NVIDIA unveils Project DIGITS AI supercomputer for $3,000

NVIDIA has announced the introduction of NVIDIA Project DIGITS, a personal AI supercomputer designed to provide AI researchers, data scientists, and students with the power of the NVIDIA Grace Blackwell platform.

Project DIGITS features the NVIDIA GB10 Grace Blackwell Superchip, delivering one petaflop of AI computing performance. It enables the prototyping, fine-tuning, and execution of large AI models using 200-billion-parameter capacity.

This new AI supercomputer allows users to develop and run inference on models on desktop systems before deploying them on cloud-based or data centre infrastructures. "AI will be mainstream in every application for every industry. With Project DIGITS, the Grace Blackwell Superchip comes to millions of developers," said Jensen Huang, founder and CEO of NVIDIA. "Placing an AI supercomputer on the desks of every data scientist, AI researcher and student empowers them to engage and shape the age of AI."

The GB10 Superchip is developed on the Grace Blackwell architecture and delivers up to a petaflop of performance at FP4 precision. It includes an NVIDIA Blackwell GPU with CUDA cores and fifth-generation Tensor Cores. The architecture integrates with a 20-core Arm-based NVIDIA Grace CPU, aided by MediaTek in design, optimising it for efficiency, performance, and connectivity.

This AI supercomputer runs on just a standard electrical outlet, providing 128GB of memory and up to 4TB of NVMe storage. It is capable of supporting 200-billion-parameter language models, enhancing AI innovation potential. Networking with NVIDIA ConnectX permits linking two Project DIGITS units to manage up to 405-billion-parameter models.

The Grace Blackwell architecture allows users to prototype and refine models locally with Linux-based NVIDIA DGX OS. These models can be scaled on NVIDIA DGX Cloud, cloud instances, or dedicated infrastructure supporting NVIDIA's AI Enterprise software platform.

Developers can access NVIDIA's AI software library for prototyping, using tools and frameworks available on the NVIDIA NGC catalogue and NVIDIA Developer portal. NVIDIA NeMo framework aids model fine-tuning, with NVIDIA RAPIDSTM boosting data science acceleration. Common frameworks like PyTorch, Python, and Jupyter notebooks are supported.

NVIDIA further supports the development of agentic AI applications through NVIDIA Blueprints and NVIDIA NIM microservices, which are available under the NVIDIA Developer Program. For production phases, NVIDIA AI Enterprise licenses offer security, support, and updates.

Project DIGITS is scheduled for availability in May through NVIDIA and its partners, with a starting point of USD $3,000.