I-GIGABYTE AI TOP ATOM vs NVIDIA DGX Spark: I-Desktop AI amandla ekhanda ukuya ekhanda

I-duel phakathi kwe-desktop mini-supercomputer iyathakazelisa: I-GIGABYTE AI TOP ATOM vs NVIDIA DGX Spark. Bobabili babelana ngenhliziyo yenkathi entsha yekhompyutha IA kusuka ku-NVIDIA, kodwa beza "ngamaphakheji" ahlukene: eyodwa njengenkundla yereferensi futhi enye njengomkhiqizo we-turnkey olungele ukubeka ukusebenza etafuleni. Uma ufuna ukusebenza kwe-AI okungathi sína ngaphandle kokusetha indawo yokubeka, kuningi okumele kuhlaziywe lapha.

Sikhuluma ngamakhompiyutha anesisindo selitha elilodwa nje elithembisa izibalo zokusebenza kuze kube muva nje abengacabangeki ku-PC encane kangaka: 1 petaFLOP yokusebenza kwe-AI ku-FP4 (1.000 TOPS), inkumbulo ehlanganisiwe enesivinini esikhulu, nayo yonke i-ecosystem yesofthiwe ye-NVIDIA ukuze ikhiqize, icule kahle, futhi iveze amamodeli esizukulwane esilandelayo. Futhi yebo, konke lokhu ngokusetshenziswa kwamandla okuphansi nokuxhumeka kwenethiwekhi okukala lapho kudingeka.

Yini ngayinye: Inkundla ye-DGX Spark kanye ne-PC encane ye-GIGABYTE

Ukuzimisa: I-NVIDIA DGX Spark iyinkundla yereferensi yedeskithophu ye-NVIDIA (phambilini eyayaziwa ngokuthi i-Project DIGITS) futhi izalwa ngokubambisana nayo I-MediaTek"Yisikhunta" lapho abakhiqizi abahlukene bangakha khona amadivaysi abo e-AI ahlangene. Ngaphakathi kwalolo hlaka, amabhrendi afana ne-MSI ne-ASUS asevele abonisa iminikelo yawo, futhi i-GIGABYTE ijoyina nge-twist yayo.

Lokho twist kuyinto I-GIGABYTE AI TOP ATOM, i-PC encane eyi-1 litre ethatha isisekelo sobuchwepheshe se-DGX Spark futhi iyipakishe ibe umkhiqizo wokugcina onemininingwane ecacile, izinsiza zayo, kanye nokutholakala kwezentengiso. Ngamanye amazwi, i-Spark ibeka izinga futhi I-AI TOP ATOM iyincazelo ephathekayo ye-GIGABYTE ukuletha lawo mandla kumadeskithophu onjiniyela, izikhungo zocwaningo, namagumbi okufundela.

I-Architecture: NVIDIA GB10 Grace Blackwell ukusuka phezulu kuye phansi

I-denominator evamile yi-superchip I-NVIDIA Grace Blackwell GB10, i-SoC ehlanganisa i-CPU ne-GPU nenkumbulo ehlanganisiwe yomkhawulokudonsa ophezulu. Ohlangothini lwe-CPU, sithola iphrosesa 20-core Armv9, enama-cores e-Cortex-X925 asebenza kahle angu-10 kanye nama-cores angu-10 Cortex-A725 ukuze asebenze kahle. Le nhlanganisela ihlinzeka ngekhono kuyo yomibili imisebenzi yekhompyutha preprocessing kanye orchestration njengamapayipi e-AI ayinkimbinkimbi.

I-GPU isebenzisa i-architecture Blackwell ngesizukulwane sakamuva se-NVIDIA cores: I-5th Generation Tensor ukusheshisa i-AI kanye Isizukulwane sesine se-RT ukulandelela imisebe. Ngaphezu kwalokho, i-GB10 ihlanganisa 1 NVENC kanye ne-1 NNVDEC, iwusizo uma usebenza ngokusakazwa kwevidiyo kuzibalo noma ezinhlelweni zokusebenza ze-multimodal. Konke lokhu kuhlanganiswa I-NVLink-C2C, “ingcina” enesivinini esikhulu kakhulu phakathi kwe-CPU ne-GPU edlula kude lokho okunikezwa yi-PCI Express 5.0 ngaphakathi kwesistimu.

