NVIDIA DGX Spark Super AI Kombiyuta

Tsananguro huru yechigadzirwa

NVIDIA DGX Spark ikombiyuta yedhijitari yedhijitari yeAI inogona kuiswa zvakananga padhesiki. Chitarisiko chayo chakakura seMac mini (150 mm × 150 mm × 50.5 mm, 1.2 kg), asi inobatanidza zvese zvakakosha zveruzivo rwechangobva kuitika rweNVIDIA rweGrace Blackwell, zvichiunza ruzivo rwe "kutamisa nzvimbo yedata kudzokera kudesktop" kune vanogadzira, vaongorori nesainzi yedata.

Taura nesu

ruzivo rwechigadzirwa

Basa rechigadzirwa uye zvakanakira

Basa uye zvakanakira zvechigadzirwa cheNVIDIA DGX Spark zvinogona kupfupikiswa mumutsara mumwe chete: chinosanganisa "data center level AI computing power" mubhokisi diki risinganyarari padesktop, zvichibvumira vanhu nemapoka madiki kugadzira, kugadzirisa, uye kuisa mamodheru makuru senge kuva ne supercomputer. ） 1、 Basa rechigadzirwa (chii chingakubatsira nacho) Kugadzira uye Kufunga Kwemodheru Yakakura Yemunharaunda Muchina mumwe chete unogona kumhanyisa ma prototypes emhando yemitauro mikuru ye200 bhiriyoni, kugadzirisa zvakanaka, uye kufungidzira pasina kuvimba nema cloud GPU. Mayuniti maviri anogona kufungidzira mamodheru makuru akadai seLlama-3.1-405B kuburikidza neNVLink-C2C interconnection. Nzvimbo yakachengeteka uye inoenderana neyakavanzika yeAI. Muzviitiko apo data rakakosha senge zvipatara, mari, nehurumende risingakwanise kuiswa ku cloud, tsvakiridzo yakazara yeAI nekuvandudzwa kunogona kupedzwa munharaunda. Kudzidzisa, Kutsvagisa, uye Kuedza Kwemunhu Makosi eYunivhesiti, makwikwi eKaggle, kukopa mapepa, plug and play, kuchengetedza mari yekutenga nekuchengetedza sevha. PoC (Proof of Concept) Accelerator Mapoka madiki emabhizinesi anokurumidza kusimbisa kushanda kwemaalgorithms munharaunda, uye mushure mekupasa kusimbiswa, anoaendesa kuDGX Cloud kana kunzvimbo dzedata kuti adzidziswe zvakanyanya, zvichiita kuti pave nekuchinja kwekodhi kubva padesktop kuenda kucloud. Edge AI neRobot Development Shandisa mamodheru eAI echizvarwa chitsva akadai seNVIDIA GR00T N1 robot base model uye Cosmos Reason world model, achitungamira zvakananga marobhoti prototypes kana zvishandiso zvemucheto.

2、 Zvakanakira zvikuru (nei zvichikosha kusarudza) Zvakanakira divi rimwe chete Kushanda kwekupedzisira kwe1000 TOPS (FP4)/1 PFLOPS AI computing power, kusvika padanho remaseva eGPU echizvarwa chekare. Vhoriyamu diki kwazvo ye150 × 150 × 50.5 mm, 1.2 kg, inogona kutakurwa neruoko rumwe, socket power supply ye170 W, hapana chikonzero chekutonhodza mukamuri rekombuta. Ultra large united memory 128 GB LPDDR5X united addressing, CPU/GPU sharing, kudzivirira kukopa data kakawanda uye kuvandudza kushanda kwemamodheru makuru. Yakagadzirira kushandiswa Ubuntu 22.04 DGX OS, PyTorch/TensorFlow, CUDA-X, NVIDIA NIM, nedzimwe AI stacks dzakakwana, dzine madhiraivha akagadziridzwa, macontainer, uye maraibhurari. Kuwedzerwa kweConnectX-7 200 GbE+NVLink-C2C kusina musono, inokwanisa kuyera yakatwasuka mumapoka akawanda emuchina; Kuchinja kamwe chete kwekodhi imwe chete kuenda kuDGX Cloud pasina chikonzero chekugadzirisa. Kutonhodza kusingadzivisiki kunochengetedza simba + feni yakanyarara, ruzha

