During the NVIDIA GTC event, the GPU giant unveiled two compact supercomputers: DGX Spark and DGX Station. Both systems are built on the NVIDIA Blackwell Ultra platform, targeting developers, researchers, data scientists, and students who need to train, infer, and deploy large language models.
"AI has transformed every layer of the computing stack. It's time for a new type of computer—one designed specifically for AI-native developers and capable of running AI-native applications," said Jensen Huang, NVIDIA's CEO and co-founder, in a press release. "With these new DGX personal AI computers, AI can now scale from cloud services to desktops and edge deployments."
DGX Spark Delivers 784 GB of Memory in a Compact Form Factor
NVIDIA claims that DGX Spark, previously known as Project Digits, is the world's smallest supercomputer. Inside, the GB10 Grace Blackwell superchip features a Blackwell GPU capable of performing 1,000 trillion AI calculations per second. The NVIDIA NVLink-C2C interconnect technology links the CPU and GPU with five times the bandwidth of PCIe Gen 5, according to NVIDIA.
Users can export AI models from DGX Spark to DGX Cloud or any accelerated cloud or data center infrastructure.
DGX Spark is compatible with NVIDIA’s recently announced GR00T N1 robot foundational model and the Cosmos world-generation system, which teaches AI-powered robots about the physical world.
The waitlist for DGX Spark is now open, priced at $3,000. DGX Spark will be manufactured by partners such as ASUS, Dell, and HP.
DGX Station Designed for Desktop AI Development
An updated version of DGX Station will be released later this year by NVIDIA's manufacturing partners, including ASUS, BOXX, Dell, HP, Lambda, and Supermicro. Pricing has not yet been disclosed.
With the current-generation DGX Station, users gain access to 784 GB of unified memory space, ideal for training and inference tasks involving large AI workloads. Internally, it is equipped with the NVIDIA GB300 Grace Blackwell Ultra desktop superchip, featuring a Blackwell Ultra GPU and an NVIDIA Grace CPU, all interconnected via NVLink-C2C.
The supercomputer also includes NVIDIA ConnectX-8 SuperNIC, supporting network connections of up to 800Gb/s. This allows multiple DGX Stations to be connected for handling massive workloads or accelerating data transfers over networks.
In addition, DGX Station integrates seamlessly with the NVIDIA CUDA-X AI platform, NVIDIA NIM microservices, and the NVIDIA AI Enterprise software suite for streamlined desktop AI development.
