NVIDIA has unveiled the world’s smallest desktop supercomputers, the Blackwell Ultra platform, which will enable AI developers, academics, data scientists, and students to prototype and configure large models on desktop PCs.
NVIDIA has unveiled the world’s smallest desktop supercomputers, the Blackwell Ultra platform, which will enable AI developers, academics, data scientists, and students to prototype and configure large models on desktop PCs.
DGX Spark (formerly known as Project DIGITS) and DGX Station allow you to run such models locally or deploy them to NVIDIA DGX Cloud or any other accelerated cloud infrastructure or data center.
They bring the powerful Grace Blackwell architecture, previously only available in data centers, to a desktop form factor. DGX Spark and DGX Station are being developed by global system manufacturers including ASUS, Dell, HP Inc. and Lenovo.
«AI has transformed every layer of the computing stack. It makes sense that a new class of computers would emerge—designed for AI-powered developers and to run AI-powered applications. With the new DGX PCs, AI can span cloud services, desktops, and edge applications,» said Jensen Huang, founder and CEO of NVIDIA.
DGX Spark is the world’s smallest AI supercomputer, enabling researchers, data scientists, and robotics developers to «push the boundaries of generative and physical AI» with high performance. It is based on the NVIDIA GB10 Grace Blackwell superchip, optimized for a desktop form factor.
The GB10 is powered by the powerful NVIDIA Blackwell GPU with 5th generation Tensor Cores and FP4 support, delivering up to 1,000 trillion operations per second of AI computation for fine-tuning and inference using the latest AI reasoning models, including the NVIDIA Cosmos Reason world foundation model and the NVIDIA GR00T N1 robot foundation model.
This superchip uses NVIDIA NVLink-C2C interconnect technology to provide a CPU+GPU coherent memory model with 5x the bandwidth of PCIe Gen 5. This allows 10GB of data access between the GPU and CPU to optimize performance for memory-intensive AI developer workloads.
NVIDIA DGX Station brings data center-class performance to desktops for AI development. It is the first desktop system built on the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, which features 784GB of coherent memory to accelerate large-scale training workloads.
The GB300 desktop superchip is equipped with an NVIDIA Blackwell Ultra GPU with the latest generation of tensor cores and FP4 precision, connected to a high-performance NVIDIA Grace CPU via NVLink-C2C.
DGX Station features NVIDIA ConnectX-8 SuperNIC, optimized for ultra-high-performance AI workloads. With support for up to 800 Gbps of network speed, the ConnectX-8 SuperNIC provides ultra-fast and efficient network connectivity.
This allows a desktop computer to achieve exceptional performance in AI development.
NVIDIA has already opened up pre-orders for DGX Spark. DGX Station is expected to be available from third-party vendors such as ASUS, BOXX, Dell, HP, Lambda, and Supermicro later this year.
