Chii chaitika
TechCrunch inoshuma kuti Nvidia's yemakwikwi mukana muAI zvivakwa zviri kuchinja kupfuura ayo magiraidhi processors. Sezvo AI data nzvimbo inoyera yakananga kugigawatt-level mashandisiro emagetsi, kambani iri kutengesa akasanganiswa masisitimu anoitirwa kurongedza ndangariro, kuchengetedza, networking uye compute yakatenderedza GPU. Chirevo ichi chinotarisa pane Nvidia's Vera Rubin architecture, iyo inosanganisa iyo Rubin GPU neVera CPU, Groq 3 LPX accelerators uye yakabatana kuchengetedza uye networking racks. Nvidia mukuru Jason Hardy akaudza TechCrunch kuti Vera CPU yakagadzirirwa kubatsira kutungamira data kuGPU panguva chaiyo. Hardy akati Nvidia yakaona kusvika pakuvandudzika kwakapetwa katatu mune mamwe mashandiro, kunyange TechCrunch isina kuzvimiririra kuzvimiririra. TechCrunch inofananidzawo maitiro eNvidia neOpenAI's Jalapeño chip, iyo yakagadzirirwa kuderedza kufamba kwedata nekuchengeta basa rese mukati meimwe yakabatana system. Maitiro acho anosiyana, asi ese ari maviri anogadzirisa dambudziko rimwechete rezvivakwa: kufambisa data nemazvo kunogona kuita sekunge kuwedzera processor kugona.
TechCrunch inoshuma kuti Nvidia ichangoburwa AI mukana uri kuwedzera kusungirirwa kune zvivakwa zvakatenderedza GPU. Iyo yekubuda inoti AI komputa iri kukura yakananga kugigawatt chiyero, zvichiita kuti zvinyanye kunetsa kushandisa nzvimbo huru dzedata nemazvo. Munzvimbo iyoyo, chinyorwa chinopokana, kurongedza kufamba kwedata uye zvikamu zvehurongwa zvinova dambudziko reinjiniya pane chechipiri chinonetsa.
Chirevo chinozivisa Nvidia's Vera Rubin dhizaini semuenzaniso weiyi zano. Sekureva kweTechCrunch, dhizaini inobatanidza iyo Rubin GPU neVera CPU, Groq 3 LPX accelerators uye masheji anofananidzwa ekuchengetedza uye network. Iyo yekuburitsa inotsanangura zvikamu izvi seakasarudzika masisitimu anoitirwa kuvandudza mashandiro ezvese zvakatenderedza GPU, pane kungogadzirisa tokeni ivo pachavo.
Jason Hardy, Nvidia's mutevedzeri wemutungamiri wetekinoroji yekuchengetedza, akaudza TechCrunch kuti Vera CPU inogadzirisa ndangariro shoma yesavha imwechete kana compute chikuva uye inobatsira kuronga data. TechCrunch inoshuma Hardy achiti Nvidia yakacherekedza kusvika pakuvandudzwa katatu mune mamwe mashandiro uye inogona kushandisa flash kuchengetedza zvakanyanya pasina kugadzira bhodhoro. Icho chirevo chekuita chinobva kuNvidia uye haina kusimbiswa yakazvimirira mune sosi.
TechCrunch inosiyanisa Nvidia's system-level maitiro neOpenAI's Jalapeño chip. Iyo yekubuda inonokora OpenAI ichiti Jalapeño yakagadzirirwa kuderedza kufamba kwedata uye kunonoka kutaurirana nekuchengeta basa rese mukati meimwe yakabatana system. Nvidia uye OpenAI vari kushandisa dhizaini dzakasiyana, asi TechCrunch inoti zvese zviri zviviri zvinotsvaga kugona kuburikidza nehungwaru kutonga kwekufamba kwedata kwete chete kuburikidza nekuwedzera mapurosesa.
Kwakabva mashoko: techcrunch.com ↗
Nei zvichikosha
Chirevo chinotsanangura shanduko mune iyo AI zvivakwa zvingaongororwa. Kuita kweGPU kunoramba kwakakosha, asi kuburitsa kunoshanda kweiyo hombe AI system inogona zvakare kutsamira pakuwana ndangariro, kuchengetedza, networking uye software uye hardware inoshandiswa kurongedza izvo zvikamu. Izvi zvinogona kuita kuti zviomere kune vanokwikwidza kupikisa Nvidia nekungopa imwe yekumhanyisa. Mukwikwidzi angada kuenderana nekuita kwehurongwa hwakakura, kusanganisira kubatana pakati pemachipisi uye masisitimu anochengeta anodhura processor akapihwa nedata. Zvazvinoreva zvinoramba zvisina chokwadi. TechCrunch's account yakavakirwa pane imwe tsananguro yeNvidia yemasisitimu ayo uye nekuita kwekuita kwemukuru. Iyo sosi haipe yakazvimiririra mabhenji mabhenji, mitengo, nhamba dzekutumira kana humbowo hwekuti Nvidia's full-system mukana unoenderera mberi sezvo hyperscalers vachigadzira yavo zvivakwa.
