NVIDIA has introduced Project DIGITS, a personal AI supercomputer designed to bring the capabilities of the NVIDIA Grace Blackwell platform to AI researchers, data scientists, and students globally. This initiative was announced at CES.

Project DIGITS is powered by the new NVIDIA GB10 Grace Blackwell Superchip, which offers a petaflop of AI computing performance. This allows users to prototype, fine-tune, and run large AI models directly from their desktop systems before deploying them on cloud or data center infrastructure.

Jensen Huang, founder and CEO of NVIDIA, emphasized the significance of this development: “AI will be mainstream in every application for every industry. With Project DIGITS, the Grace Blackwell Superchip comes to millions of developers,” he said. “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 integrates an NVIDIA Blackwell GPU with CUDA cores and Tensor Cores connected via NVLink-C2C interconnect to a high-performance NVIDIA Grace CPU. MediaTek collaborated on its design to enhance power efficiency and connectivity.

Project DIGITS provides 128GB of unified memory and up to 4TB of NVMe storage. It can handle up to 200-billion-parameter language models independently or connect two units for even larger model processing.

With this architecture, enterprises can prototype locally on Project DIGITS systems running Linux-based NVIDIA DGX OS before scaling using cloud or data center resources with consistent architecture support from NVIDIA's software platforms.

Users have access to various software tools through the NVIDIA NGC catalog and Developer portal for experimentation. They can also utilize frameworks like PyTorch and Python for development purposes.

For those looking to transition from research to production environments, the NVIDIA AI Enterprise license offers security and support services.

Project DIGITS will be available starting in May at a price point beginning at $3,000 through NVIDIA and its partners.