono ka-Naveen Rao kanye neqembu elijwayele ubungozi bezobuchwepheshe

Ukuhola iphrojekthi kuyinto Naveen RaoU-Rao, umuntu owaziwayo emhlabeni we-AI kokubili ohlangothini lwakhe lwezamabhizinisi kanye nezikhundla zakhe ezinkampanini ezinkulu zobuchwepheshe. ibhekele izinkundla zobuhlakani bokwenziwa kwa-Intel ngemuva kokuthengwa kokuqaliswa kwayo kokuqala, i-Nervana Systems, egxile kumaphrosesa okufunda ngomshini.

Kamuva, umsunguli wathatha elinye igxathu ngokusungula ngokubambisana I-MosaicML, inkundla yokuqeqeshwa eyimodeli eyazuza ukuthambekela kudatha kanye ne-AI ecosystem futhi yagcina ngokutholwa yiyo I-Databricks cishe i-$ 1.300 billionLeli rekhodi, elinokuphuma okubili okubalulekile esikhathini esingaphansi kweminyaka eyishumi, libe nesisindo esikhulu ekwenzeni ukuzethemba phakathi kwezimali manje ezisekela iphrojekthi yayo entsha.

Eceleni kukaRao, inkampani ifake amaphrofayili asezingeni eliphezulu asuka ezimpambanweni zomgwaqo i-hardware, isofthiwe, kanye nocwaningo lwezemfundo, njengoba UMichael Carbin, Sara Achour y Mina LeeLeli yiqembu elijwayele ukubhekana nobungozi obuphezulu bezobuchwepheshe, amaphrojekthi omjikelezo omude, nezinkinga ezingaxazululwa ngokuphindaphinda ama-software asheshayo, kodwa ngama-prototypes ayinkimbinkimbi kanye nokuhlanganiswa okuseduze kakhulu phakathi kwezakhiwo ezibonakalayo nama-algorithms.

URao ngokwakhe uye wachaza ukuthi uhlelo lokusebenza lwe-AI olungajwayelekile lubandakanya hlola ama-prototypes amaningi eminyakeni emininganaBahlola ukuthi iyiphi i-paradigm ekala kahle ngokusebenza kahle kanye nezindleko. Ngamanye amazwi, ababheke ukwethula umkhiqizo ngokushesha, kodwa kunalokho ukwakha isisekelo sobuchwepheshe esingenza umehluko kukhompyutha ye-AI eminyakeni eyishumi ezayo.

Lokhu kubhejwa okuthiwa "ubunjiniyela bomjikelezo omude" Lokhu kuphambene nendlela evamile yokuqalisa kwesofthiwe eminingi, egxile ekuqinisekiseni namakhasimende ngokushesha okukhulu nokulungisa umkhiqizo ngokuphindaphinda ngokushesha. Lapha, indlela ifana kakhulu neyezinkampani ezinkulu ze-semiconductor noma amaphrojekthi wengqalasizinda ebalulekile, lapho ukubuyisela ekutshalweni kwezimali kuza kamuva kodwa, uma konke kuhamba kahle, kungachaza kabusha wonke umkhakha.

Uhlobo olusha lomshini wobuhlakani bokwenziwa

Umnyombo wesiphakamiso se-AI esingajwayelekile ukwakha ikhompuyutha eyonga amandla kakhulu ngemithwalo yemisebenzi yobuhlakani bokwenziwa. U-Rao ufingqe isifiso ngegama elidonse ukunaka kulo mkhakha: ukuklama uhlelo "isebenza kahle njengebhayoloji", ukuthatha njengereferensi amandla obuchopho bomuntu okwenza izibalo eziyinkimbinkimbi ngokusebenzisa amandla amancane.

Ngenkathi iningi lemboni liqhubeka nokuphakamisa ukukala kwamamodeli-amapharamitha amaningi, idatha eyengeziwe, ama-GPU amaningi-, inkampani iqala ngesisekelo sokuthi Lelisu linomkhawulo ocacile mayelana nezindleko namandla atholakalayo.Izikhungo ezinkulu zedatha sezivele zibhekene nemikhawulo yamandla, izindleko ezikhuphukayo, nezinkinga zokusimama, into ekhathaza kakhulu eYurophu naseSpain ngenxa yezinhloso zesimo sezulu nezokulawula.

