Ibhlog

Yintoni i-NPU?

2024-08-13 16:29:49

Kwimeko yanamhlanje ekhula ngokukhawuleza Ihlabathi eliqhutywa yi-AI, imfuno yezibalo ezikhawulezayo, ezisebenza ngakumbi zinike iiprosesa ezikhethekileyo ezifana ne IYunithi yokuLungiselela iNeural (NPU). Ngokungafaniyo nemveli CPUs okanye GPUs, i-NPU yenzelwe ukukhawulezisa ingcamango yokufunda koomatshini, ngakumbi kwi iinethiwekhi ezinzulu ze-neural isetyenziswe kwi umbono wekhompyuter, ukusetyenzwa kolwimi lwendalo, kwaye ukuqondwa kwelizwi.

An NPU luhlobo lwe I-AI ye-accelerator ehambisa ukusebenza okuphezulu, ukusetyenzwa kwe-latency ephantsi kwimisebenzi efunekayo ukubala ngokunxuseneyo kwaye imisebenzi tensor. Ezi processors ngoku zifakwe kuluhlu olubanzi lwamaqonga-ukusuka ii-smartphones kwaye Izixhobo ze-IoT ukuya iinkqubo zemoto kwaye iziseko zekhompyutha ezisekupheleni-inika amandla kwisixhobo AI ngokusebenzisa amandla amancinci.

Kutheni iiNPU zibalulekile:

Uphuculo lokusebenza ngexesha lokwenyani kumthwalo wemisebenzi ye-AI
Ukusetyenziswa kwamandla aphantsi xa kuthelekiswa nezisombululo ezisekelwe kwi-GPU
Ukuncitshiswa kokuxhomekeka kwingqikelelo yelifu, ukuphucula ubumfihlo bedatha kunye ne-latency
Ukubaleka kwizixhobo ezixineneyo kunye neenkqubo ezizinzisiweyo

Ukunyuka kwe edge AI kunye nokwanda kwe Usetyenziso olunamandla e-AI yenza ukuba kubaluleke ukuqonda ukuba yintoni i-NPU kwaye yohluke njani kwezinye iiyunithi zokusetyenzwa. Kweli nqaku, siphonononga i ulwakhiwo, sebenzisa iimeko, kunye nexabiso leqhinga le-NPU kumxholo obanzi wekhompyutha ekrelekrele.

Yintoni iNPU? Isishwankathelo soBugcisa

A IYunithi yokuLungiselela iNeural (NPU) yimicroprocessor ezinikeleyo eyenzelwe ngokukodwa ukuphatha ubukrelekrele bokwenziwa (AI) imithwalo yemisebenzi, ngakumbi leyo ibandakanya ukufunda okunzulu kwaye inkcazo yenethiwekhi ye-neural. Ngokungafaniyo nenjongo jikelele CPUs okanye igrafiki-oriented GPUs, ii-NPU zilungiselelwe imisebenzi yezibalo ezinamandla iimodeli zokufunda koomatshini, njenge Uphindaphindo lwematrix, convolutions, kwaye ukusetyenzwa kwe-tensor.

Iimpawu eziphambili zeNPU:

Injongo-eyakhelwe ukuthelekelela i-AI, hayi uqeqesho
Yenza imisebenzi ehambelanayo ngokufanelekileyo
Ilungiselelwe i-low-latency, i-high-throughput imisebenzi emininzi
Isebenza ngokucokisekileyo ukusetyenziswa kwamandla aphantsi ngaphezu kweGPUs

Ngelixa CPUs zifanelekile kwilogic elandelelanayo kunye GPUs bagqwese kwimisebenzi yokubonelela kunye nemidlalo, IiNPU zilungiselelwe ukukhawulezisa imisebenzi efana nale:

Ukufunyanwa kwento
Ukuqaphela ubuso
Ukuguqulwa kwelizwi ukuya kumbhalo
Uhlalutyo lwekhamera ehlakaniphile

Uhlobo lweProsesa	Indima yokuqala	Amandla	Ukusebenza kakuhle kwamandla	Ukufaneleka kwe-AI
ICPU	Imisebenzi ngokubanzi	Zininzi izinto	Phantsi	Ilinganiselwe
GPU	Imizobo, AI	Ukusebenza okuphezulu	Phakathi	Omelele kuqeqesho
NPU	Inkcazo ye-AI	I-AI-iphuculwe	Phezulu	Eyona nto ingcono kumda we-AI

Njengoko i-AI ifakwe kwizixhobo ezininzi-ukusuka ii-smartphones ukuya IoT yoshishino-i NPU igqame njengeyona ibalaseleyo esebenzayo kwaye iyakaleka iprosesa ukwenza iimpawu ezikrelekrele ngqo kwi edge, ngaphandle kokuxhomekeka ekufikeleleni kwifu rhoqo.