Ukusebenza kwe-AI namasayizi wamamodeli

Idatha edonsa ukunaka isibalo esinembayo esiphansi saso I-AI ekhiqizayo: 1.000 TOPS ku-FP4, okulingana ne-petaFLOP engu-1Empeleni, lokhu kusho ukuthi, ku-chassis encane, ungakwazi ukugijima futhi ushune kahle ulimi lwamanje namamodeli e-multimodal, wenze i-prototyping esheshayo, futhi usebenzise okucatshangwayo okufunayo emkhiqizweni ngaphandle kokuthembela efwini.

Ngokuqondene nobukhulu bamamodeli, ulwazi oluvela emithonjeni aluyona i-monolithic futhi kuhle ukuqoqa konke: ngakolunye uhlangothi, Ukwesekwa kwasendaweni kukhonjisiwe kumamodeli anepharamitha efika ku-200.000 billion kuyunithi eyodwa; ngakolunye uhlangothi, iphinde iveze ukuthi "ithembisa ukusekelwa kwama-LLM afinyelela ku-70 trillion kanye namapharamitha angama-200 trillion (US)." Lokhu kwakhiwa kokugcina kudlala ngendlela engaqondakali phakathi "kwamabhiliyoni" nge-Spanish kanye "nebhiliyoni" ngesiNgisi, kodwa umlayezo oyinhloko usasele: Umkhawulo wokusebenza kwala maqembu ungaphakathi ngokuphelele kumaligi aphezulu kuma-LLM amanje.

Lapho amayunithi amabili exhumene, umkhawulo ohlanganyelwe ukhuphuka ufinyelele cishe 405.000 billion amapharamitha, ukuvula umnyango kumamodeli amakhulu nakakhulu kanye nemithwalo yemisebenzi esabalalisiwe. Kulabo abahlola okuqukethwe okunwetshiwe, ama-RAG anezinkomba ezinkulu, noma imigqa yezigaba eziningi, leyo ndlu eyengeziwe yenza wonke umehluko.

Lolu hlobo lokusebenza akulona nje okwalabhorethri: ukuthuthukiswa, imfundo, ucwaningo, isayensi, amadolobha ahlakaniphile namarobhothi Zivela njengezimo zemvelo. Ukuba ne-petaflop ye-AI etafuleni kusheshisa ukuhlolwa, kunciphisa izindleko zokusebenza, futhi kunika amandla ukuphindaphinda ngaphandle kokulinda kulayini emafini.

Imemori ehlanganisiwe, isitoreji kanye nomkhawulokudonsa

Inkumbulo yesistimu iyiphuzu eliqinile: 128 GB we-LPDDR5X ebhasini elingu-256-bit, enomkhawulokudonsa ocishe ube ngu-273 GB/s. Ukuba kubumbene phakathi kwe-CPU ne-GPU, yenza kube lula uhlelo futhi igwema amakhophi angenamsebenzi, okubalulekile ekuhambiseni ama-tensor amakhulu namaqoqo ekuqeqesheni okunesisindo esincane noma ukuqondiswa kwemali ephezulu.

Ngokuba isitoreji, I-GIGABYTE icabanga ukufika 4 TB ngaphakathi I-SSD I-PCIe Gen5, nakuba kukhona ukucushwa kwezentengiso okuqala kusuka 1 TB ku-PCIe 4.0. Ngale komthamo, iphuzu liwukuthi amabhodlela ayancipha lapho kulayishwa amamodeli amakhulu, noma kopisha amafayela amakhulu ngokushesha, amaqoqo okushumeka noma umbono namasethi edatha yomsindo ngesivinini esihle, okugcina iqembu lisebenza “indawo emnandi.”