Ukuze uphule lokhu okuguquguqukayo, ukuqalisa kuphakamisa i-paradigm shift in computing architectureEsikhundleni sokuqhubeka nokucwenga izakhiwo ezijwayelekile zedijithali, hlola imiklamo esebenzisa amandla izakhiwo ezibonakalayo ze-silicon ngokwayo kanye nezimiso ezigqugquzelwe ukusebenza kobuchopho, njengokuguquguquka okungaqondile kwama-neurons.

Embhalweni oshicilelwe kuwebhusayithi yayo, inkampani ichaza umgomo wayo njengokwakha i- "i-substrate entsha yobuhlakani"Umqondo uwukuthi, ngokuthola isakhiwo esifanele esixhumanisa ikhompuyutha yokwenziwa nokuziphatha kwezinhlelo zebhayoloji, kungenzeka uvule izinzuzo ezisebenza kahle kakhulu ezingaphezu kwalokho okufinyelelwayo ngokumane uthuthukise izakhiwo zedijithali zakudala.

Abatshalizimali bakwaLightspeed ababambe iqhaza emzuliswaneni bayavumelana nalokho kuxilongwa, bekhomba isidingo ukucinga "i-isomorphism efanelekile yobuhlakani" Uma umgomo uwukuthola ukuncipha okukhulu kokusetshenziswa kwamandla kagesi kwe-AI, lo mzila wokucabanga uhambisana nemizamo yocwaningo kukhompyutha ye-neuromorphic kanye nezinhlelo ezithuthukisiwe ze-analog, okuze kube manje, ziye zahlala kakhulu ngaphakathi kwezemfundo noma amaphrojekthi okuhlola abakhiqizi abakhulu.

I-Architecture: Ukusuka ku-Analog Chips kuya kuma-Pulsating Neurons

Esinye sezici ezigqama kakhulu ze-AI Engajwayelekile indlela yayo ehlanganisiwe yoku i-analog, i-mixed, ne-neuromorphic architecturesNgokungafani nama-chips amanje edijithali, amelela ulwazi olusebenzisa ama-zero ahlukene kanye nalawo, imiklamo ye-analog ivumela ukusebenza ngamavelu aqhubekayo nokusebenzisa ngokunenzuzo izenzakalo ezingokomzimba okuthi, lapho zilawulwa kahle, zingasebenza kahle kakhulu emisebenzini ethile. Le ndlela ikhomba ukuthuthuka kwe umklamo we-chip othuthukisiwe nezinqubo ezifuna ukuthuthukisa ukusebenza kahle kusukela kusisekelo somzimba.

Inkampani iyahlola ama-chips akwazi ukugcina ngokwamandla ukusatshalaliswa kwamathubaesikhundleni sokuwalinganisa ngezinombolo njengoba kwenziwa kumaphrosesa avamile. Lokhu kuvula umnyango wezethulo zemvelo ezengeziwe zamamodeli angenzeka futhi, okungenzeka, ukuze ukunciphisa ukusetshenziswa kwamandla kufika izikhathi eziyinkulungwane uma kuqhathaniswa nezinhlelo zedijithali eziphethe izikhungo zedatha namuhla.

Ukufeza lokhu, ithimba lisebenzisa imiqondo evela ama-oscillator, ama-thermodynamics nama-spiking neuronsLolu hlobo lwemodeli lugqugquzelwa yindlela ama-neuron angempela enziwa ngayo amandla ahlukene ngokuhamba kwesikhathi. Lezi zakhiwo, ezijwayelekile zenkambu ye-neuromorphic, zingenza kungasebenzi izingxenye ezinkulu ze-chip uma ingasetshenziswa, kwehlise kakhulu ukulahleka kwamandla uma kuqhathaniswa namasekhethi agcina umsebenzi oqhubekayo.