Isebenza Njani iNPU?

A IYunithi yokuLungiselela iNeural (NPU) yenzelwe ukuphumeza Imisebenzi ye-AI ngesantya esikhethekileyo kunye nokusebenza kakuhle. Ngokungafaniyo neeprosesa zemveli, iiNPU ziyasebenza iziseko zedatha esekwe kwi-tensor, ukwenza izixa ezikhulu ze imisebenzi ye-matrix ngaxeshanye—ibalulekile ekusebenzeni kwale mihla iimodeli zokufunda nzulu njenge CNNs (Convolutional Neural Networks) kwaye I-RNNs (Uthungelwano lweNeural oluQhelekileyo).

Imigaqo yokusebenza enguNdoqo:

Ukufana: IiNPU ziyaqhuba imisebenzi emininzi ngaxeshanye, inciphisa kakhulu ixesha lokucubungula imisebenzi efana ukuhlelwa komfanekiso kwaye ukuqondwa kwentetho.
Ubalo lwe-tensor olukhawulezileyo kwi-Hardware: Yakhelwe ngaphakathi tensor cores phatha ngokufanelekileyo isikali esikhulu Uphindaphindo lwematrix.
Izibalo ezichanekileyo ezisezantsi: IiNPU ezininzi zisebenzisa INT8 okanye FP16 iifomati ezichanekileyo zokubetha ulungelelwaniso phakathi ukuchaneka kwaye ukusebenza kakuhle kwamandla.
Uyilo lwemibhobho: Imisebenzi enje kulanda idatha, ukulungiswa kobunzima, kwaye ukusebenza kubalwa zenziwa ngokulandelelana kwemibhobho ukwandisa umthamo.

Ukuqukuqela okuNgenisayo okuLula:

Igalelo ledatha (umzekelo, umfanekiso okanye isignali yesandi) iqhutyelwe phambili.
Idatha ingena kwi Injini ye-tensor ye-NPU kubalo lwe-neural.
Imodeli ubunzima kunye nokuthambekela khokela inkcazo.
Iziphumo zihanjiswa ngexesha lokwenyani, rhoqo ngaphandle koncedo lwamafu.

Ngokudibanisa ukwenziwa kwe-latency ephantsi, uyilo oluwonga amandla, kwaye ezinikezelweyo AI logic, iiNPU zenza izixhobo zisebenze Iimodeli ze-AI ezintsonkothileyo ekuhlaleni. Oku kubenza babaluleke kakhulu edge computing, izixhobo smart, kwaye izicelo zamashishini AI apho ukwenza izigqibo ngexesha lokwenyani ibalulekile.

Izinto eziNzuzo eziPhambili zeeNPU

Ukwamkelwa kwe IiYunithi zeNeural Processing (NPUs) itshintshe ukusasazwa kwe Inkcazo ye-AI kuzo zombini izixhobo zomphetho kwaye iindawo zamafu. IiNPU zibonelela a ulwakhiwo olukhethekileyo lokulungisa eyenza ukuba ukusebenza okubalulekileyo kunye neenzuzo zokonga amandla ngaphezu kwesiqhelo ICPU kwaye GPU izisombululo xa kusenziwa iimodeli zokufunda nzulu.

IiNzuzo eziphezulu zokusebenzisa ii-NPU:

Ukusebenza okuphezulu kwe-AI Inference
IiNPU zenzelwe injongo AI imithwalo yomsebenzi, ukuphumeza imisebenzi yenethiwekhi ye-neural ngesantya esikhulu kunye ne-latency esezantsi. Baqhuba Ubalo lwexesha elinxuseneyo ngokufanelekileyo, ukuhambisa ulwazi lwexesha lokwenyani kwizicelo ezifana ukubhaqwa kwento, ukwahlulwa kwemifanekiso, kwaye ukuqondwa kwentetho.
Ukusebenza kakuhle kwamandla
Ii-NPU zisebenzisa amandla amancinci kakhulu kune-GPUs, ngenxa yazo izibalo ezichanekileyo ezisezantsi (umz., INT8, FP16) kunye nolwakhiwo oluphuculweyo. Oku kubenza balungele mobile AI kwaye Izixhobo ze-IoT apho ubomi bebhetri kwaye imida thermal zibalulekile.
Kwisixhobo se-AI yezakhono
Ngokuvumela ingqikelelo yendawo, ii-NPU zinciphisa imfuno yokulungiswa kwamafu, ukuphucula ubumfihlo bedatha, ixesha lokuphendula, kwaye Ukuphumelela kwenethiwekhi ye-bandwidth.
Uyilo oluQinisekileyo nolunoCalekayo
IiNPU zinokudityaniswa Inkqubo-kwi-Chip (SoC) uyilo, evumela abavelisi ukuba bazinzise AI accelerators zibe izixhobo compact ezifana ii-smartphones, ezinxitywayo, edge gateways, kwaye iinkqubo ezizimeleyo.

Itheyibhile yokuthelekisa: I-NPU vs. GPU kunye ne-CPU

Uphawu	NPU	GPU	ICPU
Injongo	Inkcazo ye-AI	Imizobo, jikelele AI	Icomputing yenjongo ngokubanzi
Ukusebenza kakuhle kwamandla	⭐⭐⭐⭐⭐	⭐⭐	⭐
Ukubambezeleka (Ngexesha lokwenyani)	Phantsi	Phakathi	Phezulu
I-Parallel AI Workloads	Egqwesileyo	Kakuhle kakhulu	Ilinganiselwe
Imeko yokusetyenziswa efanelekileyo	Edge AI, Mobile AI	Uqeqesho lwe-AI, imizobo	Ingqiqo yokulawula, imisebenzi ye-OS

Ngokudibanisa ii-NPUs kwizisombululo zekhompuyutha edge ye-AI, amashishini ayakwazi ukuvula i-high-speed machine learning inference kunye neendleko zokusebenza eziphantsi, okwenza ii-NPU zibe sisiseko se-hardware ye-AI yesizukulwana esilandelayo.

I-NPU vs. CPU, GPU, kunye ne-TPU: Uhlalutyo lokuthelekisa

Njengoko imfuneko ixesha lokwenyani AI inference ikhula kuzo zonke izixhobo-ukusuka kwii-smartphones kunye nezithuthi ezizimeleyo ukuya kwiirobhothi zamashishini-ingxoxo-mpikiswano malunga nokuba yeyiphi iprosesa ephethe kakuhle lo mthwalo wemisebenzi iyaqina. Ngelixa CPUs, GPUs, TPUs, kwaye IiNPU ingenza konke ukufunda koomatshini imisebenzi, zilungiselelwa iinjongo ezahlukeneyo kakhulu. Ukuqonda lo mahluko kubalulekile xa ukhetha ihardware ye Usetyenziso lwe-AI, ngakumbi kwi edge.

CPU

I ICPU ihlala iyeyona iprosesa enemisebenzi emininzi. Iphatha uluhlu olubanzi lwemisebenzi yecomputing yenjongo jikelele, kubandakanywa imisebenzi yenkqubo yokusebenza, Ulawulo lwe-I/O, kwaye logic elandelelanayo. Ngelixa ii-CPU zikwazi ukwenza i-AI inference, azilungiselelwanga ukusetyenzwa ngokunxuseneyo ukuba iinethiwekhi ze-neural zifuna.

IGPU (IYunithi yokuLungiselela imizobo)

Ekuqaleni iphuhliselwe ukunikezelwa kwemifanekiso kunye nevidiyo, i GPU ibonakalise ukusebenza kakuhle kakhulu imisebenzi ehambelanayo, iyenza ifaneleke zombini Uqeqesho lwe-AI kwaye uthelekiso. IiGPU ziyagqwesa imimandla yeziko ledatha, apho ukusetyenziswa kwamandla kunye nendawo ingathinteli kakhulu.

I-TPU (iYunithi yoLungiso lweTensor)

Iphuhliswe nguGoogle, i TPU yenzelwe tensor computation kwi-AI yomthwalo wemisebenzi. Inikezela ngoqeqesho olunamandla kunye nokusebenza okuthelekelelwayo kodwa ifumaneka ngokuyintloko Isiseko selifu okanye ihardware yodidi lweshishini.