Ukuxhumana kwangaphakathi I-NVLink-C2C iphawula omunye umehluko obalulekile. Ngokunikeza umkhawulokudonsa ophakeme ngokusobala kuno PCIe 5.0 kokuxhumana kwe-CPU-GPU ngaphakathi kwe-SoC, kunciphisa ukubambezeleka futhi kusizakala ngenkumbulo ehlanganisiwe ezimeni ezizwelayo ukudluliswa, okuyizona kanye lezo ezilawula i-AI ne-ML ekhiqizayo.

Ukuxhumana nokunwetshwa ukuze kuqhubekele phambili

Emachwebeni nangaphambili kokuxhumana, ayifinyeleli. Njengokwejwayelekile, okokusebenza kuhlanganisa 10GbE kunethiwekhi enezintambo ezisebenza kahle, I-Wi‑Fi 7 ne-Bluetooth 5.3 ngokuxhumeka kwesimanje okungenantambo, kanye I-HDMI 2.1a futhi ahlukahlukene I-USB 3.2 Gen 2x2 Type-C (kufaka phakathi imbobo ye-PD IN yamandla). Umsindo weziteshi eziningi uphuma nge-HDMI, ngakho futhi uhlanganisa nezidingo ze-multimedia ngaphandle kwanoma yiziphi izesekeli ezengeziwe.

Ucezu oluvumela ukukhula yi- I-NVIDIA ConnectX‑7 SmartNIC. Ixhuma amasistimu amabili anezixhumanisi zokufika ku I-200 Gbps, ivumela ikhompuyutha yeqoqo le-low-latency. Imithombo iphinde ichaze lesi sikhombimsebenzisi njenge I-ConnectX‑7 InfiniBand NIC, indlela yokugcizelela ukugxila ekubambeni okuncane kanye nomkhawulokudonsa oqhubekayo. Empeleni, lokhu kuvumela hamba kusuka ku-200B kuya ku-~405B amapharamitha lapho uhlanganisa amanodi amabili.

Ngaphezu kwale nyunyana yamaqembu amabili, ngokombono wokusebenza kungenzeka beka amayunithi amaningi e-AI TOP ATOM futhi uzixhume ndawonye usebenzisa i-ConnectX‑7 ukuze uzenze zisebenze kanyekanye futhi usize ukuvumelanisa ifolda yenethiwekhiEzingqungqutheleni zocwaningo noma amagumbi okufundela athuthukile, ukuba "namabhulokhi" ahlanganisa ngendlela eyimojuli kuyasebenziseka ikakhulukazi.

Isoftware, izinhlelo zokusebenza nezinsiza

Akukho ukuyekethisa ku-software: ithimba liyasekela I-NVIDIA DGX OS y Ubuntu Linux, futhi ihlanganisa ne Isitaki sesoftware ye-NVIDIA AI, okungukuthi, isethi yamathuluzi, izinhlaka, nemitapo yolwazi esetshenziselwa ukwakha nokuphakela amaphrojekthi e-AI asezingeni eliphezulu. Lokhu kwehlisa umgoqo wokungena kubo bobabili asebevele baku-ecosystem ye-CUDA nasemaqenjini okufundisa, futhi kuvumela, isibonelo, buyekeza i-firmware ku-Linux njengokujwayelekile.

I-GIGABYTE futhi yengeza yayo I-AI TOP Utility, isendlalelo esenza lula ukugeleza okuvamile: Ukulanda imodeli, inkomba, i-RAG, nokufunda ngomshini kusuka kusixhumi esibonakalayo esiqonde kakhudlwana. Isiphakamiso sigcizelela ubumfihlo bedatha yendawo kanye nokuvikeleka, okuthile okufanele izinhlangano okufanele zithobele imithetho noma ezikhetha nje ukugcina ulwazi lwazo olubucayi ngaphakathi endlini.