Le ndlela ngandlela-thile ikhumbuza imizamo yangaphambilini yezinkampani ezifana ne-Intel namaphrosesa azo we-neuromorphic, aqeda iwashi elivamile elivamile futhi avumele i-chip ukuthi isebenze ngokuvumelanayo, isebenze izingxenye ezidingekayo kuphela kuye ngomthwalo womsebenzi. Nokho, I-AI engajwayelekile ifuna ukuqhubekela phambilihhayi kuphela ngokulingisa ukuziphatha kwe-neuronal, kodwa ngokuhlanganisa eduze ukwakheka komzimba we-silicon namamodeli we-AI aklanyelwe leyo ndawo.

Le nhlanganisela ye Izingxenyekazi zekhompuyutha eziyisipesheli namamodeli aklanywe ngokuhlanganyela Ikhomba esikhathini esizayo lapho umngcele phakathi kwe-chip ne-algorithm ifiphalisa, nalapho ukusebenza kungasancikile kangako ekutheni mangaki ama-GPU angastakwa, kodwa ekusetshenzisweni kahle kwezakhiwo ezijulile zezinto ezibonakalayo namasekhethi.

I-SoC eyenzelwe ngokwezifiso igagasi elilandelayo le-AI

Ngale kokubuka konke, imininingwane yobuchwepheshe iyavela mayelana nohlobo lwe-chip Unconventional AI okuhloswe ngayo ukuletha ekukhiqizeni. Ukuthunyelwa kwemisebenzi eyahlukene eshicilelwe yinkampani kukhomba... i-accelerator ye-AI esekelwe kumklamo wesistimu-on-a-chip (SoC).Okusho ukuthi, ingxenye eyodwa ehlanganisa amamojula amaningana ekhompiyutha akhethekile.

Ngokwalezi zincazelo, i-SoC izofaka i-central processor (CPU) unesibopho semisebenzi yokuqala efana nokuhlela kanye nokulungiselela idatha yezinzwa ngaphambi kokuba idluliselwe kumayunithi aqondile we-AI. Ngokusekelwe kulesi sisekelo esijwayelekile, amabhulokhi athuthukisiwe azokwengezwa ukuze enze imisebenzi ye-algebra yomugqaokuyinhliziyo yezibalo cishe yazo zonke izinhlobo zokufunda ezijulile, kusukela kumamodeli wolimi amakhulu kuya ezinhlelweni zokubona kwekhompyutha.

Umklamo futhi ucabangela ukusetshenziswa impahla yengqondo yomuntu wesithathu Kwamanye amamojula, lokhu kuwumkhuba ovamile embonini ye-semiconductor, lapho kusebenza kahle kakhulu ukunikeza amabhulokhi amabhulokhi aqinisekisiwe kunokuwathuthukisa kusukela ekuqaleni. Ukusuka lapho, inani elingeziwe le-AI elingajwayelekile lizogxilwa ezingxenyeni eziqanjwa kakhulu ze-SoC.

Lezi zakhi ezihlukanisayo zihlanganisa izifunda zesignali ezixubileLawa masekethi, akwazi ukucubungula kokubili ulwazi lwe-analog nolwazi lwedijithali, awusizo kakhulu ekulawuleni idatha evela kuzinzwa noma ekusebenziseni ngokuqondile imisebenzi ephefumulelwe yi-physics. Lolu hlobo lokujikeleza luwukhiye ukuze i-chip isebenzise amandla angaqondile kanye nezethulo ezingase zibe khona ezilandelwa inkampani.

Elinye iphuzu elibalulekile intshisekelo yenkampani kulo izinkumbulo ezivelayo ezingaguquki, njenge-RRAMLobu buchwepheshe bugcina ulwazi ngisho nalapho amandla elahlekile. Banganikeza izinzuzo zokusebenza ngaphezu kwememori yefuleshi evamile ezimweni ezithile, nakuba basabhekene nezinselele zobuchwepheshe ezikhawulele ukusabalalisa kwabo okusabalele kuzikhungo zedatha. Ukuvela kwemakethe yenkumbulo nezinqumo zabakhiqizi abanjengo I-Micron ehlobene nemigqa yomkhiqizo Bagqamisa lezi zinselelo namathuba.