Chirevo chine basa nekuti chinokudza tsananguro yeAI komputa kugona. A accelerator ine simba inogona kushandiswa pasi kana data, modhi paramita kana mhedzisiro yepakati ikasasvika pairi nekukurumidza zvakakwana. Chinhu chepakati chechinyorwa ndechekuti dhizaini dhizaini inogona kuona kuti zvinodhura sei maAI processors anoshandiswa.
Izvi zvinogadzira chipingamupinyi chinobvira kune vanopikisa. TechCrunch inoshuma kuti ma hyperscalers akadai seAmazon neGoogle vakagadzira machipisi avo, vachidzikisa chimiro cheNvidia seye ega mupi wepamberi AI GPUs. Asi imwe GPU kana accelerator inogona kunge isina kukwana kana ichishaya ndangariro dzakafanana, kuchengetedza, networking uye orchestration kugona.
Kuchinja uku kunogona kukanganisawo kushandiswa kwezvivakwa. Kana kufamba kwedata uye kurongeka kuri zvipingaidzo zvakakura, vatengi vangangoda kuongorora racks dzakakwana uye masisitimu ekuvaka pane kuenzanisa accelerator yakatarwa chete. Izvo zvinogona kusimbisa chinzvimbo chevatengesi vanogona kupa yakasimba yakabatanidzwa hardware uye software, nepo zvichigona kuwedzera kuoma uye mutengo wekuchinja vatengesi.
Uchapupu hune miganhu inokosha. Chirevo cheTechCrunch chinosanganisira nhaurirano nevashandi veNvidia uye chimiro chekushanda chakapihwa nemukuru weNvidia, asi sosi yacho haipe yakazvimiririra kuyedzwa, data rekutumira vatengi, mitengo, zviyero zvemagetsi kana kuenzanisa neanokwikwidza masisitimu. Chirevo chinotsigira kuvepo kweNvidia's yakakura system system, asi hazviratidze kuti kambani ichachengetedza kutungamira kwakasimba.
Interactive Mechanism: Iyo Inonyatsoshanda
Ongorora ari pasi tekinoroji kuseri kwekusimudzira uku uchipindirana.
Which component of an AI application is the machine-learning model itself?
Zvekutarisa zvinotevera
Tarisa uone kana Nvidia ichidhinda yakazvimirira, zviyero zvinogoneka zveVera Rubin system, kusanganisira kupera-kusvika-kumagumo, kushandiswa kwesimba, ndangariro uye kuchengetedza kuita, uye mamiriro ari kuseri kweyakashumwa kuvandudzwa katatu. Tarisa kuti Amazon, Google uye mamwe mahombe anoshanda makore anopindura sei. TechCrunch inoshuma kuti hyperscalers vanga vachigadzira yavo machipisi, asi iwo anonyanya kukosha makwikwi anogona kusanganisira akazara masisitimu ekuronga komputa, ndangariro, kuchengetedza uye network. Tarisa uone kana vatengi vanotenga Nvidia's yakasanganiswa zvivakwa sepakeji kana zvinotsiva zvinhu kubva kune akawanda vatengesi. Chirevo chinoratidza kuti orchestration iri kuita chikamu chikuru chemakwikwi, asi hairatidze kuti masisitimu eNvidia anowanikwa sei, anodhura, kana maitiro aanoita mukugadzira.
Humbowo hunotevera hunobatsira hungave hwakajeka, wechitatu-bato kuyedzwa kweVera Rubin masisitimu. Kuyedzwa kwakadai kunofanirwa kuparadzanisa kuita kweGPU kubva kune zvakawanikwa kuVera CPU, flash kuchengetedza, networking uye orchestration, uye inofanirwa kutsanangura mitoro yebasa uye yekutanga inoshandiswa kune chero yainzi kuvandudzwa.
Kuvepo uye mazwi ekutenga ane basa zvakare. Sosi yacho inotsanangura magadzirirwo eNvidia uye kuburitswa kwazvino asi haitaure kuti ndevapi vatengi vanogona kuwana masisitimu, muhuwandu hupi, nemutengo upi, kana kuti zvinhu zvega zvinogona kusanganiswa nehardware kubva kune vamwe vatengesi. Iwo madhiri anozoona kana zano racho richishanda zvakanyanya kana kunyanya rakasanganiswa Nvidia inopa.
Mhinduro dzevakwikwidzi dzicharatidza kana musika uri kuenda kumakwikwi e-system-level. Vanopa Cloud vanogona kugadzira mamwe mavakirwo akatenderedza machipisi avo, nepo vamwe vanogadzira chip vanogona kutarisa ndangariro, networking, kuchengetedza kana kubatanidza pane kukwikwidza zvakananga paGPU kuita. Iyo sosi hairatidze chaiwo masisitimu emakwikwi kana kupa humbowo nezve kugona kwavo kwazvino.
OpenAI's Jalapeño dhizaini kuenzanisa kunobatsira, asi haifanirwe kubatwa sehumbowo hwekuti imwe nzira iri nani. Chirevo cheOpenAI, chakatorwa neTechCrunch, chinotsanangura kuedza kudzikisa kufamba kwedata mukati meiyo chip yakabatana. Rumwe ruzivo runodiwa kuenzanisa dhizaini neiyo Nvidia's yakagoverwa orchestration modhi pamabasa chaiwo, kushandiswa kwesimba, kuvimbika uye mutengo wekushandisa.