Ukuklama ngokomzimba, ukusetshenziswa kanye nefomethi

Isethi ilingana ku-chassis cishe 1 ilitha. Kuyibhokisi elihlangene ngendlela emangalisayo elinezilinganiso ezisemthethweni ze 150 × 150 × 50,5 mm kanye nesisindo nxazonke 1,2 kg. Amandla anikezwa i-adaptha yangaphandle. 240 W, okwanele ukusekela ukusebenza okuphezulu ngaphandle kwezinkinga, futhi, njengoba kuchazwe imithombo, isistimu injalo iyahambisana nezitolo ezijwayelekile zasendlini ngaphandle kwesidingo sengqalasizinda ekhethekile.

Ngale kosayizi, i-GIGABYTE icabange mayelana nokuthunyelwa okusheshayo nokuhlelekile: iyasebenza nqwabelanisa ngokomzimba amayunithi amaningi futhi uwaxhume nge-SmartNIC efanelekile, esiza ukwakha iqoqo elincane endaweni encane kakhulu. Kubasebenzi bephrojekthi, ukukwazi "ukwengeza amabhokisi" nokukhipha ama-node njengoba kudingeka kuyigolide elihlanzekile.

Ukucaciswa okubalulekile kwe-GIGABYTE AI TOP ATOM

• I-SoC - NVIDIA GB10
  • CPU - 20 Armv9 cores ene-10× Cortex‑X925 kanye ne-10× Cortex-A725
  • Izakhiwo – NVIDIA Grace Blackwell
  • GPU - I-Blackwell Architecture
  • Izinhlobo ze-CUDA – Blackwell Generation
  • Ama-Tensor Cores - Isizukulwane sesi-5
  • Ama-RT cores - Isizukulwane sesi-4
  • Ukusebenza kwe-Tensor - 1.000 AI TOPS (FP4)
  • I-VPU - 1 × VENNC, 1 × NNVDEC
• Inkumbulo yesistimu – 128 GB 256-bit LPDDR5X (≈273 GB/s)
• Isitoreji - Kufika ku-4 TB PCIe Gen5 SSD
• Ividiyo – HDMI 2.1a okukhiphayo
• Audio - Umsindo weziteshi eziningi nge-HDMI
• Red
  • 1× 10GbE RJ45
  • I-SmartNIC I-ConnectX‑7 ukuze ujoyine amasistimu amabili afika ku-200 Gbps
  • I-Wi-Fi 7 y I-Bluetooth 5.3
• USB
  • 1× USB 3.2 Gen 2×2 Type‑C (PD IN)
  • 3× USB 3.2 Gen 2×2 Type‑C (kufika ku-20 Gbps)
• Ukusetshenziswa – 240 W i-adaptha yangaphandle
• Ubukhulu - 150 × 150 × 50,5 mm
• Isisindo - 1,2 kg

Intengo nokutholakala

I-GIGABYTE AI TOP ATOM isiyatholakala ezitolo ezifana I-Newegg, ngentengo yokuqala cishe Amadola ka-3.499,99 yenguqulo ene-1 TB PCIe 4.0 SSD. Ngenhla kukhona inketho ye 4 TB PCIe 4.0 (cishe u-$3.899,99) nokuhluka okuphezulu nge 4 TB PCIe 5.0 (cishe u-$3.999,99). Njengenhlalayenza, kuyafaneleka qinisekisa ukutholakala nabasabalalisi bendawo futhi uthintane nekhasi lomkhiqizo elisemthethweni kanye nokukhululwa kwabezindaba ukuze uthole ukucushwa kanye nezinsuku zokugcina.

I-ecosystem nezinye izindlela ngesisekelo esifanayo

Ukufika kwale nkundla akukhawulelwe ku-GIGABYTE. I-MSI ne-ASUS sebevele bakhombisile i-PC yabo encane kusithenjwa esifanayo se-NVIDIA DGX Spark, esiqinisekisa ukuthi sibhekene nezinga le-de facto ku-“desktop AI”. Kumsebenzisi wokugcina, lokhu kuhumushela ekukhetheni isihlanganisi esilingana kangcono ngokuya ngedizayini, izimbobo, isevisi nezinye izinsiza ezengeziwe, ukwazi ukuthi i-technical core (GB10) Kuyafana.

