Riaan Joyina abaphathi abaphezulu eSan Francisco ngoJulayi 11-12, ukuzwa ukuthi abaholi bahlanganisa kanjani futhi bathuthukisa kanjani ukutshalwa kwezimali kwe-AI ukuze kuphumelele. Funda kabanzi
Engqungqutheleni ye-GPU Technology (GTC) yalo nyaka, iNvidia iqhubekile nokuphusha kwayo kwehardware ye-AI igxile kakhulu ekwenzeni ubuchwepheshe bayo bufinyeleleke kalula amabhizinisi kuzo zonke izimboni kanye nokwenza lula ukuthuthukiswa kwezinhlelo zokusebenza ezikhiqizayo ze-AI njenge-ChatGPT.
Okulandelayo ukuphindaphinda kwansuku zonke kwezimemezelo ezinkulu ezenziwa yinkampani yase-Santa Clara, e-California ngezixhumanisi zokufakwa okujulile.
Qasha ingqalasizinda ye-AI supercomputing nge-DGX Cloud
Ngenkathi uNvidia akha i-Hardware ye-AI isikhathi eside, ubuchwepheshe buthathe isikhathi ukubona ukutholwa kwabantu abaningi – ngokwengxenye ngenxa yezindleko eziphezulu. Emuva ngo-2020, ibhokisi layo leseva ye-DGX A100 lathengiswa ngama- $ 199,000. Ukushintsha lokhu, inkampani namuhla imemezele I-DGX Cloudisevisi ezovumela amabhizinisi ukuthi afinyelele ingqalasizinda yayo ye-AI supercomputing kanye nesofthiwe ngesiphequluli sewebhu.
>>Landela okuqhubekayo kwe-VentureBeat I-Nvidia GTC entwasahlobo ka-2023 ukufakwa <<
I-DGX Cloud iqashisa amabhokisi e-DGX Server, ngalinye line-Nvidia H100 noma i-A100 GPUs eyisishiyagalombili kanye nenkumbulo engu-640GB, futhi libiza u-$36,999 ngenyanga ngenodi eyodwa.
Ngokusebenzisa amandla e-DGX Cloud, inkampani iphinde yamemezela ukwethulwa kwe I-AI Foundations ukusiza amabhizinisi ukuthi adale futhi asebenzise amamodeli enziwa ngokwezifiso e-AI. Umnikelo, uNvidia uthe, uhlinzeka ngezinsizakalo ezintathu zamafu: I-Nvidia NeMo kumamodeli wezilimi ezinkulu (LLMs), Nvidia Picasso yezithombe, ividiyo nezinhlelo zokusebenza ze-3D, kanye I-BioNeMO ukukhiqiza imibhalo yesayensi ngokusekelwe kudatha yezinto eziphilayo.
Izingxenyekazi zekhompuyutha ezintsha zokuqondiswa kwe-AI nezincomo
Eceleni kwe-DGX kanye ne-AI Foundations, uNvidia uphinde wadalula amapulatifomu amane wokucabanga yakhelwe ukusiza onjiniyela bakhe ngokushesha izinhlelo ezikhethekile ezikhiqizayo ze-AI. Lokhu kufaka i-Nvidia L4 yokukhiqiza ividiyo ye-AI; I-Nvidia L40 yokukhiqiza isithombe se-2D/3D; I-Nvidia H100 NVL yokufaka amamodeli ezilimi ezinkulu; kanye no-Nvidia Grace Hopper – oxhumanisa i-Grace CPU ne-Hopper GPU ngaphezu kwesixhumi esibonakalayo se-chip-to-chip esinesivinini esikhulu esingu-900GB/sec – ngezinhlelo zokuncoma ezakhelwe kumadathasethi amakhulu.
Inkampani ithi i-L4 ingaletha ukusebenza kwevidiyo okunamandla angu-120x ngaphezulu kwe-AI kunama-CPU, kuhlanganiswe nokusebenza kahle kwamandla ka-99%; kuyilapho i-L40 isebenza njengenjini ye-Omniverse, iletha ukusebenza okucatshangelwayo okungu-7x kwe-Stable Diffusion nokusebenza okungu-12x Omniverse ngaphezu kwesizukulwane sangaphambilini.
Ama-Chipmaker athola i-cuLitho e-Nvidia GTC
Kulo mcimbi, isikhulu esiphezulu sakwaNvidia, uJenson Huang, singene esiteji ukuze simemezele I-Nvidia kuLitho umtapo wolwazi wesoftware we-computational lithography. Umnikelo, njengoba u-Huang echaza, kuzokwenza amabhizinisi asebenzisa i-semiconductor akwazi ukuklama futhi athuthukise ama-chips anama-transistors ama-ultrasmall nezintambo ngenkathi kusheshisa isikhathi sokumaketha futhi kukhuphule ukusebenza kahle kwamandla kwezikhungo zedatha ezinkulu ezisebenza ngo-24/7 ukushayela inqubo yokukhiqiza i-semiconductor.
“Imboni yama-chip iyisisekelo sayo yonke enye imboni emhlabeni,” kusho uHuang. “Nge-lithography emikhawulweni yefiziksi, ukwethulwa kwe-NVIDIA kwe-cuLitho nokusebenzisana nozakwethu i-TSMC, i-ASML kanye ne-Synopsys kuvumela ama-fabs ukuthi akhuphule izinga lokuphuma, anciphise indawo yawo ye-carbon futhi abeke isisekelo se-2nm nangaphezulu.”
Ekugcineni, inkampani iphinde yamemezela ukusebenzisana ne-Medtronic kanye Microsoft. Owokuqala, uthe, uzoholela ekuthuthukisweni komphakathi Ipulatifomu ye-AI kumishini yezokwelapha echazwe nge-software ekwazi ukwenza ngcono ukunakekelwa kwesiguli. Ngaleso sikhathi, laba bakamuva bazobona umphathi weMicrosoft Azure uNvidia Omniverse kanye neNvidia DGX Cloud.
Umcimbi we-2023 Nvidia GTC uqhubeka kuze kube nguMashi 23.
Umsebenzi we-VentureBeat kufanele kube isikwele sedolobha esidijithali sabenzi bezinqumo zobuchwepheshe ukuze bathole ulwazi mayelana nobuchwepheshe bebhizinisi obushintshayo kanye nokuhwebelana. Thola Okufingqiwe kwethu.
