UDell uQinisekisa iServer eNtsha kunye neePlatifomu zokuGcina kwi-AI ePhakamileyo kunye ne-HPC yoMsebenzi woMsebenzi

I-Dell Integrated Rack 7000 (IR7000) ijongana neemfuno ezikhawulezayo zekhompyutha ezinoxinano oluphezulu, ulawulo lwamandla oluzinzileyo kunye nobuchwepheshe bokupholisa obukwinqanaba eliphezulu. Le Projekthi ye-Open Compute (i-OCP) yemigangatho-based rack ilungele ukusasazwa kwizinga elikhulu kwaye ibonakalisa uyilo lwexesha elizayo lwe-multi-generation kunye ne-heterogenous technology environments.

Iimpawu eziphambili ziquka:

Yenzelwe ukuxinana, i-21-intshi i-Dell IR7000 yenzelwe ukuxhasa i-CPU ehamba phambili kwishishini kunye noxinaniso lwe-GPU.

Ilungele ikamva kwaye iyasebenza, i-rack ibonakalisa ububanzi, iileyidi zeseva ezinde ukuhlalisa i-CPU enkulu kunye ne-GPU yoyilo. Le rack yayiyinjongo yakhelwe ukupholisa ulwelo ngokwemveli, ekwazi ukupholisa ukusasazwa kwexesha elizayo ukuya kuthi ga kwi-480KW, kwaye iyakwazi ukubamba phantse i-100% yobushushu obudaliweyo.

Yenzelwe ukhetho olukhulu kunye nokuguquguquka, le rack edibeneyo inika inkxaso kuzo zombini iDell kunye ne-off-the-shelf networking.

Ukusasazwa kulula kwaye ukonga amandlakunye neDell Integrated Rack Scalable Systems (IRSS). I-IRSS inikezela ngeziseko zophuhliso lwe-rack-scale elungiselelwe umthwalo womsebenzi we-AI, okwenza inkqubo yokuseta ingabinamthungo kwaye isebenze kakuhle nge-plug-and-play rack isikali esidityanisiweyo ngokupheleleyo.

Dell PowerEdge

IDell Technologies yazisa amaqonga alungele i-AI alungiselelwe i-Dell IR7000:

Inxalenye yeDell AI Factory eneNVIDIA, iI-Dell PowerEdge XE9712ibonelela ngokusebenza okuphezulu, ukukhawulezisa okuxineneyo kuqeqesho lwe-LLM kunye nokunyanzeliswa kwexesha lokwenyani lokusasazwa kwe-AI enkulu. Yenzelwe uxinaniso lweGPU oluhamba phambili kushishino ngeNVIDIA GB200 NVL72, eli qonga liqhagamshela ukuya kuthi ga kwi-36 ye-NVIDIA Grace CPUs kunye ne-72 NVIDIA Blackwell GPUs kuyilo lwesikali. I-72 GPU NVLink domain isebenza njenge-GPU enye ukuya kuthi ga kwi-30x ekhawulezayo yexesha lokwenyani letriliyoni-parameter LLM inferencing. Ulwelo olupholileyo lwe-NVIDIA GB200 NVL72 lufikelela kwi-25x lusebenza ngakumbi kuneenkqubo eziqhutywa ngumoya ze-NVIDIA H100.

IDell PowerEdge M7725ibonelela ngentsebenzo ephezulu eshinyeneyo icomputing efanelekileyo kuphando, urhulumente, i-fintech kunye nemeko yemfundo ephakamileyo. Yenzelwe ukuba ibekwe kwi-IR7000 rack, iDell PowerEdgeI-M7725 ihambisa i-computing eninzi kwindawo encinci kunye nokuphucuka komgangatho wokusebenza phakathi kwe-24K-27K ii-cores kwi-rack nganye, kunye ne-64 okanye i-72 ye-socket nodes ezimbini, inikwe amandla yi-5th Gen AMD EPYC CPUs Front IO slots yenza isantya esiphezulu se-IO uqhagamshele kwaye ibonelele uqhagamshelwano olungenamthungo kwizicelo ezifunayo. Ifomu yefomu eyonga amandla yomncedisi ivumela ukusasazwa okuzinzileyo ngokupholisa ulwelo ngokuthe ngqo (i-DLC) ukuya kwii-CPU kunye nokupholisa umoya ngokuqhagamshelwa ngokukhawuleza kwirack edibeneyo.

Ukugcinwa okungacwangciswanga kunye noLawulo lweeNkcukacha ezintsha ze-AI Era

IDell Technologies engacwangciswanga yogcino lwedatha entsha iphucula ukusebenza kwesicelo se-AI kunye nokuhambisa ulawulo lwedatha olulula lwehlabathi.

I-Dell PowerScale, indawo yokuqala yokugcina i-Ethernet yehlabathi eqinisekisiweyo ye-NVIDIA DGX SuperPOD, ihambisa uhlaziyo olutsha oluphucula izicwangciso zokulawula idatha, ukuphucula ukusebenza komsebenzi kunye nokubonelela ngenkxaso enkulu yomthwalo we-AI.

Ukufumaneka okuphuculweyo:Vula ulwazi lwedatha ukuze wenze izigqibo ezikhawulezayo ezikrelekrele usebenzisa i-PowerScale metadata kunye neDell Data Lakehouse. Umlayishi ozayo we-Dell ovulekileyo woxwebhu lweenkonzo ze-NVIDIA NeMo kunye nezicwangciso ze-RAG zenzelwe ukunceda abathengi baphucule ixesha lokungeniswa kwedatha kunye nokunciphisa i-computing kunye neendleko ze-GPU.

Ugcino oluxineneyo:Abathengi banokuhlengahlengisa iimodeli zabo ze-AI ngokubaqeqesha kwiiseti zedatha ezinkulu ezine-61TB ezintsha zokuqhuba ezonyusa umthamo kunye nokusebenza kakuhle ngelixa zinciphisa indawo yokugcina idatha ngesiqingatha.

Ukusebenza okuphuculweyo kwe-AI:Ukusebenza komsebenzi we-AI kuphuculwe ngekhono langaphambili le-NVIDIA InfiniBand kunye ne-200GbE Ethernet inkxaso ye-adapter ehambisa ukuya kwi-63% ngokukhawuleza.

Ngophuculo olutsha kwi-platform yolawulo lwedatha yeDell Data Lakehouse, abathengi banokugcina ixesha kunye nokuphucula imisebenzi kunye neempawu ezintsha ezifana nokubuyisela intlekele, ukufunyanwa kwe-schema ngokuzenzekelayo, ii-APIs zolawulo olubanzi, kunye nokuphuculwa kwe-stack epheleleyo.

Abathengi banokwenza lula uhambo lwabo oluqhutywa yidatha kwaye balinganise ngokukhawuleza i-AI yabo kunye namatyala okusetyenziswa kweshishini kunye neeNkonzo zokuLungisa iiNkcukacha zeDatha kunye neeNkonzo zokuPhunyezwa kwiiMibhobho zeDatha. Ezi nkonzo zandisa ukufikeleleka kwedatha ekumgangatho ophezulu ngokufunyanwa, umbutho, ukuzenzekelayo kunye nokudibanisa.


Ixesha lokuposa: Nov-02-2024