Ibhulogi

Iyini i-NPU?

2024-08-13 16:29:49

Ekuguqukeni kwanamuhla ngokushesha Umhlaba oqhutshwa yi-AI, isidingo sokubala esisheshayo, esisebenza kahle kakhudlwana sibangele amaphrosesa akhethekile afana ne I-Neural Processing Unit (NPU). Ngokungafani nendabuko Ama-CPU noma Ama-GPU, i-NPU yakhelwe inhloso ukusheshisa isisekelo sokufunda ngomshini, ikakhulukazi ye amanethiwekhi e-neural ajulile esetshenziswa ku umbono wekhompyutha, ukucutshungulwa kolimi lwemvelo, futhi ukuqashelwa kwezwi.

An I-NPU wuhlobo lwe I-accelerator ye-AI eletha ukusebenza okuphezulu, ukucubungula okuphansi kwemisebenzi edingekayo ukubala ngokuhambisana futhi ukusebenza kwe-tensor. Lawa maphrosesa manje ashumekwe ezinhlobonhlobo zezinkundla—kusuka ama-smartphone futhi Amadivayisi we-IoT ku izinhlelo zezimoto futhi ingqalasizinda yekhompyutha- kuvumela kudivayisi AI ngokusebenzisa amandla amancane.

Kungani ama-NPU Abalulekile:

Ukusebenza kwesikhathi sangempela okuthuthukisiwe ngemithwalo yemisebenzi ye-AI
Ukusetshenziswa kwamandla okuphansi uma kuqhathaniswa nezixazululo ezisekelwe ku-GPU
Ukuncishiswa kokuncika ekuqondeni kwefu, ukuthuthukisa ubumfihlo bedatha nokubambezeleka
I-Scalability kumadivayisi ahlangene namasistimu ashumekiwe

Ukunyuka kwe onqenqemeni AI kanye nokwanda kwe Izinhlelo zokusebenza ezinikwe amandla yi-AI ikwenze kubaluleke ukuqonda ukuthi iyini i-NPU nokuthi ihluke kanjani kwamanye amayunithi okucubungula. Kulesi sihloko, sihlola i- izakhiwo, sebenzisa amacala, kanye nenani lesu lama-NPU kumongo obanzi wekhompyutha ehlakaniphile.

Yini i-NPU? Ukubuka konke Kwezobuchwepheshe

A I-Neural Processing Unit (NPU) iyi-microprocessor ezinikele eklanyelwe ukuphatha ngqo i-artificial intelligence (AI) imisebenzi eminingi, ikakhulukazi leyo ehilelekile ukufunda okujulile futhi i-neural network inference. Ngokungafani nenhloso evamile Ama-CPU noma ihluzo-ngamakhasimende Ama-GPU, ama-NPU enzelwe imisebenzi yezibalo amandla amamodeli wokufunda womshini, njenge ukuphindaphinda kwe-matrix, ama-convolutions, futhi i-tensor processing.

Izici ezibalulekile ze-NPU:

Inhloso eyakhelwe ukuqondiswa kwe-AI, hhayi ukuqeqeshwa
Ikhiphe imisebenzi ehambisanayo ngempumelelo
Kulungiselelwe i-low-latency, i-high-throughput imithwalo yemisebenzi
Isebenza ngokuphawulekayo ukusetshenziswa kwamandla okuphansi kunama-GPU

Ngenkathi Ama-CPU zilungele ukulandelana kwe-logic kanye Ama-GPU uphumelele emisebenzini yokuhlinzeka kanye nemidlalo, Ama-NPU zenzelwe ukusheshisa imisebenzi efana nale:

Ukutholwa kwento
Ukuqashelwa kobuso
Ukuguqulwa kwezwi ukuya kombhalo
Izibalo zekhamera ehlakaniphile

Uhlobo Lwephrosesa	Indima Eyinhloko	Amandla	Ukusebenza kahle kwamandla	Ukufaneleka kwe-AI
CPU	Imisebenzi ejwayelekile	Okuhlukahlukene	Phansi	Inomkhawulo
I-GPU	Graphics, AI	Ukusebenza okuphezulu	Maphakathi	Iqinile ekuqeqesheni
I-NPU	Incazelo ye-AI	I-AI-ilungiselelwe	Phezulu	Kungcono kakhulu kwe-AI enqenqemeni

Njengoba i-AI ishumeka kumadivayisi amaningi—kusuka ama-smartphone ku IoT yezimboni-the I-NPU igqama kakhulu ngempumelelo futhi scable iprosesa yokuvumela izici ezihlakaniphile ngqo ku- umphetho, ngaphandle kokuthembela ekufinyeleleni kwamafu njalo.