NPU (Neural Processing Unit)

Yenzelwe ukusuka phantsi ukuya phezulu Inkcazo ye-AI, i NPU inikeza ukusebenza okungahambelaniyo ngewatt nganye. Ngoku ifakwe ngokubanzi ngaphakathi Inkqubo-kwi-Chip (SoC) uyilo kuzo zonke ii-smartphones, izixhobo zomda we-AI, kunye noomatshini abazimeleyo.

Itheyibhile yokuthelekisa iProsesa

Uphawu	ICPU	GPU	TPU	NPU
Yenzelwe	Iinjongo jikelele	Imizobo kunye ne-AI	I-Tensor Ops	Inkcazo ye-AI
Uqeqesho lwe-AI	Ubuhlwempu	Egqwesileyo	Kakuhle kakhulu	Ilinganiselwe
Inkcazo ye-AI	Ilinganiselwe	Kulungile	Egqwesileyo	Egqwesileyo
Ukusebenza kakuhle kwamandla	Phantsi	Phakathi	Phantsi	Phezulu
Ukubambezeleka (Edge Tasks)	Phezulu	Phakathi	Phantsi	Isezantsi kakhulu
Ukufaneleka kweMobile/IoT	Ubuhlwempu	Ubuhlwempu	Ilinganiselwe	Egqwesileyo
Inkxaso yeNkqubo-sikhokelo	Banzi	Banzi	Imxinwa (Google)	Phakathi (kuyahluka)

Izinto eziphambili zokuThatha:

GPUs zikhethwa Uqeqesho lwemodeli ye-AI kwaye ukhomyutha yamafu.
TPUs zisemafini kwaye zingcono ngaphakathi I-Google AI ecosystem.
CPUs ziingcali ngokubanzi, azilungele ukufunda nzulu.
IiNPU zezona zikhethwayo kwisixhobo AI, ukunika i eyona ibhalansi yesantya, ukusebenza kakuhle, kunye namandla aphantsi ukuze kuqikelelwe umthwalo womsebenzi.

Njengoko i-AI iqhubeka nokufudukela kwi edge, i-NPU igqame njengesona sisombululo sisebenza ixesha elide ixesha lokwenyani, ukufunda ngoomatshini okuzinzisiweyo.

Zisetyenziswa phi iiNPU? Usetyenziso oluphambili

IiYunithi zeNeural Processing (NPUs) ngoku ziyinxalenye yesiseko kuluhlu olubanzi lwe Itekhnoloji eqhutywa yi-AI, ngakumbi abo bafuna uthelekiso lwesixhobo, ukusetyenziswa kwamandla aphantsi, kwaye ukuphendula ngexesha lokwenyani. Njengoko amashishini eqhubeka nokudityaniswa kukubhadla okungeyonyani kwiinkqubo zabo zokusebenza, iiNPU zamkelwe kuzo zombini i-elektroniki yabathengi kwaye iinkqubo zoshishino.

1. Ii-Smartphones kunye nezixhobo eziphathwayo

Yangoku ii-smartphones baxhotyiswe ngeeNPUs ukuya phambili Iimpawu ze-AI njenge:

Ukuqaphela ubuso (umzekelo, uqinisekiso lwebhayometriki)
Ukomelezwa kweefoto kwaye ukubhaqwa kwento kwiiapps zekhamera
Abancedisi belizwi kwaye ukuguqulelwa kolwimi
Inyani eyandisiweyo (AR) ukunikezela ngexesha langempela

I-chipsets ehamba phambili ehamba phambili njenge Apple's A-series Neural Engine, Qualcomm Snapdragon, kwaye IHuawei's Kirin SoC hlanganisa ii-NPU ngokufanelekileyo edge processing AI.

2. I-Edge Computing kunye ne-IoT

Ehlabathini le edge AI, iiNPU ziyasebenza iikhamera smart, izixhobo ezinxitywayo, kwaye iihabhu ezizisebenzelayo zasekhaya ukubaleka nzima iimodeli zokufunda koomatshini ekuhlaleni. Oku kunciphisa ukuthembela kwiiseva zefu kunye nokuqinisekisa ngokukhawuleza ukwenza isigqibo, nakwiindawo ezinoqhagamshelo olunyiniweyo.

