Mini-PCs: DGX Spark Systems with Nvidia Hardware Finally Appear
With a five-month delay, mini-PCs with Nvidia's GB10 CPU are arriving. Numerous partners like Asus and MSI are on board.
Elon Musk officially receives the first DGX Spark from Nvidia CEO Jensen Huang personally.
(Image: Nvidia)
Nvidia begins shipping its first own mini-PC, the DGX Spark. CEO Jensen Huang handed over the first unit to Elon Musk in his capacity as SpaceX CEO. Such photo opportunities have become a tradition: in 2016, OpenAI received the first DGX-1 server and most recently the first DGX B200.
In an email announcement, Nvidia states that the DGX Spark will be available for order on its website, Nvidia.com, starting Wednesday, October 15. The first mini-PCs are reportedly on their way from Asia to Europe. Nvidia is initially serving previous reservations from its marketplace.
Several PC manufacturers are offering their versions of the DGX Spark: Acer (Veriton GN100), Asus (Ascent GX10), Dell (Pro Max with GB10), Gigabyte (AI Top Atom), HP (ZGX Nano AI Station), Lenovo (Thinkstation PGX), and MSI (EdgeXpert MS-C931). All use the same mainboard and differ only in their case, cooling, and SSD.
Months ago, Nvidia was charging 3689 Euros for a DGX Spark with a 4 TB SSD in Germany. Asus is selling its Ascent GX10 a bit cheaper in its shop (starting from 3431,84 €) and plans to deliver from October 20. Mifcom lists the Gigabyte AI Top Atom at a high price.
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmung wird hier ein externer Preisvergleich (heise Preisvergleich) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (heise Preisvergleich) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
20-core processor with powerful GPU
The heart of the mini-PCs is Nvidia's GB10 processor, each with ten ARM cores of type Cortex-X925 and Cortex-A725. The integrated graphics unit consists of 6144 shader cores. The CPU and GPU jointly access 128 GB of LPDDR5X memory.
Videos by heise
Nvidia is focusing on AI performance: the GPU can perform one quintillion operations per second (1 Petaflops) in the simple FP4 data format. The 128 GB of memory should be sufficient for local AI models with up to 200 billion parameters.
Actually, the DGX Spark and partner systems were supposed to appear in May were supposed to be released in May, but were then delayed by five months. According to rumors, problems with the GB10 processor were to blame. They are said to be responsible for even longer delays for the end-user version N1X, partly due to alleged issues with different display resolutions.
(mma)