Isebenza Kanjani I-NPU?

A I-Neural Processing Unit (NPU) yakhelwe ukwenza Imisebenzi ye-AI yokubonisa ngesivinini esingavamile kanye nokusebenza kahle. Ngokungafani nama-processor wendabuko, ama-NPU ayasebenza izakhiwo zedatha esekwe kwi-tensor, sisebenzisa amanani amakhulu we imisebenzi ye-matrix ngokuhambisana—kubalulekile ekusebenzeni kwesimanje amamodeli okufunda ajulile njenge Ama-CNNs (Convolutional Neural Networks) futhi Ama-RNN (Amanethiwekhi Avamile We-Neural).

Izimiso Zokusebenza Eziyinhloko:

Ukufana: Ama-NPU ayadlala imisebenzi eminingi kanyekanye, kunciphisa kakhulu isikhathi sokucubungula semisebenzi efana ukuhlukaniswa kwezithombe futhi ukuqashelwa kwenkulumo.
Ukubalwa kwe-tensor okusheshiswe ngezingxenyekazi zekhompuyutha: Yakhelwe ngaphakathi ama-tensor cores phatha kahle izikali ezinkulu ukuphindaphinda kwe-matrix.
Izibalo ezinemba ephansi: Ama-NPU amaningi asebenzisa I-INT8 noma FP16 ukunemba amafomethi ukuze uthole ibhalansi phakathi ukunemba futhi ukusebenza kahle kwamandla.
Isakhiwo esinamapayipi: Imisebenzi efana ukulanda idatha, ukucubungula isisindo, futhi ukwenza kusebenze ukubala zenziwa ngokulandelana kwamapayipi ukuze kukhuliswe ukuphuma.

Ukugeleza Kwemibono Elula:

Idatha yokufaka (isb, isithombe noma isignali yomsindo) icutshungulwa ngaphambili.
Idatha ingena ku Injini ye-tensor ye-NPU ngokubala kwe-neural.
Imodeli izisindo nokuchema qondisa isiphetho.
Imiphumela ilethwa ngesikhathi sangempela, ngokuvamile ngaphandle kosizo lwamafu.

Ngokuhlanganisa ukwenza ukubambezeleka okuphansi, idizayini eyonga amandla, futhi i-AI enengqondo ezinikele, ama-NPU anika amandla amadivayisi ukuthi asebenze amamodeli e-AI ayinkimbinkimbi endaweni. Lokhu kubenza babaluleke kakhulu kubo ikhompuyutha enqenqemeni, amadivaysi ahlakaniphile, futhi izinhlelo zokusebenza ze-AI zezimboni lapho ukwenza izinqumo ngesikhathi sangempela kubalulekile.

Izinzuzo Eziyinhloko zama-NPU

Ukwamukelwa kwe I-Neural Processing Units (NPUs) iguqule ukuthunyelwa kwe Incazelo ye-AI kukho kokubili amadivaysi onqenqema futhi izindawo zamafu. Ama-NPU ahlinzeka nge ukwakhiwa okukhethekile kokucubungula enika amandla ukusebenza okubalulekile kanye nezinzuzo zokonga amandla kunokwejwayelekile CPU futhi I-GPU izixazululo lapho kwenziwa amamodeli okufunda ajulile.

Izinzuzo eziphezulu zokusebenzisa ama-NPU:

Ukusebenza okuphezulu kwe-AI Inference
Ama-NPU akhelwe inhloso Imisebenzi ye-AI, ukwenza imisebenzi yenethiwekhi ye-neural ngesivinini esikhulu kanye nokubambezeleka okuphansi. Bayacubungula i-parallel tensor computations kahle, iletha imininingwane yesikhathi sangempela kuzinhlelo zokusebenza ezifana ukutholwa kwento, ukuhlukaniswa kwesithombe, futhi ukuqashelwa kwenkulumo.
Ukusebenza Kwamandla
Ama-NPU asebenzisa amandla amancane kakhulu kunama-GPU, ngenxa yawo izibalo zokunemba okuphansi (isb, INT8, FP16) kanye nezakhiwo ezithuthukisiwe. Lokhu kubenza balungele I-AI yeselula futhi Amadivayisi we-IoT lapho impilo yebhethri futhi imikhawulo ezishisayo zibucayi.
Amakhono e-AI akudivayisi
Ngokunika amandla ukucatshangelwa kwendawo, ama-NPU anciphisa isidingo sokucubungula amafu, ukuthuthukisa ubumfihlo bedatha, isikhathi sokuphendula, futhi ukusebenza kahle komkhawulokudonsa wenethiwekhi.
Idizayini Encane Neyokukala
Ama-NPU angahlanganiswa ku Isistimu-on-Chip (SoC) imiklamo, evumela abakhiqizi ukuthi bashumeke Ama-accelerator e-AI kumadivayisi ahlangene njenge ama-smartphone, okugqokwayo, emaphethelweni amasango, futhi izinhlelo ezizimele.

Ithebula lokuqhathanisa: I-NPU vs. GPU ne-CPU

Isici	I-NPU	I-GPU	CPU
Inhloso	Incazelo ye-AI	Graphics, jikelele AI	Ikhompyutha yenhloso evamile
Ukusebenza kahle kwamandla	⭐⭐⭐⭐⭐	⭐⭐	⭐
Ukubambezeleka (Isikhathi sangempela)	Phansi	Maphakathi	Phezulu
I-Parallel AI Workloads	Kuhle kakhulu	Kuhle kakhulu	Inomkhawulo
I-Ideal Use Case	I-Edge AI, i-Mobile AI	Ukuqeqeshwa kwe-AI, ihluzo	Control logic, OS imisebenzi

Ngokuhlanganisa ama-NPU kuzixazululo zekhompiyutha enqenqemeni ye-AI, amabhizinisi angavula inkomba yokufunda yomshini enesivinini esikhulu ngezindleko eziphansi zokusebenza, okwenza ama-NPU abe isisekelo sehadiwe ye-AI yesizukulwane esilandelayo.

I-NPU iqhudelana ne-CPU, i-GPU, ne-TPU: Ukuhlaziya Okuqhathanisayo

Njengesidingo sokuba real-time AI inference ikhula kuwo wonke amadivayisi—kusuka kuma-smartphones kanye nezimoto ezizimele kuya kumarobhothi ezimboni—impikiswano ngokuthi iyiphi iphrosesa eyisingatha kangcono le mithwalo yomsebenzi iyaqina. Ngenkathi Ama-CPU, Ama-GPU, Ama-TPU, futhi Ama-NPU konke kungakwenza ukufunda ngomshini imisebenzi, yenzelwe izinjongo ezihluke kakhulu. Ukuqonda lo mehluko kubalulekile lapho ukhetha ihadiwe Izinhlelo zokusebenza ze-AI, ikakhulukazi e- umphetho.

I-CPU (Iyunithi Yokucubungula Emaphakathi)

I CPU ihlala iyiprosesa enemisebenzi eminingi. Iphatha uhla olubanzi lwemisebenzi yekhompuyutha yenhloso ejwayelekile, okuhlanganisa imisebenzi yesistimu yokusebenza, Abaphathi be-I/O, futhi logic elandelanayo. Ngenkathi ama-CPU ekwazi ukwenza inkomba ye-AI ngokobuchwepheshe, awathuthukisiwe ukucubungula okuhambisanayo lokho kudinga amanethiwekhi e-neural.

I-GPU (Iyunithi Yokucubungula Imifanekiso)

Ekuqaleni yakhelwe ukunikeza izithombe namavidiyo, i I-GPU ibonakale iphumelela kakhulu imisebenzi ehambisanayo, okwenza ifaneleke kokubili Ukuqeqeshwa kwe-AI futhi ukucabangela. Ama-GPU ahamba phambili izindawo zesikhungo sedatha, lapho ukusetshenziswa kwamandla nesikhala kungavinjelwa kancane.

I-TPU (Iyunithi Yokucubungula I-Tensor)

Ithuthukiswe i-Google, i I-TPU yakhelwe tensor computation emisebenzini ye-AI. Iletha ukuqeqeshwa okunamandla nokusebenza okuqondile kodwa itholakala ngokuyinhloko ingqalasizinda yefu noma ihadiwe yezinga lebhizinisi.