I-Co-design yamamodeli wehadiwe ne-AI

I-AI engajwayelekile ayifuni ukuhlala kuphela kusendlalelo esibonakalayo sephrosesa. Isu futhi libandakanya ukuthuthukisa amamodeli e-AI ajwayelane nama-chips awo., ukusebenzisa imajini yokuthuthukisa enikezwa ngokudala isofthiwe nehadiwe ndawonye kusukela ekuqaleni.

Le ndlela ye co-design Ivumela ukulawula okuphezulu kokuthi idatha imelelwa kanjani, imiphi imisebenzi eyenziwa, nokuthi umsebenzi usatshalaliswa kanjani ngaphakathi kwe-chip. Esikhundleni sokushintsha amamodeli akhona aklanyelwe ama-GPU enhloso ejwayelekile, inkampani ingaklama ama-algorithms asebenzisa izici ezihlukile zamasekhethi ayo e-analog, ama-pulsating neurons, noma amamojula enkumbulo angajwayelekile.

Inkampani inethemba lokuthi lokhu kuhlanganiswa kuzokuvumela ukuthi kufezeke ukusebenza kahle ngokulandelana kwezikhathi eziyi-1.000 uma kuqhathaniswa ne-silicon yamanje ngaphansi kwemithwalo ethile yomsebenzi. Yize lezi zibalo zizodinga ukuqinisekiswa lapho kuvela ama-prototypes okuqala azimele namabhentshimakhi, zinikeza umbono wesilinganiso sesifiso iqembu elihlose ukusifinyelela.

Lolu hlobo lwendlela lufaneleka kakhulu EYurophu naseSpainlapho impikiswano ngobukhosi bezobuchwepheshe kanye nokuncika kubahlinzeki bezingxenyekazi zekhompiyutha bangaphandle iba namandla. Ukuba nezakhiwo ezintsha, ezisebenza kahle kakhulu ze-AI kuvula umnyango wezikhungo zedatha ezizinzile nezingabizi kakhulu.Lokhu kuhambisana namandla esifunda kanye nezinto ezibalulekile zokulawula. Ukubambisana phakathi kwabahlinzeki abakhulu bamafu nabakhiqizi bezingxenyekazi zekhompuyutha, njengalezo ezisanda kushintsha isimo somkhakha, ziyisibonelo somongo lapho lezi zixazululo zingangena khona.ukusebenzisana phakathi kwamafu nabakhiqizi).

Uma imodeli ye-AI engajwayelekile ekugcineni ikhombisa ukuncintisana, Ngeke kumangaze ukubona izinkampani zamafu zaseYurophu, amalebhu ocwaningo, nezinkampani ezinkulu zihlanganisa lezi zinhlobo zezixazululo. engqalasizinda yayo, efuna ukunciphisa izindleko zamandla kanye ne-carbon footprint ngaphandle kokudela amakhono e-AI athuthukile.

Umongo wemakethe: Imijikelezo eyi-Mega kanye nomjaho wengqalasizinda ye-AI

Indaba ye-AI Engajwayelekile iyingxenye yethrendi ebanzi: ukuvela kweziqalo ze-AI ezikhuphula amakhulu ezigidi zamaRandi ngezigaba zakuqala kakhulu, ngokulinganisa eminyakeni embalwa edlule ebegcinelwe izinkampani ezisohlwini noma izinkampani ezinemali engenayo ehlanganiswe kakhulu.

Eminyakeni yamuva, amagama afana I-OpenAI, I-Anthropic noma izinhlelo ezikhuthazwa izibalo ezifana nalezi U-Ilya Sutskever o Mira Murati Bebezibandakanya emizuliswaneni eyingqophamlando yemali yebhizinisi. Ngo-2025, inqwaba yeziqalo ze-AI zeqa ingqophamlando Imali engu-$100 millionukuhlanganisa umthamo wotshalomali ongakaze ubonwe kule ngxenye.

