A NVIDIA akaunza ruzivo nezve AI Factories, kana kuti "AI Factories", nzira yakabudirira yekugadzira zvigadzirwa nemasevhisi anofambiswa ne artificial intelligencel, kushandisa yepamusoro komputa zviwanikwa uye Kugadzira AI. Kuita otomatiki zvese kubva kuunganidzwa kwedata uye kugadzirisa kusvika pakutumirwa kweAI modhi, mafekitori aya anogonesa kukurumidza scalability kuzadzisa zvinodiwa zviri kukura, ichiita basa rakakosha mukuenderera mberi hunyanzvi uye kugadzira mhinduro dzakasarudzika pachiyero. Nzwisisa ipapo! 2c6z38
Chii chinonzi AI Factories 606nb
Aya mafekitori anogadzirisa maitiro kubva mukuunganidza data uye kugadzirisa kusvika pakugadzira uye kutumira kweAI modhi, zvichiita a Rapid scalability kusangana nekukura zvinodiwa. Kuvimba nehukuru hwemakomputa zviwanikwa, zvinowanzo kupihwa ne cloud data centers, AI Factories inogona kuita mabasa akaomarara eAI pamwero, kudzidzisa mamodheru akaomesesa anoda kugadzirwa kwakasimba uye kuchengetedza mavhoriyamu makuru e data.
A IA generative, iyo inogadzira zvitsva zvemukati senge zvinyorwa, mifananidzo, mavhidhiyo uye mimhanzi kubva kune yekupinda data (yekukurumidza), iri pakati pepakati AI Factories. Inobvumira kugadzirwa kwezvigadzirwa senge vabatsiri vechokwadi, masisitimu ekurudziro uye zvemukati zvemukati. Iyo chigadzirwa lifecycle muAI Fekitori inosanganisira kusimudzira, kuyedza, kusimbiswa, kuendesa, kuchengetedza uye kuenderera mberi nekuvandudza, kuve nechokwadi chemhando yepamusoro uye zvigadzirwa zvekuita. Makambani makuru etekinoroji akadai Google e Amazon, uye akawanda ekutanga, kunyanya mu niches senge fintech uye hutano hwehutano, vanoshandisa yavo yega AI Factories.
Mabhenefiti makuru eAI Mafekitori anosanganisira kuenderera mberi hunyanzvi, kuderedzwa kwemitengo yekushanda uye kugona kugadzira mhinduro dzakasarudzika pachiyero. Nekudaro, mafekitori aya anotarisana nematambudziko akakura akadai sekuve nechokwadi chemhando yedata, kutevedzera kuvanzika e tsika, uye gadzirisa yakaoma tekinoroji zvivakwa.
Kuchinja kwemaindasitiri kunotevera kwakatotanga. Makambani nenyika dziri kubatana neNVIDIA kushandura echinyakare matrillion-dollar data centers kuti ikurumidze komputa uye kuvaka mhando nyowani yedata data - mafekitori eAI - kugadzira chinhu chitsva: hungwaru hwekugadzira. Kubva pasevha, networking uye vagadziri vezvivakwa kune vanogadzira software, iyo indasitiri yese iri kugadzirira Blackwell kuti ikurumidze kufambiswa neAI-inotungamirwa hunyanzvi munzvimbo dzese.
Akataura panguva yehurukuro yake kuCOMPUTEX, muvambi uye CEO weNVIDIA, Jensen Huang
NVIDIA MGX modular architecture 2gw1g
Iyo modular architecture NVIDIA MGX (Modular GPU Kuwedzera) ipuratifomu yehardware yakagadziridzwa kuti itsigire yakakura computing mabasa anodiwa muhungwaru hwekugadzira uye nharaunda yekudzidza yemuchina - yakanyatsokodzera kushanda kweAI Factories. Kubva pamamodules anoshandurwa, MGX architecture inobvumira kugadzirisa nyore uye kugadziriswa kwemaitiro, kuita kuti kugadziriswa kwezvikamu zvega pasina kudiwa kwekutsiva hurongwa hwose.
Kunyangwe NVIDIA MGX isiri nyowani, tekinoroji iripo semhinduro iyo mhando inopa kumakambani ari kutsvaga iyi tekinoroji, kana tichitaura nezvezvinodiwa zveAI. Iyi modularity inokurudzira hupfumi uye kuenderera mberi, kubvumira makambani kuti agadzirise maitiro avo eIT sezvinodiwa.
A kugadzirisa Iyo MGX architecture inobvumira makambani kuti agadzirise zvigadziriso zvawo zvinoenderana nezvinodiwa chaizvo, zvingave zvine chekuita nekugadziriswa kwemifananidzo, kuchengetedza, network kana zvimwe zvikamu. Kugadzirisa uku kunoita kuti MGX ikwane kune akasiyana maapplication, anosanganisira AI komputa, kuongororwa kwedata, sainzi simulations, uye girafu kupa. With the kugona kusanganisa akasiyana ma module, makambani anogona kugadzira mhinduro dzakarongedzerwa kusangana neakasiyana mabasa ekuita.
Chimwe chinhu chakakosha cheNVIDIA MGX modular architecture ndeye scalability. Masisitimu anogona kukura zvishoma nezvishoma nekuwedzera mamodule matsva sezvo kudiwa kwezviwanikwa kunowedzera, kuchengetedza huwandu hwepamusoro hwekuita kunyangwe nekuwedzera. Iyi scalability inobvumira makambani kuti ashande akadzika uye akasiyana mabasa ekuita nemazvo, kuve nechokwadi chekuti zvivakwa zvavo zvinogona shanduka pasina kukanganisa kukuru.