I-NPU (Neural Processing Unit)

Idizayinelwe kusukela phansi kuya phezulu Incazelo ye-AI, i I-NPU inikeza ukusebenza okungenakuqhathaniswa nge-watt ngayinye. Manje isishumekwe kakhulu phakathi Isistimu-on-Chip (SoC) iklama kuwo wonke ama-smartphones, amadivayisi we-AI onqenqemeni, nemishini ezimele.

Ithebula Lokuqhathanisa Lomprosesa

Isici	CPU	I-GPU	I-TPU	I-NPU
Idizayinelwe	Inhloso-jikelele	Imidwebo ne-AI	I-Tensor Ops	Incazelo ye-AI
Ukuqeqeshwa kwe-AI	Impofu	Kuhle kakhulu	Kuhle kakhulu	Inomkhawulo
Incazelo ye-AI	Inomkhawulo	Kuhle	Kuhle kakhulu	Kuhle kakhulu
Ukusebenza kahle kwamandla	Phansi	Maphakathi	Phansi	Phezulu
Ukubambezeleka (I-Edge Tasks)	Phezulu	Maphakathi	Phansi	Phansi Kakhulu
Ukufaneleka kweselula/IoT	Impofu	Impofu	Inomkhawulo	Kuhle kakhulu
Ukusekela Uhlaka	Okubanzi	Okubanzi	Narrow (Google)	Okumaphakathi (kuyahluka)

Okuthathwayo Okubalulekile:

Ama-GPU ziyakhethwa Ukuqeqeshwa kwemodeli ye-AI futhi i-cloud computing.
Ama-TPU amafu-centric futhi ezihamba phambili ngaphakathi I-Google AI ecosystem.
Ama-CPU bangama-generalists, abakulungele ukufunda okujulile.
Ama-NPU kukhona ongakhetha kukho kudivayisi AI,enikela i ibhalansi engcono kakhulu yejubane, ukusebenza kahle, namandla aphansi ukuze kutholwe imithwalo yemisebenzi.

Njengoba i-AI iqhubeka nokuthuthela ku- umphetho, i-NPU igqama njengesixazululo sesikhathi eside esisebenza kakhulu ngesikhathi sangempela, ukufunda komshini okushumekiwe.

Asetshenziswa kuphi ama-NPU? Izinhlelo zokusebenza ezibalulekile

I-Neural Processing Units (NPUs) manje ziyingxenye yesisekelo ezinhlobonhlobo ze Ubuchwepheshe obuqhutshwa yi-AI, ikakhulukazi labo abadinga okucatshangwayo kudivayisi, ukusetshenziswa kwamandla okuphansi, futhi ukuphendula kwesikhathi sangempela. Njengoba izimboni ziqhubeka nokudidiyela ukuhlakanipha okungekhona okwangempela ekuhambeni kwawo komsebenzi, ama-NPU amukelwa kukho kokubili umthengi electronics futhi izinhlelo zezimboni.

1. Omakhalekhukhwini kanye Namadivayisi Eselula

Okwesimanje ama-smartphone afakwe ama-NPU ukuze anikeze amandla athuthukile Izici ze-AI njenge:

Ukuqashelwa kobuso (isb, ukuqinisekiswa kwebhayomethrikhi)
Ukuthuthukisa isithombe futhi ukutholwa kwento ezinhlelweni zokusebenza zekhamera
Abasizi bezwi futhi ukuhumusha ulimi
I-Augmented reality (AR) ukunikeza ngesikhathi sangempela

I-chipset ehamba phambili yeselula efana I-Apple's A-series Neural Engine, I-Qualcomm Snapdragon, futhi IHuawei Kirin SoC yeHuawei hlanganisa ama-NPU ukuze asebenze kahle ukucubungula kwe-AI.

2. I-Edge Computing ne-IoT

Emhlabeni we onqenqemeni AI, ama-NPU anika amandla amakhamera ahlakaniphile, izinto ezigqokekayo, futhi amahabhu e-automation yasekhaya ukugijima eziyinkimbinkimbi amamodeli wokufunda womshini endaweni. Lokhu kunciphisa ukuthembela kumaseva wamafu futhi kuqinisekisa ngokushesha ukwenza isinqumbo, ngisho nasezindaweni ezinokuxhumeka okulinganiselwe.

