{"rewrite":{"id":"r_0f4838d77733a648379fd466","clusterId":"c_1cdf93c84d4a672de04e8982","slug":"nvidia-s-dgx-spark-developer-kit-links-four-units-into-a-512gb-cluster","model":"deepseek-v4-1-flash","headline":"NVIDIA's DGX Spark Developer Kit Links Four Units Into a 512GB Cluster","summary":"NVIDIA announced the DGX Spark Developer Kit, a system that connects four DGX Spark compact AI supercomputers into one cluster. The combined cluster carries 512GB of memory and 20TB of storage, and NVIDIA says it can run models with up to 700 billion parameters locally, including DeepSeek-V4.1. The four units link through ConnectX-7, so no separate network switch is needed. The kit is on display at SC26 in Chicago, which runs November 15 to 20, 2026.","whyItMatters":"NVIDIA is selling the interconnect as much as the box: four handheld-sized DGX Sparks become a 512GB cluster through ConnectX-7, which puts a 700B-parameter model on a desk without a separate switch purchase.","webCardHtml":"\u003cp\u003eEach DGX Spark is small enough to hold in one hand and carries the GB10 Grace Blackwell Superchip with 128GB of memory. The Developer Kit does not change the individual unit. It changes how many of them count as one machine.\u003c/p\u003e\u003cp\u003eThe four units are joined by ConnectX-7, NVIDIA\u0026#39;s networking system, so a buyer does not have to add a switch to the deployment. The cluster totals 512GB of memory and 20TB of storage. NVIDIA lists DeepSeek-V4.1 among the models it can run locally at up to 700 billion parameters.\u003c/p\u003e\u003cp\u003eThe hardware is on the floor at SC26, the international supercomputing conference in Chicago, which runs November 15 to 20, 2026. The DGX Sparks and the cables for the cluster are sold through OEM vendors.\u003c/p\u003e","blueskyPost":"NVIDIA's DGX Spark Developer Kit runs a 700B model on four linked units, and ConnectX-7 removes the switch from the bill. Local inference at that scale turns cluster networking into the cost line.","twitterPost":"NVIDIA's DGX Spark Developer Kit links four units over ConnectX-7, so the network switch drops out of the build. The cabling is the cost saving.","threadsPost":"NVIDIA's DGX Spark Developer Kit runs models up to 700B parameters across four linked units, and ConnectX-7 means no separate network switch. The interconnect choice is what makes a desk-side cluster cheap to assemble, not the 512GB of memory alone.","newsletterBlurb":"NVIDIA announced the DGX Spark Developer Kit, which connects four DGX Spark compact AI supercomputers into a single cluster. The combined system carries 512GB of memory and 20TB of storage and can run models up to 700 billion parameters, including DeepSeek-V4.1. The four units link through ConnectX-7, so no separate network switch is needed, and the kit is on display at SC26 in Chicago through November 20.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20261117-dgx-spark-developer-kit/\",\"title\":\"NVIDIA Announces 'DGX Spark Developer Kit' That Connects Four Compact AI Supercomputers DGX Spark to Achieve 512GB of Memory, Capable of Running Models Up to 700B\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":5632,"outputTokens":853,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1794889265,"createdAt":"2026-11-17T04:16:05.000Z","publishedAt":"2026-11-17T04:19:57.000Z","updatedAt":"2026-11-17T04:19:57.000Z"},"cluster":{"id":"c_1cdf93c84d4a672de04e8982","canonicalTitle":"NVIDIAが小型AIスパコンDGX Sparkを4台つなげてメモリ容量512GBを実現する「DGX Spark Developer Kit」を発表、最大700Bのモデルを実行可能","representativeArticleId":"a_af987381be5da771fb556d5a","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[\"DGX Spark Developer Kit\",\"DGX Spark\"],\"studios\":[\"NVIDIA\"],\"people\":[],\"type\":\"announcement\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-11-17T03:37:00.000Z","lastSeenAt":"2026-11-17T03:37:00.000Z","updatedAt":"2026-11-17T04:19:57.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20261117-dgx-spark-developer-kit/","title":"NVIDIAが小型AIスパコンDGX Sparkを4台つなげてメモリ容量512GBを実現する「DGX Spark Developer Kit」を発表、最大700Bのモデルを実行可能"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":["DGX Spark Developer Kit","DGX Spark"],"studios":["NVIDIA"],"people":[],"type":"announcement","domain":"other","is_roundup":false},"keyFacts":["NVIDIA announced the DGX Spark Developer Kit, which connects four DGX Spark units into one cluster.","The cluster totals 512GB of memory and 20TB of storage.","NVIDIA says the cluster can run models with up to 700 billion parameters, including DeepSeek-V4.1.","The four DGX Sparks are interconnected using ConnectX-7, so no separate network switch is needed.","The kit is exhibited at SC26 in Chicago, held November 15 to 20, 2026."],"pageLinks":{"healthSlugs":["dgx-spark-developer-kit","dgx-spark","nvidia"],"storySlugs":[],"entities":[{"slug":"dgx-spark-developer-kit","supported":2,"disputed":0,"contradicted":0,"unverified":4,"overall":"green","story_count":1,"has_dossier":false},{"slug":"dgx-spark","supported":1,"disputed":0,"contradicted":0,"unverified":5,"overall":"green","story_count":3,"has_dossier":true},{"slug":"nvidia","supported":6,"disputed":0,"contradicted":1,"unverified":40,"overall":"red","story_count":56,"has_dossier":true}],"stories":[],"entitiesOk":true,"storiesOk":true}}