Ukuqhathanisa okuqondile: I-AI TOP ATOM vs DGX Spark

Ngakho konke okungenhla, ukuqhathanisa kuqondwa kangcono. I-DGX Spark Kungumqondo kanye nenkundla yereferensi: ichaza ifomethi ye-1 litre, ukusetshenziswa kwe- GB10 Grace Blackwell kanye nomgomo wokuletha i-petaFLOP FP4 etafuleni. Ngaleso sisekelo, i I-AI TOP ATOM I-GIGABYTE iwumkhiqizo wokugcina ogqamisa umbono ngenkumbulo ehlanganisiwe engu-128GB, isitoreji esisheshayo, amachweba azinikele, kanye nokusetshenziswa kwawo ukwenza lula ukugeleza kwe-AI.

Ngakho-ke, ngokwamandla aluhlaza kanye nohlu lwamamodeli sibheka izibalo ezicishe zilingane (1.000 TOPS FP4 nangaphezulu. I-200B amapharamitha endaweni eyodwa, nge ~ 405B kumanodi amabili axhunyiwe). Umehluko ongokoqobo ubonakala ku- ukuqaliswa okuphathekayo:: izinsiza ezifakiwe (I-AI TOP Utility), inhlanganisela eqondile yamachweba, izinketho zokugcina futhi kunjalo, Inani nokutholakalaUma ufuna umuzwa we-OEM "out-of-the-box", i-ATOM ilingana nebhili.

Ukuvala umbuthano, kufanele sikhumbule ukuthi imithombo yaseSpain iveze ini: I-GIGABYTE ikhuluma ngokucacile ukupakisha amayunithi amaningi kanye nokusetshenziswa kusukela imfundo emhlabeni wesayensi, ngokusebenzisa amadolobha ahlakaniphile namarobhothi. Umoya we-Spark uhlala unjalo: amandla we-AI wangempela ngefomethi encane futhi enokuqina ngokushesha lapho kudingeka imisipha eyengeziwe.

Kuyacaca ukuthi sikhuluma ngokugxuma kokukhiqiza ku-"desktop AI": i-PC encane efana ne- I-GIGABYTE AI TOP ATOM ifa ingqikithi ye I-NVIDIA DGX Spark (owayekade eyi-Project DIGITS), uyagxila I-1 petaFLOP FP4 ngelitha elilodwa, linikeza 128 GB ehlanganisiwe nge-NVLink‑C2C, khuphuka uye ku ~405B amapharamitha lapho ubhanqa ama-node amabili nge-ConnectX‑7, futhi iletha Isitaki se-NVIDIA nge-DGX OS/Ubuntu kanye nensiza ye-AI TOP ukuze kube lula ukulanda, ukuqagela, i-RAG ne-ML; uma sengeza i 10GbE, the I-HDMI 2.1a, the I-USB 3.2 Gen 2 × 2 kanye nezintengo esezivele zishicilelwe ku-Newegg, sinesiphakamiso esiqinile kakhulu socwaningo, amakilasi, namathimba afuna ukusheshisa imijikelezo yawo ye-AI ngaphandle kokusetha iseva ezinikezele.

I-athikili ehlobene:

Ungayithola kanjani i-NPU ku-Copilot+ PC: Umhlahlandlela Ophelele Wokuthola Okuningi Ku-AI

Isaka

Umbhali oshisekayo ngomhlaba wamabhayithi nobuchwepheshe ngokujwayelekile. Ngiyathanda ukwabelana ngolwazi lwami ngokubhala, futhi yilokho engizokwenza kule bhulogi, ngikubonise zonke izinto ezithakazelisayo kakhulu ngamagajethi, isofthiwe, ihadiwe, izitayela zobuchwepheshe, nokuningi. Inhloso yami ukukusiza ukuthi uzulazule emhlabeni wedijithali ngendlela elula nejabulisayo.