Nvidia Corp. (NVDA) said it has begun shipping DGX Spark, a compact desktop system it bills as the world’s smallest AI supercomputer, aiming to bring data-center-class performance to individual developers. Built on the company’s new Grace Blackwell architecture, the unit delivers up to one petaflop of AI performance and 128GB of unified CPU-GPU memory, supporting local inference on models with as many as 200 billion parameters and fine-tuning up to 70 billion. The machine integrates Nvidia GPUs and CPUs with ConnectX-7 networking and ships with the CUDA and Nvidia AI software stack preinstalled. Major PC makers including Acer, Asus, Dell, HP, Lenovo, Gigabyte and MSI will offer versions, with orders opening Oct. 15 via Nvidia.com and U.S. retailer Micro Center. Chief Executive Jensen Huang marked the launch by hand-delivering a unit to Elon Musk at SpaceX’s Starbase site, echoing Nvidia’s 2016 delivery of its first DGX system to OpenAI. Early users include Anaconda, Google, Hugging Face and Microsoft. Nvidia included customary cautionary language about forward-looking statements.
Related articles:
NVIDIA Spectrum-X Ethernet Switches Speed Up Networks for Meta and Oracle
NVIDIA Accelerates Robotics Research and Development With New Open Models and Simulation Libraries
NVIDIA and Intel to Develop AI Infrastructure and Personal Computing Products
