Riaan Phawula Njenge-salmon ebuyela emanzini okhokho babo, ama-GPU akwaNvidia abuyela ezimbonini zama-chip.
I-TSMC, i-ASML, ne-Synopsys zonke zisebenzisa izisheshi ze-Nvidia ukusheshisa noma ukukhulisa i-lithography yekhompyutha. Ngaleso sikhathi, i-KLA Group, Applied Materials, kanye ne-Hitachi manje zisebenzisa ikhodi yokufunda ejulile esebenza ku-silicon yokucubungula ukufana kwe-Nv yokuhlola i-e-beam ne-optical wafer.
Vele, izindwangu zigcwele izinto zikagesi ngakho-ke ama-chips azo zonke izinhlobo abuyela endaweni ethile ngendlela. Kepha kuyajabulisa ukubona ama-GPU esheshisa le ngxenye yenqubo yokukhiqiza. Njengoba sishwabadela amasango e-transistor, okusisiza ukuthi siwapakishe amaningi kumaphrosesa ethu, i-computational lithography iyadingeka – njengoba kwenzeka iminyaka eminingana – ukukhiqiza ama-photomasks akwazi ukufaka izici zamasango ezihlala zincane ku-silicon.
Kuze kube muva nje le mithwalo yemisebenzi isebenze kakhulu kuma-CPU cores, noma kuvele ukuthi uma kunikezwe ama-tweaks afanele, ama-GPU mahle kakhulu ekusheshiseni imisebenzi enjalo.
Ku-computing lithography, inqubo iqondile. Ama-chips amaningi agxilwa ku-silicon ngokuveza ubude beza obuthile bokukhanya – ngokuvamile engxenyeni ye-ultraviolet eyeqisayo yesibonisi – nge-photomask. Ukuze kukhiqizwe ama-transistors amancane namancane ku-silicon efa, onjiniyela kuye kwadingeka babe nobuciko ukuze bavimbele, okokuqala, ukuhlanekezela ekufiphaliseni izici. Namuhla, lawa ma-photomasks ahlotshiswe kakhulu futhi akhiqizwa ngamaqoqo amakhulu wekhompiyutha angathatha amaviki ukuthi aqedwe.
Esebenzisa ukusheshisa kwe-GPU, i-CEO yakwaNvidia uJensen Huang uthi le nqubo ingasheshiswa ngo-50x. “Amashumi ezinkulungwane zamaseva e-CPU angathathelwa indawo yizinhlelo ze-DGX ezingamakhulu ambalwa, ukunciphisa amandla nezindleko ngokuhleleka kobukhulu,” esho kuleli sonto.
Ukusebenzisa i-AI ukwakha ama-chips angcono
Noma ngubani obenaka kusukela sekwephuzile uzokwazi ukuthi ama-graphic processors alungele ngaphezu kwekhompyutha esebenza kahle kakhulu. Futhi bayinhliziyo eshayayo lapho i-AI enkulu yanamuhla incike kuyo. Futhi ngenkathi uNvidia eyeka ukusho ukuthi ubuhlakani bokwenziwa bukhona kuyo kuLitho isitaki se-software ngoMashi, u-Vivek Singh, i-VP yeqembu lezobuchwepheshe elithuthukisiwe le-Nvidia, wakwenza kwacaca ukuthi kuyindaba yesikhathi ngaphambi kokuba i-AI isetshenziswe kabanzi ku-computing lithography.
Ku-semiconductor ye-ITF ingqungquthela NgoLwesibili kuleli sonto, uHuang kaNvidia uveze amandla e-AI okuphefumula impilo entsha kuMthetho kaMoore. “Ukukhiqizwa kwe-Chip kuwuhlelo olufanele lwe-Nvidia esheshisiwe kanye nekhompyutha ye-AI,” egcizelela. Hhayi-ke, ubengakusho lokho.
Nakuba izindwangu sezivele zizenzakalela kakhulu, i-Nvidia ibona ithuba lokusebenzisa umsebenzi awenzayo mayelana namarobhothi, izimoto ezizimele, nama-chatbots ekwenzeni ama-chip – futhi enze izigidi zithengise ama-GPU kubakhi bamaphrosesa, abathengisi bemishini, nabasunguli.
Ngesikhathi ekhuluma, u-Huang wagcona I-VIMAi-multimodal “imodeli ye-AI ehlanganisiwe eqeqeshelwe ukwenza imisebenzi esekelwe ekwazisweni kombhalo obonakalayo, njengokuhlela kabusha izinto ukuze zifane nesigcawu.”
U-Huang akashongo ngokucacile ukusebenzisa lokhu ku-software yokuklama i-chip, kodwa uma ingqungquthela igcizelela ekukhiqizeni ama-semiconductor, kulula ukubona ukuthi ingaba kanjani. “Ngibheke phambili kumarobhothi e-physics-AI kanye namawele edijithali asekelwe ku-Omniverse esiza ukuthuthukisa ikusasa lokwenza ama-chipmaking,” wanezela.
Isikhathi sesivuthiwe
Intshisekelo engazelelwe kaNvidia yokuthengisa imboni ye-semiconductor kukhompyutha esheshisiwe ayimangazi neze.
Ukulandela i- bhidlika yemakethe ye-GPU yabathengi kanye nobusika obuqhubekayo be-crypto, uphiko lwedathacenter ye-Nv yenze umsebenzi omkhulu wokuphakamisa inkampane. U-Huang waze wathatha a ukunqanyulwa kweholo kule kota edlule ngenxa yokusebenza kahle kwenkampani yakhe.
Naphezu kokwehla kwakamuva kwesidingo se-chip – ikakhulukazi ekupheleni – opharetha be-Foundry, njenge-Samsung, TSMC, Intel, SK hynix phakathi kwabanye, baqhubekela phambili ngamaphrojekthi amasha wokusungula. Lokhu kwanda kugqugquzelwe kakhulu uchungechunge lwezinyathelo esifundeni sase-US, Europe, ne-Asia Pacific, ezihlangene zibiza imali engaphezu kwezigidigidi eziyi-100 zamaRandi. ikhefu lentela kanye nezibonelelo emhlabeni we-semiconductor. I-Nvidia isesimweni esihle sokuhlomula ekuthuthukisweni kwe-chip.
I-AI ingahle ikhathaze abasebenzi bekhola emhlophe – qiniseka ukuthi uyahlola Igedlela laleli sonto ukuze uthole okwengeziwe ngalokho – kodwa kungabhekana nokushoda kwabasebenzi abanamakhono endaweni ye-semiconductor.
Ekuqaleni konyaka odlule iCentre for Security and Emerging Technology waxwayisa ukuthi “ukukhishwa kabusha” kokukhiqizwa kwe-semiconductor yase-US kungase kuvinjwe ukuntuleka kwabasebenzi abanamakhono. Ngokwesilinganiso sabo iMelika izodinga ukuqeqesha, ukuqasha, noma ukungenisa abanye abasebenzi abangu-27,000.
Umsunguli we-TSMC uMorris Chang usephindelela waveza ukuthi kubiza kakhulu kangakanani ukwenziwa kwe-semiconductor e-US uma kuqhathaniswa ne-Taiwan.
Ukuze ugcwalise lesi sikhala futhi uqukathe izindleko, ngeke kusimangaze ukubona abenzi bamashiphu benza imizamo emikhulu yokushintsha inqubo yokukhiqiza nokuklama ngokuzenzakalelayo ngokusizakala okukhulu kwe-AI/ML. ®