3. Izimoto kanye namarobhothi

Ama-NPU aya ngokuya ethulwa kuwo izimoto ezizimele futhi amarobhothi amasistimu ukucubungula:

Ukuhlanganiswa kwezinzwa
Ukutholwa komzila
Ukuqashelwa kwabahamba ngezinyawo
Amasistimu womyalo wezwi

Ngokusebenzisa inkomba ngqo emotweni noma irobhothi, ama-NPU aqinisekisa ukucubungula ngesikhathi sangempela ngokubambezeleka okuncane.

4. I-AI Yezimboni Nokunakekelwa Kwezempilo

Ama-NPU nawo asetshenziswa ku i-automation yemboni, ukugcinwa okubikezelwayo, futhi izithombe zezokwelapha. Ku ukukhiqiza okuhlakaniphile, ama-NPU anika amandla Izinhlelo zombono ze-AI ukuhlonza amaphutha, ukuhlukanisa izinto zokwakha, nokuqapha ukusebenza ngokuqhubekayo ngaphandle kwesidingo sokuxhumeka kwamafu okunomkhawulokudonsa ophezulu.

Abathengisi Abakhulu be-NPU Hardware kanye ne-Ecosystems

Ukukhula kwe I-AI ekugcineni kuholele ekutshalweni kwezimali okubalulekile ku Ukuthuthukiswa kwezingxenyekazi zekhompuyutha ze-NPU yizinkampani ezinkulu zobuchwepheshe. Laba bathengisi bakha isimo sezwe I-AI eshumekiwe ngokunikeza izixazululo ezakhelwe inhloso ezilungiselelwe isivinini sokucabanga, ukusebenza kahle kwamandla, futhi ukuguquguquka kokuhlanganisa. Uhlobo ngalunye lunolwalo I-NPU Architecture, uchungechunge lwamathuluzi, nokusekelwa kwe-ecosystem.

1. I-Apple - Neural Engine

I-Apple's I-Neural Engine, ihlanganiswe kuyo A-series kanye ne-M-series chips, inika amandla imisebenzi ye-AI ethuthukisiwe kuma-iPhone, ama-iPads, nama-Mac. Iletha:

Isikhathi sangempela ukuqashelwa kobuso
Ukucutshungulwa kwe-Siri ekudivayisi
Izithuthukisi zesithombe nevidiyo ehlakaniphile

2. I-Huawei - I-Da Vinci Architecture

I-Huawei Enyuka ama-NPU futhi Da Vinci Architecture ziwumgogodla wayo I-Kirin SoCs futhi Ingqalasizinda ye-AI:

Ukuqonda okusheshisiwe kwe-AI ku amadivaysi eselula
Ukuthunyelwa ku izikhungo zedatha yebhizinisi i-AI
Isekela I-MindSpore Uhlaka lwe-AI

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

I-Qualcomm's I-Hexagon DSP + AI Engine ishumekwe kumapulatifomu e-Snapdragon, enikeza:

Ukusheshisa kwe-AI yezinhlelo zokusebenza ze-Android
Isebenza kahle ukucubungula kudivayisi
Ukuhambisana ne I-TensorFlow Lite futhi I-ONNX

4. I-Google – Edge TPU

I I-TPU ye-Edge, ingxenye ye-Google Ipulatifomu yamakhorali, yenzelwe I-IoT ne-edge deployments:

Kulungiselelwe I-TensorFlow Lite
I-ultra-low power AI inference
I-Compact form factor ye izinzwa ezihlakaniphile futhi amasango

5. I-NVIDIA – NVDLA kanye ne-Jetson Platform

I-NVIDIA I-NVDLA (I-Deep Learning Accelerator) kanye Jetson series hambisa ikhompuyutha ye-AI esebenza kahle kakhulu:

Isetshenziswe ku amarobhothi, ama-drones, futhi imishini ezimele
Isekela OKUHLUKILE, I-TensorRT, futhi I-DeepStream SDK

Izinselelo kanye nemikhawulo yama-NPU

Ngenkathi I-Neural Processing Units (NPUs) ukunikeza izinzuzo eziphawulekayo ku Ukusebenza kwe-AI, ukusebenza kahle kwamandla, futhi ikhompuyutha enqenqemeni, azinayo imingcele. Onjiniyela nabahlanganisi bohlelo kufanele bacabangele izinselele ezimbalwa zobuchwepheshe nezokusebenza lapho bamukela ama-NPU Amadivayisi anikwe amandla i-AI.