3. Iimoto kunye neeRobhothi

IiNPU ziya zisanda ukusasazwa ngaphakathi izithuthi ezizimeleyo kwaye iinkqubo zerobhothi ukuqhubekeka:

Ukudityaniswa kwenzwa
Ukufunyanwa kwendlela
Ingqwalasela yabahambi ngeenyawo
Iinkqubo zomyalelo welizwi

Ngokuqhuba intelekelelo ngqo kwisithuthi okanye kwirobhothi, ii-NPU ziqinisekisa ukusetyenzwa kwexesha lokwenyani nge-latency encinci.

4. I-AI yoShishino kunye nezeMpilo

IiNPU zikwasetyenziswa kwi automation yefektri, ukugcinwa kwangaphambili, kwaye umfanekiso wezonyango. Kwi ukwenziwa kobuchule, iiNPU ziyasebenza Iinkqubo zombono we-AI ukuchonga iziphene, ukwahlula izixhobo, kunye nokubeka iliso kwimisebenzi ngokuqhubekayo ngaphandle kwesidingo soqhagamshelo lwamafu aphezulu.

Abathengisi abakhulu beNPU kunye ne-Ecosystems

Ukukhula kwe I-AI ekupheleni kukhokelele kutyalo-mali olubalulekileyo kwi Uphuhliso lwezixhobo ze-NPU ziinkampani ezinkulu zetekhnoloji. Aba bathengisi babumba ubume belizwe I-AI elungisiweyo ngokunika izisombululo ezakhelwe injongo ezilungiselelwe isantya sokucinga, ukusebenza kakuhle kwamandla, kwaye ukuguquguquka kokudibanisa. Uphawu ngalunye lunolwalo Uyilo lwe-NPU, izixhobo, kunye nenkxaso ye-ikhosistim.

1. I-Apple - Injini ye-Neural

I-Apple's Injini yeNeural, idityaniswe kuyo A-series kunye M-series chips, inika amandla imisebenzi ye-AI ephezulu kwii-iPhones, ii-iPads, kunye neeMacs. Inika:

Ixesha elilungile ukuqondwa kobuso
Ukuqhubekeka kweSiri kwisixhobo
Uphuculo lweefoto kunye nevidiyo ehlakaniphile

2. Huawei – Da Vinci Architecture

IHuawei Ukunyuka kweeNPU kwaye Uyilo lweDa Vinci zingundoqo kuyo IiKirin SoCs kwaye Iziseko ze-AI:

Inkcazo ekhawulezileyo ye-AI ku izixhobo eziphathwayo
Ukusasazwa ngaphakathi amaziko data yeshishini AI
Ixhasa I-MindSpore Isakhelo se-AI

3. I-Qualcomm – i-Hexagon DSP ene-AI Engine

I-Qualcomm's Hexagon DSP + AI Injini ifakwe kumaqonga e-Snapdragon, anikezela:

Ukukhawuleza kwe-AI yeapps ze-Android
Esebenzayo ukusetyenzwa kwesixhobo
Ukuhambelana ne I-TensorFlow Lite kwaye ONNX

4. Google – Edge TPU

I Umgca weTPU, inxalenye yeGoogle Iqonga lekorale, yenzelwe IoT kunye nokuthunyelwa komda:

Ilungiselelwe I-TensorFlow Lite
Ultra-low amandla AI inference
Compact form factor ye abenzi boluvo abakrelekrele kwaye amasango

5. I-NVIDIA - i-NVDLA kunye ne-Jetson Platform

Kwi-NVIDIA NVDLA (I-Accelerator yokuFunda nzulu) kunye Jetson series hambisa ikhompuyutha ye-AI esebenza kakhulu:

Isetyenziswa kwi iirobhothi, iidrones, kwaye oomatshini abazimeleyo
Ixhasa EZAHLUKENEYO, I-TensorRT, kwaye DeepStream SDK

Imingeni kunye neMida yeeNPU

Ngelixa IiYunithi zeNeural Processing (NPUs) zibonelela ngeenzuzo ezimangalisayo kwi Ukusebenza kwe-AI, ukusebenza kakuhle kwamandla, kwaye edge computing, azinamida. Abaphuhlisi kunye nabahlanganisi benkqubo kufuneka baqwalasele imingeni emininzi yobugcisa kunye nokusebenza xa besamkela iiNPUs Izixhobo ezenziwe nge-AI.

