NVIDIA unveils DGX Spark & DGX Station AI desktops

NVIDIA has announced the release of its DGX Spark and DGX Station personal AI supercomputers, which are powered by the NVIDIA Grace Blackwell platform.

These desktop supercomputers are designed to provide AI researchers, developers, and data scientists the ability to prototype, fine-tune, and run large AI models from their desktops.

According to NVIDIA, these systems enable users to either run their models locally or deploy them on NVIDIA DGX Cloud or other accelerated cloud or data center infrastructures.

The new DGX Spark and DGX Station systems bring the Grace Blackwell architecture, traditionally available in data centers, directly to the desktop environment. ASUS, Dell, HP, and Lenovo are among the global system builders involved in the development of these systems.

Jensen Huang, the founder and CEO of NVIDIA, said, "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. With these new DGX personal AI computers, AI can span from cloud services to desktop and edge applications."

The DGX Spark, described as the world's smallest AI supercomputer, aims to provide researchers, data scientists, robotics developers, and students with enhanced capabilities for both generative and physical AI projects. The core of the DGX Spark features the NVIDIA GB10 Grace Blackwell Superchip. This chip is designed specifically for desktop use and includes a powerful Blackwell GPU with fifth-generation Tensor Cores and FP4 support, capable of up to 1,000 trillion operations per second in AI compute tasks.

The GB10 Superchip is enhanced by NVIDIA's NVLink-C2C interconnect technology, which provides a cohesive memory model between the CPU and GPU.

This interconnect allows for 5x the bandwidth compared to fifth-generation PCIe connectivity, benefiting memory-intensive AI workloads significantly.

For those using the DGX Spark, NVIDIA's comprehensive AI platform supports seamless model transition from desktops to the DGX Cloud or other accelerated infrastructure with minimal code alterations.

The DGX Station is NVIDIA's solution for bringing data centre-level AI performance to desktop users. It features the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip with a significant 784GB of coherent memory, which is crucial for handling large-scale training and inferencing tasks.

This system incorporates the NVIDIA Blackwell Ultra GPU, connecting it to a high-performance Grace CPU through NVLink-C2C, enhancing system communication and overall performance.

Additionally, the DGX Station comes equipped with the NVIDIA ConnectX-8 SuperNIC, enabling network speeds of up to 800Gb/s. This feature is particularly beneficial for high-speed connection between multiple DGX Stations, facilitating larger workload management and network-accelerated AI data handling.

The audio of these systems is bolstered by the NVIDIA CUDA-X AI platform, which aids in optimisation and speed of AI development on desktop systems. Users of the DGX Station will also have access to NVIDIA NIM microservices under the NVIDIA AI Enterprise software, providing optimized inference microservices and enterprise-level support.

While reservations for DGX Spark systems are currently open, DGX Station systems are expected to be available from manufacturers like ASUS, BOXX, Dell, HP, Lambda, and Supermicro later in the year.