NVIDIA launches DGX Spark: A compact AI supercomputer for scale-out clustering
The DGX Spark is designed to support local AI engineering efforts with its compact 1.2kg build. Equipped with high-speed networking, it enables seamless scalability. The system features 273GB/s memory bandwidth, four USB4 ports, an HDMI port, a 10GbE port, and a dual-port 200GbE NIC, allowing cluster integration with other Spark units.
Hyderabad: NVIDIA unveiled the DGX Spark as a portable AI mini supercomputer suitable to scale-up clustering capabilities. It is powered by the Grace Blackwell GB10 Superchip, which features 20 Arm cores, a Blackwell GPU, a confident 128GB of LPDDR5X memory, and 200GbE NVIDIA ConnectX-7 network.
The DGX Spark is intended to assist local AI engineering pursuits since it is lightweight, weighing just 1.2kg. It has high-speed networking, making it easy to scale-out. The machine also comes with 273GB/s memory bandwidth, four USB4 ports, HDMI, a 10GbE port, and dual-port 200GbE NIC, which supports cluster mode with other Spark units.
Given competitive AI computing capabilities, NVIDIA positions the DGX Spark at an affordable price of $3,999, with an available bundle consisting of a pair of DGX Sparks and a QSFP cable to facilitate clustering.
The ace in its sleeve is the NVIDIA NVLink-C2C interconnect in use, which establishes a 5x faster CPU-GPU coherent memory model than PCIe 5.0, making the system exceedingly suitable for AI inference and large-scale model tuning. The performer boasts its software suite based on NVIDIA DGX OS, promising compatibility with such products as DGX Cloud and NCCL for heavy multi-node AI workloads.
In the meantime, NVIDIA has completed the DGX Station, an even more compact superpower AI desk housing the 128-core Grace Blackwell Ultra chip, 784GB unified memory, and 800Gb/s NVIDIA ConnectX-8 networking. While DGX Spark is now available for reservation, the DGX Station should be out later this year.
The likes of Dell, ASUS, HP, and Lenovo are anticipated to be the manufacturer and distributor of these systems, providing easy AI access for developers, researchers, and businesses. In view of its size and power, the DGX Spark is setting the trend in AI for local model building, edge AI processing, and large-scale AI clustering solutions.
