If you're in the market for a personal AI supercomputer, Nvidia has introduced Project Digits, set to launch in May. At its core is the GB10 Grace Blackwell Superchip, delivering the processing power necessary for sophisticated AI models while remaining compact enough to fit on a desk and run on a standard power outlet—previously, such power demanded much larger, energy-intensive systems. This desktop-sized device supports AI models with up to 200 billion parameters and starts at $3, 000. Its design resembles a Mac Mini. “AI will become essential in every application across all industries. Project Digits, with the Grace Blackwell Superchip, brings this to millions of developers, ” stated Nvidia CEO Jensen Huang.

“Providing an AI supercomputer for data scientists, AI researchers, and students empowers them to lead in the AI era. ” Project Digits comes with 128GB of unified memory (a typical laptop might have 16GB or 32GB of RAM) and up to 4TB of NVMe storage. For more intensive needs, two systems can link to manage models with up to 405 billion parameters, comparable to Meta’s 405-billion parameter Llama 3. 1 model. The GB10 chip offers up to 1 petaflop of AI performance (equivalent to 1 quadrillion calculations per second) at FP4 precision, optimizing speed through approximations. The system includes Nvidia's latest CUDA cores and fifth-generation Tensor Cores, linked via NVLink-C2C to a power-efficient Grace CPU with 20 Arm-based cores, co-designed with MediaTek to enhance efficiency and performance. Users gain access to Nvidia’s AI software, which includes development kits, orchestration tools, and pre-trained models available via the Nvidia NGC catalog. Operating on Linux-based Nvidia DGX OS, it supports frameworks like PyTorch, Python, and Jupyter notebooks. Developers can fine-tune models through Nvidia’s NeMo framework and enhance data science workflows with Nvidia RAPIDS libraries. With Project Digits, users can develop and test AI models locally and then deploy them to cloud services or data centers using the same Grace Blackwell architecture alongside Nvidia AI Enterprise software. Nvidia offers other accessible devices: in December, it launched a $249 version of the Jetson computer for AI tasks, aimed at hobbyists and startups, called the Jetson Orin Nano Super, which handles models with up to 8 billion parameters.