NVIDIA Unleashes Desktop AI Supercomputing with DGX Spark and DGX Station
NVIDIA has dramatically expanded access to its cutting-edge AI technology with the unveiling of the DGX Spark and DGX Station, personal AI supercomputers powered by the groundbreaking NVIDIA Grace Blackwell platform. These desktop powerhouses bring data centre-level AI capabilities directly to developers, researchers, and data scientists, marking a significant leap forward in AI accessibility.
Democratising AI Development
The newly announced DGX Spark, formerly known as Project DIGITS, and DGX Station are built to enhance AI development. They allow users to prototype, fine-tune, and run large AI models directly on their desktops. With these powerful tools, AI professionals can streamline their workflow and accelerate innovation. This eliminates the traditional reliance on remote cloud or data centre resources, enabling rapid iteration and development. Users retain the flexibility to deploy their models on NVIDIA DGX Cloud or any other accelerated infrastructure when needed.
“AI has transformed every layer of the computing stack. It stands to reason a new class of computers would emerge — designed for AI-native developers and to run AI-native applications,” stated Jensen Huang, founder and CEO of NVIDIA. “With these new DGX personal AI computers, AI can span from cloud services to desktop and edge applications.”
DGX Spark: The World’s Smallest AI Supercomputer
The DGX Spark is powered by the NVIDIA GB10 Grace Blackwell Superchip, optimised for a desktop form factor. This superchip is equipped with a powerful NVIDIA Blackwell GPU. It features fifth-generation Tensor Cores and FP4 support, enabling exceptional AI performance. In fact, it can deliver up to 1,000 trillion operations per second of AI computing.
Furthermore, it is designed to handle the most demanding AI workloads. This includes tasks such as fine-tuning and inference with advanced models like NVIDIA Cosmos Reason and NVIDIA GR00T N1.
A key innovation is the NVIDIA NVLink-C2C interconnect technology, which provides a CPU+GPU-coherent memory model with 5x the bandwidth of fifth-generation PCIe. This allows for efficient data access and optimised performance for memory-intensive AI tasks.
DGX Station: Data Centre Performance on Your Desktop
The DGX Station, powered by the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, not only brings data centre-level performance to the desktop but also delivers exceptional efficiency. Furthermore, featuring a massive 784GB of coherent memory space, the DGX Station significantly accelerates large-scale training and inferencing workloads. In addition, the GB300 Desktop Superchip includes an NVIDIA Blackwell Ultra GPU connected to a high-performance NVIDIA Grace CPU via NVLink-C2C, thereby ensuring seamless communication and performance.
Further enhancing its capabilities, the DGX Station incorporates the NVIDIA ConnectX-8 SuperNIC, supporting up to 800Gb/s networking. This enables high-speed connectivity for multi-DGX Station setups and network-accelerated data transfers.
Both systems leverage the NVIDIA CUDA-X AI platform for enhanced performance. Additionally, they provide access to NVIDIA NIM microservices through the NVIDIA AI Enterprise software platform. This ensures optimized and enterprise-supported inference capabilities.
Industry Support and Availability
Global system builders are actively preparing to develop and release DGX Spark and DGX Station systems. Notably, this includes industry leaders such as ASUS, Dell Technologies, HP Inc., and Lenovo. With their combined expertise, these companies aim to deliver cutting-edge AI computing solutions to the market.
Reservations for DGX Spark systems are open immediately. DGX Station is expected to be available from manufacturing partners like ASUS, BOXX, Dell, HP, Lambda, and Supermicro later this year.
For more information, viewers can watch the NVIDIA GTC keynote and register for sessions at the event, which runs through 21st March.