1. Ukuhambisana Kohlaka

Esinye sezithiyo eziyinhloko ukwesekwa okulinganiselwe Izinhlaka ze-AI. Ngokungafani nama-GPU, asekela uhla olubanzi lwamapulatifomu afana I-TensorFlow, I-PyTorch, futhi I-ONNX, ama-NPU avame ukudinga ama-toolchains angokwezifiso noma ama-SDK okuphathelene. Lokhu kungaholela kokuthi:

Imijikelezo yokuthuthukiswa emide
Kunzima ukungena ukuguqulwa kwemodeli nokwenza kahle
Ukusebenza okungahambisani kubo bonke abathengisi bezingxenyekazi zekhompuyutha

2. Ikhono Lokucabanga Kuphela

Ama-NPU amaningi akhelwe ngokukhethekile ukucabangela, hhayi ukuqeqeshwa. Lokhu kusho:

Amamodeli e-AI kumele aqeqeshwe kuwo Ama-GPU noma ama-TPU
Noma yikuphi ukuqeqeshwa kabusha noma ukulungiswa kahle kudinga ingqalasizinda yangaphandle
Ukunciphisa ukuguquguquka ku ukufunda kudivayisi

3. I-Hardware Integration Izingqinamba

Ukuhlanganisa ama-NPU kumasistimu ashumekiwe noma Isistimu-on-Chip (SoC) imiklamo ihlanganisa ukuhwebelana:

Imikhawulo yendawo ye-silicon
Izithiyo zokuklama ezishisayo
Izingxabano ezingaba nazo I/O ekhona kanye nomkhawulokudonsa wememori

Ikusasa lama-NPU kanye ne-AI Hardware

Njengesidingo se real-time AI inference iyaqhubeka nokukhula kuzo zonke izimboni, ikusasa le I-Neural Processing Units (NPUs) kulindeleke ukuthi ibe semqoka ekuziphendukeleni kwemvelo kwe I-AI hardware. Lawa maphrosesa ayisipesheli aya eba namandla kakhulu, ahlangene, futhi awonga amandla—awabeka njengengxenye eyinhloko yesizukulwane esilandelayo amadivaysi e-AI enqenqemeni, izinzwa ezihlakaniphile, futhi izinhlelo ezizimele.

Amathrendi Asafufusa Ekuthuthukisweni kwe-NPU

Ukuhlanganiswa kwe-SoC okuqinile
Ama-NPU aya ngokuya ahlanganiswa ku Isistimu-on-Chip (SoC) iklama eduze kwama-CPU, ama-GPU, nama-ISP. Lesi sakhiwo esihlanganisiwe sinika amandla ngokushesha amapayipi okucubungula idatha, yehlisa ukubambezeleka, futhi yehlisa ubunkimbinkimbi besistimu.
Usekelo lwe-Heterogeneous Computing
Ama-NPU esikhathi esizayo azosebenza ngaphandle komthungo namanye ama-accelerator e-AI, asabalalise imithwalo yemisebenzi kuwo wonke. Ama-CPU, ama-GPU, nama-TPU ukuze usebenze kahle.
Ukuvumelana Kohlaka Oluthuthukisiwe
Abathengisi bakha ukusekelwa kwezinhlaka ze-AI ezijwayelekile ezifana I-ONNX, I-TensorFlow Lite, futhi I-PyTorch Mobile, ukunciphisa i ijika lokufunda likanjiniyela kanye nokukhuthaza ukuthunyelwa kwezingxenyekazi ezihlukene.
Ukunwetshwa Ngaphandle Kwamadivayisi Abathengi
Ama-NPU azodlala indima enkulu ku ezishintshayo zezimboni, ukunakekelwa kwezempilo okuhlakaniphile, amarobhothi onqenqemeni, ezolimo AI, futhi izinhlelo zokugada, lapho amandla aphansi, ukuqagela kwesivinini esikhulu kubalulekile.

Ezweni eliya ngalapho ubuhlakani bendawo yonke, ama-NPU ngeke asheshise i-AI kuphela kodwa futhi ayenze ibe ngaphezulu kuyafinyeleleka, esimeme, futhi ivikelekile. Indima yabo ekubumbeni amasistimu ashumekiwe esizukulwane esilandelayo,kusuka Amakhamera asebenza nge-AI ku ezizimele navigation ezisekelweni, igcizelela ukubaluleka okukhulayo kwe ukwakheka kwekhompuyutha okuqondene nesizinda enkathini ye-AI.