Uyezve, iyo MGX dhizaini yakagadzirirwa kuve simba rinoshanda, kubatsira kuderedza kushandiswa kwesimba uye mari yekushandisa. Kushanda kwesimba hakungosimudziri kuchengetedza chete, asiwo kunobatsira kune mamwe maitiro akasimba uye ecological. Kuenderana uye kusanganisa kwakadzama nemamwe matekinoroji eNVIDIA uye mhinduro, senge advanced GPUs uye AI software, inova nechokwadi chekuita kwakagadziridzwa uye nepamusoro mushandisi ruzivo.
Vagadziri vanotanga neiyo yakakosha sisitimu dhizaini yavo server chassis vobva vagadzirisa kusarudzwa kwe GPU, DPU uye U kuti isangane nezvinodiwa chaizvo zvemabasa akasiyana. Kusvikira zvino, anopfuura 90 masisitimu, ichibva kune vanopfuura makumi maviri nevaviri vanobatana, yakatangwa kana iri muchikamu chebudiriro, ichisimudzira iyo MGX referensi yekuvaka. Izvi zvinomiririra kuwedzera kukuru kubva mugore rapfuura, paine masisitimu gumi nemana chete anobva kune vatanhatu vanobatana.
Kushandisa iyo MGX yekuvakisa kunogona kukonzera kudzikiswa kwe kusvika pamatatu (3 / 4) mumitengo nguva yebudiriro uye kudzikiswa kwezvikamu zviviri muzvitatu munguva inodiwa mubudiriro, kupfupisa kutenderera kwekugadzira kusvika kumwedzi mitanhatu chete.
AMD e Intel vari kushandirapamwe paMGX architecture, vachiunza yavo yeU host processor module magadzirirwo kekutanga. Izvi zvinosanganisira chikuva AMD Turin chizvarwa chinotevera uye processor Intel Xeon 6 ine P-cores. Aya mareferensi madhizaini anogona kushandiswa nechero server system mugadziri, kuchengetedza nguva yekusimudzira uye kuve nechokwadi chekuenderana mukugadzira nekuita.
NVIDIA yazvino chikuva, iyo GB200 NVL2, inobatanidza zvivakwa MGX e Blackwell. Inoratidzira scalable, single-node dhizaini, iyo GB200 NVL2 inopa akasiyana masisitimu masisitimu uye netiweki sarudzo, zvichiita kuti iwedzere kukurumidza kusanganisa komputa mune iripo data center zvivakwa. Iyo GB200 NVL2 inoha iyo Blackwell chigadzirwa mutsara, iyo inosanganisira NVIDIA Blackwell Tensor Core GPUs, GB200 Nyasha Blackwell superchips, uye iyo GB200 NVL72. Mutsara uyu unopa mhinduro dzakasimba dzekusangana nekukasira kudiwa kwekombuta mune dzakasiyana siyana data data data.
AI Fekitori Zvishandiso 1kx1q
Munzvimbo dzedata, inofambisa kuvakwa kwezvivakwa zvinokwanisa kubata mavhoriyamu makuru e data uye akasiyana mabasa. Mumaindasitiri ekugadzira, inoshandiswa mukati firimu, dhizaini uye animation masitudiyo emhando yepamusoro graphic rendering. Mundima yekutsvagisa nekusimudzira, inotsigira yakaoma sainzi simulations uye hombe yekuongorora data, kukurumidza kuwanikwa uye zvitsva. Uyezve, muindasitiri otomatiki, inogonesa kudzora kwepamberi uye otomatiki masisitimu mumafekitori akangwara.
jensen huang zvakaratidza kuti makambani anotungamira muTaiwan ari kukurumidza kutora tekinoroji yeBlackwell kubatanidza hungwaru hwekugadzira mumabasa avo. O Chang Gung Chirangaridzo Chipatara, nzvimbo yezvokurapa yakakurumbira muTaiwan, ine zvirongwa zvekubatanidza chikuva chekombiyuta pamwe neBlackwell architecture mukutsvagisa kwavo biomedical. Ichi chirongwa chine chinangwa chekumhanyisa magadzirirwo emifananidzo uye mutauro, kugadzirisa makiriniki maitiro uye, pakupedzisira, kusimudza mwero wekutarisira varwere.
Kune rimwe divi, iyo Foxconn, imwe yehofori dzepasi rose dzemagetsi, iri kunangisa kuedza kwayo kukushandiswa kwehunyanzvi NVIDIA Nyasha Blackwell. Mapurojekiti ayo anosanganisira kugadzira mhinduro dzakangwara dzemotokari dzemagetsi dzinofambiswa neAI uye mapuratifomu erobhoti. Pamusoro pezvo, ivo vari kuwedzera kupa kwavo kwemitauro-yakavakirwa masevhisi, vachivavarira kupa ruzivo rwakanyanya kumunhu kune vatengi.
Uye iwe, wakafungei nezvenhau? Tiudze mashoko!
Onawo:
NVIDIA CUDA-Q inounza quantum computing kune yazvino supercomputer.
Neruzivo kubva: Dell.
Yakaongororwa na Glaucon Vital musi wa2/6/24.
Cool news!