I-Dell Integrated Rack 7000 (IR7000) isingatha izimfuno zekhompuyutha ezisheshisiwe ezinobuningi obuphakeme, ukuphathwa kwamandla okuzinzile kanye nobuchwepheshe bokupholisa obuthuthukile. Le rakhi esekwe kumazinga ye-Open Compute Project (OCP) ilungele ukuthunyelwa kwenani elikhulu futhi ifaka idizayini engenasikhathi esizayo yezindawo zobuchwepheshe zesizukulwane esiningi kanye ne-heterogenous.
Izici ezibalulekile zifaka:
Idizayinelwe ukuminyana, i-21-inch Dell IR7000 yakhelwe ukusekela i-CPU ehamba phambili embonini kanye nokuminyana kwe-GPU.
Ikusasa-ngomumo futhi ephumelelayo, indawo yokubeka ifaka izihlibhi zeseva ezibanzi, ezinde ukuze kuhlangatshezwane nezakhiwo zakamuva, ezinkulu ze-CPU ne-GPU. Le rack yayakhelwe ukupholisa uketshezi ngokomdabu, ekwazi ukupholisa ukuthunyelwa kwesikhathi esizayo okungafika ku-480KW, futhi iyakwazi ukuthwebula cishe u-100% wokushisa okudaliwe.
Iklanyelwe ukukhetha okukhulu nokuvumelana nezimo, le rack ehlanganisiwe inikeza ukusekelwa kokubili kwe-Dell kanye ne-off-the-shelf networking.
Ukuthunyelwa kulula futhi konga amandlangeDell Integrated Rack Scalable Systems (IRSS). I-IRSS iletha ingqalasizinda ye-rack-scale ethuthukisiwe yomthwalo we-AI, okwenza inqubo yokusetha ingabi namthungo futhi isebenze kahle ngohlelo oludidiyelwe ngokugcwele lwe-plug-and-play rack scale.
I-Dell Technologies yethula amapulatifomu alungele i-AI aklanyelwe i-Dell IR7000:
Ingxenye ye-Dell AI Factory ene-NVIDIA, iI-Dell PowerEdge XE9712ihlinzeka ngokusebenza okuphezulu, ukusheshisa okuminyene kokuqeqeshwa kwe-LLM kanye nokwehliswa kwesikhathi sangempela kokuthunyelwa kwe-AI ngezinga elikhulu. Idizayinelwe ukuminyana kwe-GPU ehamba phambili embonini nge-NVIDIA GB200 NVL72, le nkundla ixhuma ama-NVIDIA Grace CPUs angafika kwangu-36 nama-72 NVIDIA Blackwell GPUs edizayini yesikali. Isizinda se-72 GPU NVLink sisebenza njenge-GPU eyodwa kufika ku-30x ngokushesha okukhulu kwe-trillion-parameter LLM inferencing. Uketshezi olupholile lwe-NVIDIA GB200 NVL72 lusebenza kahle ngokungu-25x kunamasistimu anamandla e-NVIDIA H100 aphethwe umoya.
II-Dell PowerEdge M7725inikeza ukusebenza okuphezulu kwekhompiyutha eminyene elungele ucwaningo, uhulumeni, i-fintech kanye nezindawo zemfundo ephakeme. Idizayinelwe ukuthi isetshenziswe ku-IR7000 rack, iI-Dell PowerEdgeI-M7725 iletha ikhompuyutha eyengeziwe esikhaleni esincane ngokukalwa okuthuthukisiwe kokusebenza phakathi kwama-cores angama-24K-27K rack ngayinye, enama-socket node amabili angama-64 noma angama-72, anikwa amandla yi-5th Gen AMD EPYC CPUs Front IO Slots inika amandla ukuxhumana kwe-IO okunesivinini esikhulu futhi ihlinzeka ngoxhumano olungenamthungo lwezinhlelo zokusebenza ezifunayo. Isici sefomu esonga amandla seseva sivumela ukusetshenziswa okuzinzile kukho kokubili ukupholisa uketshezi oluqondile (i-DLC) kuma-CPU kanye nokupholisa komoya ngokuxhuma ngokushesha endaweni yokubeka ehlanganisiwe.
Isitoreji Esingahlelekile Nokuphathwa Kwedatha Yesikhathi se-AI
Iphothifoliyo ye-Dell Technologies engahlelekile yokugcina idatha ithuthukisa ukusebenza kohlelo lwe-AI futhi ilethe ukuphathwa kwedatha yomhlaba wonke okwenziwe lula.
I-Dell PowerScale, isitoreji sokuqala se-Ethernet emhlabeni esigunyazwe i-NVIDIA DGX SuperPOD, iletha izibuyekezo ezintsha ezithuthukisa amasu okuphatha idatha, ezithuthukisa ukusebenza komsebenzi futhi zinikeze ukwesekwa okukhulu kwemithwalo yemisebenzi ye-AI.
Ukutholakala okuthuthukisiwe:Vula imininingwane yedatha ukuze wenze izinqumo ezihlakaniphile ngokushesha usebenzisa imethadatha ye-PowerScale kanye ne-Dell Data Lakehouse. Isilayishi sedokhumenti yomthombo ovulekile ye-Dell ezayo samasevisi e-NVIDIA NeMo kanye nezinhlaka ze-RAG yakhelwe ukusiza amakhasimende athuthukise isikhathi sokungenisa idatha futhi ehlise izindleko zekhompuyutha kanye ne-GPU.
Isitoreji esiminyene:Amakhasimende angakwazi ukushuna kahle amamodeli awo e-AI ngokuwaqeqesha kumadathasethi amakhudlwana anamadrayivu amasha angu-61TB akhuphula umthamo nokusebenza kahle kuyilapho enciphisa indawo yokugcina isikhungo sedatha ngesigamu.
Ukusebenza kwe-AI okuthuthukisiwe:Ukusebenza komthwalo womsebenzi we-AI kuthuthukiswa ngamakhono e-NVIDIA InfiniBand angaphambili kanye nosekelo lwe-adaptha ye-Ethernet engu-200GbE iletha okokusebenza okusheshayo okungafika ku-63%.
Ngezithuthukisi ezintsha zeplathifomu yokuphatha idatha ye-Dell Data Lakehouse, amakhasimende angonga isikhathi futhi athuthukise ukusebenza ngezici ezintsha ezifana nokutholwa kwenhlekelele, ukutholwa kwe-schema okuzenzakalelayo, ama-API okuphatha aphelele, kanye nokuthuthukiswa kwezitaki ezigcwele zokuzisiza.
Amakhasimende angakwazi ukwenza uhambo lwawo olushayelwa idatha lube lula futhi akale ngokushesha i-AI yawo kanye namacala okusetshenziswa kwebhizinisi Ngezinsizakalo Zokuthuthukisa Zekhathalogu Yedatha kanye Nezinsizakalo Zokuzisebenzisa Zamapayipi Edatha. Lezi zinsizakalo zikhulisa ukufinyeleleka kudatha yekhwalithi ephezulu ngokutholwa, ukuhlela, ukuzenzela kanye nokuhlanganiswa.
Isikhathi sokuthumela: Nov-02-2024