Nzvimbo dzekushandisa dzechigadzirwa

Kunyangwe NVIDIA DGX Spark iri kombiyuta huru yeAI yemhando yedesktop, ine simba rekuverenga re 1 PFLOPS, 128 GB ye memory yakabatana, uye 170 W yekushandisa simba, inogona kutoshandisa mamodheru makuru ega ane ma parameters ari pakati pe7-200 bhiriyoni munharaunda. Saka, yakakodzera zvakanyanya kudzika ma workflows makuru eAI anogona kungoiswa mugore kana pakati pedata ku "desktop/on-site/edge".
Pazasi pane nzvimbo dzakajairika dzekushandisa dzinogona kushandiswa zvakananga zvichienderana nemamiriro ezvinhu eindasitiri (★ inomiririra mukana wekuti "muchina mumwe chete unogona kushandisa mamodheru makuru" ayo akakodzera DGX Spark).
Magetsi neZvishandiso zveVeruzhinji
★ Kuronga nguva chaiyo kwegridi yakangwara
Zvichibva pamamodeli makuru ekufanotaura mutoro uye kufanotaura simba idzva, muchina mumwe chete unogona kushandisa network yekufanotaura nguva ye70B parameter ine 15 minute rolling prediction, ichitsiva dhizaini yechinyakare yetier two-tier ye cloud training edge inference.
Kuiswa kwemagetsi panzvimbo iyi kuri muimba yebasa renzvimbo yekutumira magetsi, uye data harigone kutumirwa kubva muintranet kuti zvienderane nekuchengetedzeka uye kutevedzera mitemo yegridi yemagetsi.
★ Kugadzirisa mafambiro emidziyo yekufambisa nguva
Uchishandisa modhi yakakura inoonekwa kuona zvikanganiso zviri muchiedza chinoonekwa uye mifananidzo ye infrared yemotokari/marobhoti asina vanhu panguva yekuongorora, zvinokwanisika kuita fungidziro pamifananidzo 2000 ye4K pasekondi imwe chete munharaunda, zvichiita kuti "ibhururuke uye itonge panguva imwe chete".
Ita ongororo ye multimodal fusion ye vibration partial discharge oil chromatography pamidziyo ine voltage yakawanda yakadai se transformers ne GIS, uye yambirai zvikanganiso mwedzi 3-6 isati yasvika.
★ Dhijitari yemagetsi maviri uye chibooreso chekukurumidzira
Shandisa 1000 node level power grid simulation + reinforcement learning model kuti ugadzire nzira dzekufambisa mumaminitsi; Unogona kukurumidza kuongorora mamiriro ekunze akaipisisa mukamuri rebasa rekutumira kana mota yekuraira emergency.
Kuchengetera simba idzva nemagetsi
★ Kufanotaura kwenguva pfupi kwemhepo nesimba rezuva
Kushanda kwemuchina mumwe chete kwe34 B meteorological power joint large model, kubatanidza radar, mifananidzo yesatellite cloud, uye data re wind lidar, kunogadziriswa maminetsi gumi nemashanu ega ega, zvichideredza zvikanganiso ne15-20%.
★ Kugadziriswa kwehurongwa hwekuchengetedza simba reEMS munharaunda
Shandisa modhi yekuronga zvinhu ine zvinangwa zvakawanda (kuwedzera mari inowanikwa, kuwedzera hupenyu hwebhatiri, uye kudzora kudiwa kwemagetsi) munzvimbo yako mukamuri rekudzora nzvimbo yekuchengetera magetsi, uye nzira dzekuchaja nekuburitsa magetsi mumasekonzi mashoma pasina chikonzero chekuadzosera kugore.
mafuta negesi
★ Dudziro yekudengenyeka kwenyika uye kufanotaura nezvemadhizeti
Nzira yekugadzira AI yeLocal 70 B parameter inoshandiswa kutsanangura otomatiki zvikanganiso uye mitumbi yejecha rerwizi zvichibva padata re pre stack/post stack seismic, zvichigadzira huwandu hwe sweet spot probability hwemativi matatu. Vashandi vekududzira panzvimbo vanogona kutaurirana zvakananga nekuzvigadzirisa.
★ Kufanotaura njodzi dzekuboora nekupedzisa
Kushandisa LSTM+Transformer hybrid model pakuongorora data repaburi panguva yekuboorwa (MWD/LWD), yambiro yekuboorwa kwetsime nekudzikiswa kwemvura inogona kuitika maminetsi mashanu kusvika gumi pachine nguva, uye muchina mumwe chete unogona kusangana ne20 Hz level real-time inference.
Simba renyukireya nesimba rekupisa
★ Kutarisa AI yezinga rekuchengetedza
Shandisa modhi ye30 B parameter visual acoustic fusion mu network yemukati mekamuri yekudzora chiteshi chesimba renyukireya, ongorora data reCCTV vhidhiyo uye acoustic sensor munguva chaiyo, uye tsvaga kubuda kwemvura uye ruzha rusina kujairika rwemidziyo, zvichizadzisa zvinodiwa zvekuzviparadzanisa kwesimba renyukireya.
★ Kugadziriswa kwekupisa muzviteshi zvemagetsi zvinopisa
Nzira yekugadzirisa kupisa yakavakirwa pakudzidza kwakadzika kwekusimbisa, kubatanidza data reDCS renguva chaiyo (tembiricha, kumanikidzwa, huwandu hweokisijeni), kudzora kwemuchina mumwe chete kwe100 ms closed-loop, kuderedza kushandiswa kwemarasha ne1-2%. Simba rakabatanidzwa/microgrid/nzvimbo yemaindasitiri
★ Brain yedanho rekambasi simba router
Isa DGX Spark mukamuri rekudzora microgrid, mhanyisa modhi ye50B parameter multi energy flow optimization, batanidza photovoltaic, kuchengetedza simba, kuchaja mapombi, uye kutonhora kwakabatana, kudziyisa, uye kupa simba munguva chaiyo, uye uwane "source grid load storage" kuzvitonga kwechipiri.
★ Kuverenga uye kutengeserana kwemhepo inoburitsa kabhoni online
Modhi yakakura yemuno inoverenga marogi ehurongwa akadai seEMS, MES, ERP, nezvimwewo, inogadzira mishumo yekupisa kabhoni yemaminitsi mashoma, uye inobatanidza zvakananga neAPI yekuchinjana kabhoni yedunhu pasina data rinobuda mupaki. Pfupiso mumutsara mumwe chete:
DGX Spark yakaunza zvechokwadi "mamodheru makuru anogona kungoiswa mugore/mukamuri remakombiyuta" kumitsetse yekugadzira yakadai semakamuri emagetsi anotumira magetsi, makamuri ebasa rezviteshi zvemagetsi, nzvimbo dzekufuridzira mapurazi emhepo, makamuri ekudzora nzvimbo dzekuchengetera magetsi, mahofisi epuratifomu yekuboorera, nezvimwewo kekutanga, ichizadzisa "data risingabude mumunda, mamodheru asiri mugore", ichideredza zvikuru njodzi dzekutevedzera mitemo yeAI muindasitiri yesimba nemagetsi.

