NVIDIA Launches AI-Powered PC Solutions: DGX Spark & DGX Station Desktops Featuring 72-Core Grace CPU and Blackwell GPUs with Up to 784 GB Memory

NVIDIA has officially unveiled its latest offerings in the realm of AI computing with the introduction of the DGX Spark and DGX Station AI PCs, which are built on the powerful Grace CPUs and Blackwell GPUs.

Unleashing Desktop Power with NVIDIA’s DGX Spark and DGX Station

Press Release: The DGX Spark, previously known as Project DIGITS, along with the DGX Station, represents NVIDIA’s new line of supercomputers designed for desktop environments. These innovative systems harness the capabilities of the NVIDIA Blackwell Ultra platform, enabling AI developers, researchers, data scientists, and students to efficiently prototype, fine-tune, and run large AI models. Users have the flexibility to execute these models locally or utilize the robust NVIDIA DGX Cloud, as well as other accelerated cloud infrastructures.

Both the DGX Spark and DGX Station bring the formidable power of the Grace Blackwell architecture—previously exclusive to data centers—directly to desktop users. Leading global system builders, including ASUS, Dell, HP Inc., and Lenovo, are collaborating to develop these groundbreaking technologies.

Empowering Innovation with DGX Spark

Hailed as the world’s smallest AI supercomputer, DGX Spark is designed to enable millions of researchers, data scientists, and robotics developers, as well as students, to explore the frontiers of both generative and physical AI. Its architecture promises to deliver exceptional performance and capabilities.

At its core is the NVIDIA GB10 Grace Blackwell Superchip, meticulously designed to fit a desktop format. The GB10 is equipped with a powerful NVIDIA Blackwell GPU, featuring fifth-generation Tensor Cores and FP4 support, capable of executing an astounding 1, 000 trillion operations per second for AI compute tasks, including fine-tuning and inference using advanced AI reasoning models.

The GB10 Superchip leverages NVIDIA’s NVLink-C2C interconnect technology, providing a coherent CPU+GPU memory model that boasts five times the bandwidth of fifth-generation PCIe. This allows for more efficient data access between the GPU and CPU, significantly enhancing performance for memory-intensive AI tasks.

Accelerating Development with DGX Station

The NVIDIA DGX Station stands out by delivering data-center-level capabilities right to desktop workstations aimed at AI development. As the first desktop to incorporate the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, it features an astonishing 784GB of coherent memory to support extensive training and inference workloads. The GB300 is linked to an NVIDIA Blackwell Ultra GPU equipped with the latest Tensor Cores and FP4 precision, connected through NVLink-C2C for optimal system performance.

Furthermore, DGX Station includes the NVIDIA ConnectX-8 SuperNIC, specifically designed to enhance hyperscale AI computing operations. With capabilities supporting network speeds of up to 800Gb/s, this technology facilitates rapid, efficient network connections, allowing multiple DGX Stations to collaborate on larger workloads and enabling accelerated data transfers for demanding AI applications.

Anticipated Availability

The DGX Station is projected to become available later this year through various manufacturing partners, including ASUS, BOXX, Dell, HP, Lambda, and Supermicro.

Source & Images