If NVIDIA’s massive market share in industrial and world-scale AI supercomputing doesn’t give enough hints, I guess the announcement of Project DIGITS will definitely show that Team Green is gunning for the prosumer group.
This program incorporates the new NVIDIA GB10 Grace Blackwell Superchip into a “NUC-like” chassis that delivers 1 petaflop of AI performance (FP4), allowing users to prototype, fine-tune, and run large-scale AI models directly on their desktop systems which can be later deployed on cloud and data center infrastructure seamlessly since most AI applications nowadays are supported or revolve around the NVIDIA ecosystem most of the time.
At the heart of the Superchip is the latest CPU and GPU architecture currently offered by Team Green which is Grace CPU and Blackwell GPU with the latest generation of CUDA and Tensor Cores. NVIDIA also credited MediaTek for helping the system achieve top-tier power efficiency and performance.
Memory capacity-wise, it comes with 128GB of RAM and up to 4TB of NVMe storage presumably self-upgradable which is an adequate baseline for handling LLMs with up to 200B parameters. Additionally, NVIDIA’s ConnectX networking seamlessly connects 2 units of Project DIGITS to make up 405B parameter-capable AI computing.
And as expected, NVIDIA’s extensive list of AI software suites like NeMo and RAPIDS in addition to PyTorch and Python which are some of the fundamental languages and basics of AI computing are supported. Ready for some industrial-level development? NVIDIA Blueprints and NIM microservice are ready to offer their service small scale to Project DIGITS users.
NVIDIA will be making these machines available starting in May with a starting price of US$3,000 but early sign-ups are open so be sure to get on the list if you want to try it out.