Gwaro reKushanda kweChigadzirwa

Gwaro reKushanda kweChigadzirwa cheNVIDIA DGX Spark
(Inoshanda kuRetail/Founders Edition yakatangwa kubva muna Chikunguru 2025 zvichienda mberi) Kuburitsa zvinhu uye kutanga kuvhura
Rondedzero yekurongedza: Host, 170W Gallium Nitride Power Adapter (kusanganisira multi country plug), 1m USB-C kusvika C cable, 1m 10 GbE Ethernet cable, Quick Start Card, Warranty Card.
Matanho ekubatanidza
① Isa adapta yemagetsi muchiteshi cheUSB-C PD 3.1 kuruboshwe rwemuviri;
② Batanidza HDMI 2.1 kune monitor kana kushandisa USB-C kuenda kuDP/HDMI cable kuti ubatanidze skrini dzakawanda;
③ Network ine waya: RJ-45 yakabatana ne10 GbE; Network isina waya: Mushure mekubatidza, sarudza Wi Fi 7 zvichienderana nemirairo;
④ Dzvanya bhatani repamusoro rekudzima magetsi kwesekondi imwe chete, uye chiedza chechiratidzo chinoshanduka kubva pachena kuenda pabhuruu kuratidza kutanga. Kutanga kwekutanga kunotora masekondi angangoita makumi mapfumbamwe kuti kupinde pakombiyuta yeDGX OS.
Kutanga kwesystem uye account
Sisitimu yekutanga: Ubuntu 22.04 LTS+NVIDIA DGX OS 5.2 (kernel 6.8+CUDA 12.6). Murongi wekutanga wekutangisa
① Mutauro → ② Nguva Yenzvimbo → ③ Gadzira account yemutariri wemunharaunda (ine kodzero dzekushandisa sudo) → ④ Sarudza kana uchida kunyoresa kuNVIDIA AI Enterprise (kuyedzwa kwemazuva makumi mapfumbamwe).
Remote SSH: Sisitimu iyi ine opensh server yakatoiswa; Ita mushure me network yekutanga
sudo systemctl inogonesa --ikozvino ssh
chiteshi chekare

Kugadziriswa kwechigadzirwa mushure mekutengeswa nekugadzirisa

NVIDIA DGX Spark Super AI Kombiyuta
Pfupiso yemasevhisi ekugadzirisa mushure mekutengesa nekugadzirisa
(Inoshanda kune vhezheni yekutengesa yakatumirwa zviri pamutemo muChinese Mainland kubva muna Chikunguru 2025)
Mutemo wewarandi wepamutemo
|Nguva yewarandi | Waranti yemakore matatu yemuchina wese (kusanganisira host, power adapter, uye tambo dzinoperekedza) IT Home|
|Kutanga kweWaranti | Zvichibva pazuva reinvoice kana zuva refekitori pamwe nemazuva makumi matatu, chero ripi zvaro riripo kare|
|Kuvharwa kwewarandi | Kukundikana kwehardware, zvikanganiso zvepakutanga zvefirmware, kukuvara kwesoftware yakamboiswa|
|Chiyero chisiri chewarandi | Kukuvadzwa kwevanhu, kupinda kwemvura, kurohwa nemheni, kuzviparadzanisa, kupenya kwefirmware isina mvumo

tsananguro2