Project DIGITS allows one to enjoy NVIDIA's full-stack AI computing and development from just US$3K

NVIDIA's dominance in industrial and large-scale AI supercomputing is already well-established, but the announcement of Project DIGITS reinforces its commitment to targeting the prosumer segment.

This innovative program introduces the NVIDIA GB10 Grace Blackwell Superchip, housed within a compact, “NUC-like” chassis capable of delivering 1 petaflop of AI performance (FP4). With this setup, users can prototype, refine, and run large-scale AI models directly from their desktops, ensuring seamless deployment to cloud or data center infrastructure—an approach that aligns with NVIDIA's ecosystem, which remains central to most AI applications today.

The Superchip is built on NVIDIA's cutting-edge CPU and GPU architecture, featuring the Grace CPU and Blackwell GPU with the latest iterations of CUDA and Tensor Cores. MediaTek also contributed to optimizing the system’s power efficiency and performance, earning recognition from NVIDIA.

For memory and storage, the system includes 128GB of RAM and up to 4TB of NVMe storage, which is likely user-upgradable. This configuration provides a solid foundation for handling large language models (LLMs) with up to 200 billion parameters. Moreover, NVIDIA’s ConnectX networking allows two units of Project DIGITS to work together, enabling computing capabilities for LLMs with up to 405 billion parameters.

NVIDIA’s robust suite of AI software, including NeMo and RAPIDS, comes pre-integrated, alongside support for essential tools like PyTorch and Python. For users ready to engage in industrial-level development, NVIDIA Blueprints and the NIM microservice offer scalable solutions tailored to the needs of Project DIGITS users.

Set to launch in May with a starting price of $3,000, Project DIGITS is now open for early sign-ups.