1. Ukuhambelana kweNkqubo-sikhokelo

Omnye wemiqobo ephambili yinkxaso encinci ye Izikhokelo ze-AI. Ngokungafaniyo neeGPUs, ezixhasa uluhlu olubanzi lwamaqonga anje I-TensorFlow, I-PyTorch, kwaye ONNX, iiNPU zihlala zifuna izixhobo zesiko okanye ii-SDKs zobunini. Oku kunokukhokelela ekubeni:

Imijikelo emide yophuhliso
Ubunzima kwi uguqulo lwemodeli kunye nokwenza ngcono
Ukusebenza okungahambelaniyo kubo bonke abathengisi behardware

2. UkuQikelela-Kuphela Ukukwazi

Uninzi lweeNPUs ziyilelwe kuphela uthelekiso, hayi uqeqesho. Ithetha ukuba:

Iimodeli ze-AI kufuneka ziqeqeshwe IiGPUs okanye iiTPUs
Nakuphi na ukuqeqeshwa kwakhona okanye ukulungiswa kakuhle kufuna iziseko ezingundoqo zangaphandle
Ukunciphisa ukuguquguquka kwi ukufunda kwisixhobo

3. Imiqobo yokuDityaniswa kweHardware

Ukudibanisa ii-NPU kwiinkqubo ezizinzisiweyo okanye Inkqubo-kwi-Chip (SoC) Uyilo lubandakanya urhwebo:

Ukulinganiselwa kwendawo ye-silicon
Izithintelo zoyilo lwe-Thermal
Iingxabano ezinokwenzeka kunye ekhoyo I/O kunye bandwidth memory

Ikamva leeNPU kunye ne-AI Hardware

Njengemfuno ye ixesha lokwenyani AI inference iyaqhubeka ukukhula kuwo wonke amashishini, ikamva le IiYunithi zeNeural Processing (NPUs) kulindeleke ukuba ibe sembindini wenguquko ye AI hardware. Ezi processors zikhethekileyo ziba namandla ngakumbi, zihlangene, kwaye zisebenza ngokufanelekileyo-zizibeka njengeyona nto ibalulekileyo kwisizukulwana esilandelayo. edge izixhobo AI, abenzi boluvo abakrelekrele, kwaye iinkqubo ezizimeleyo.

IiNdlela ezisakhulayo kuPhuhliso lwe-NPU

Udibaniso lwe-SoC oluqinile
IiNPU ziya zidityaniswa ngakumbi Inkqubo-kwi-Chip (SoC) uyilo ecaleni kwe-CPU, ii-GPU, kunye nee-ISPs. Olu lwakhiwo olumanyeneyo lunika amandla ngokukhawuleza imibhobho yokucubungula idatha, kunciphisa i-latency, kwaye yehlisa ubunzima benkqubo.
Inkxaso ye-Heterogeneous Computing
Ii-NPU zexesha elizayo ziya kusebenza ngaphandle komthungo kunye nezinye ii-accelerators ze-AI, zisasaza imithwalo yemisebenzi kulo lonke. IiCPU, iiGPU, kunye neeTPU ukusebenza ngokugqibeleleyo.
Ukuhambelana kweNkqubo-sikhokelo ephuculweyo
Abathengisi bakha inkxaso kwiinkqubo ze-AI eziqhelekileyo ezifana ONNX, I-TensorFlow Lite, kwaye PyTorch Mobile, ukunciphisa i ijiko lokufunda lophuhlisi kunye nokukhuthaza ukusasazwa kweqonga.
Ukwandiswa okuNgaphaya kweziXhobo zabaThengi
IiNPU ziyakudlala indima enkulu i-automation yemizi-mveliso, ukhathalelo lwezempilo oluhlakaniphile, iirobhothi zomda, ezolimo AI, kwaye iinkqubo zokucupha, apho amandla aphantsi, isantya esiphezulu ibalulekile.

Kwihlabathi elisingise ngakwicala ubukrelekrele kwindawo yonke, ii-NPU aziyi kukhawulezisa i-AI kodwa ziyenze ibe ngaphezulu iyafikeleleka, zizinzileyo, kwaye ikhuselekile. Indima yabo ekubumbeni iinkqubo ezizinzisiweyo ezizayo, ukusuka Iikhamera ze-AI-powered ukuya amaqonga okukhangela azimeleyo, igxininisa ukubaluleka kokukhula kwe i-domain-specific compute architecture kwixesha le-AI.