Kuleli gagasi, impi yengqalasizinda Ama-chips, amafu akhethekile, ama-accelerator, kanye nezinhlelo zokuqeqesha sekungenye yezindawo okuqhunyiswa kakhulu kuzo. ukuncika kweprosesa Ukushoda kwedlanzana labakhiqizi, futhi ikakhulukazi ama-GPU asezingeni eliphezulu, kuye kwashukumisela abatshalizimali nosomabhizinisi ukuba bafune ezinye izindlela ezidambisa izingqinamba zokuhlinzekwa kanye nentengo.

I-AI engajwayelekile ingena kulo mjaho ngokuhlongoza indlela ehlukile kunokuncintisana nje okukhulayo nabakhiqizi abakhulu be-GPUEsikhundleni sokulwela nje ukusebenza okwengeziwe, gxila ekuzuzeni imiyalo yokuthuthukiswa kobukhulu ekusebenzeni kahle kwamandla, into ebalulekile esikhathini esimaphakathi ukuze amasistimu e-AI aqhubeke nokukhula ngaphandle kokugijimela emikhawulweni yomzimba nezomnotho.

Ohlelweni lwe-ecosystem yaseYurophu, lapho izindleko zamandla kanye nezidingo zokulawula ekukhishweni kwesisi ziqine kakhulu, impumelelo yeziphakamiso zalolu hlobo ingaba yisinqumo esibalulekile. Ihadiwe ye-AI esebenza kahle kakhulu Lokhu kuzofanelana namasu okushintsha okuluhlaza, kuyilapho kuvumela izinkampani nabaphathi ukuthi basebenzise izinhlelo zokusebenza ze-AI ezithuthukisiwe ngaphandle kokwandisa ukusetshenziswa kwazo.

Iphrojekthi ye I-AI engajwayelekile Ihlanganisa izitayela eziningi ezinkulu zamanje: ama-mega-round ezigabeni zembewu, i-hardware eklanyelwe i-AI kusukela phansi, ugqozi oluqondile oluvela kubhayoloji, kanye nokuthatheka ngokusebenza kahle kwamandla okusabela eqinisweni elibonakala liqhubeka libonakala. Uma inkampani ikwazi ukwenza izithembiso zayo ku-silicon, ingaba ngomunye wabadlali ababalulekile abachaza ukuthi amamodeli obuhlakani bokwenziwa aqeqeshwa futhi asebenze kanjani eminyakeni eyishumi ezayo, e-United States naseYurophu, futhi, ngokwengeza, ezimakethe ezifana neSpain.

Isihloko esihlobene:

I-Nvidia iqinisa umfelandawonye wayo wesu ne-Synopsys enhliziyweni yomklamo we-chip

U-Alberto Navarro

Ngingumshisekeli wezobuchwepheshe oguqule izintshisekelo zakhe "ze-geek" zaba umsebenzi. Ngichithe iminyaka engaphezu kwengu-10 yempilo yami ngisebenzisa ubuchwepheshe obusezingeni eliphezulu kanye nokukitaza ngazo zonke izinhlobo zezinhlelo ngenxa yelukuluku lokufuna ukwazi. Manje sengiqeqeshelwe ubuchwepheshe be-computer nemidlalo yama-video. Lokhu kungenxa yokuthi sekuphele iminyaka engaphezu kwengu-5 ngisebenza ngokubhalela amawebhusayithi ahlukahlukene ezobuchwepheshe nemidlalo yevidiyo, ngenza izindatshana ezifuna ukukunikeza imininingwane oyidingayo ngolimi oluqondakala yiwo wonke umuntu.

Uma unemibuzo, ulwazi lwami lusukela kuyo yonke into ehlobene nesistimu yokusebenza ye-Windows kanye ne-Android yomakhalekhukhwini. Futhi ukuzibophezela kwami ​​​​kuwe, ngihlala ngizimisele ukuchitha amaminithi ambalwa futhi ngikusize uxazulule noma yimiphi imibuzo ongase ube nayo kulo mhlaba we-inthanethi.