At CES 2025, AMD and Nvidia have given it their all trying to outshine one another, and while both companies made headway with their AI gaming chips, Nvidia gained an edge with the announcement of its AI supercomputer. Nvidia’s Project Digits debut has been nothing short of fascinating, providing us with a clear picture of how central AI is expected to be for the future of computing.
Keeping the Nvidia GB10 Grace Blackwell Superchip at the heart of the project, the AI company showed off its plans to compress advanced AI computing capabilities into a compact form factor that makes it the “world’s smallest AI supercomputer capable of running 200B-parameter models.” The company went on to explain, “With Project Digits, users can develop and run inference on models using their own desktop system, then seamlessly deploy the models on accelerated cloud or data center infrastructure.”
Image: Keynote by Jensen Huang, CEO of NVIDIA
Nvidia’s AI Supercomputer Just Made a Grand Entrance at CES 2025
From everything we’ve heard at the Project Digits overview event, we know that the compact AI supercomputer is set to launch in May. Interested users will then be able to pursue their own AI development projects through Nvidia’s Digits, as long as they are willing to spend $3,000 to purchase it.
The device has evidently been designed to empower every data scientist, AI researcher, and student with the full force of AI and all of its processing capabilities, however, you can also use it for your own AI passion projects if money is of no object.
Project Digits Overview—What Can You Do With Nvidia’s Supercomputer?
Those who bite the bullet and make the purchase can expect to see the Nvidia GB10 Grace Blackwell Superchip powering the gadget, with Nvidia’s customized Linux distro DGX OS running the show. Support for frameworks such as PyTorch, Python and Jupyter notebooks make it an efficient system to have on hand.
Not only will the Nvidia AI supercomputer be capable of running and supporting AI, but it will also give users access to NVIDIA’s library of AI software from the Nvidia NGC catalog and NVIDIA Developer portal. This will give users more room to experiment and test out models with tools such as the NeMo framework, optimizing work with the Nvidia RAPIDS libraries.
Nvidia also explained that for constructing agentic AI applications, users can turn to the Nvidia Developer Program for access to Nvidia Blueprint and Nvidia NIM microservices. “When AI applications are ready to move from experimentation to production environments, the Nvidia AI Enterprise license provides enterprise-grade security, support, and product releases of Nvidia AI software,” so consider this a one-stop shop for all your AI needs.
Image: Nvidia
Project Digits Specs—What will Power these AI Developments?
The GB10 super chip in the upcoming AI supercomputer combines the Nvidia Blackwell GPU—with the latest CUDA cores and 5th-gen Tensor cores—with the Nvidia Grace GPU, which features 20 power-efficient Arm-based cores. The SoC delivers up to 1 petaflop of AI performance at FP4 precision, which sets up the system for 1 quadrillion AI calculations per second.
Nvidia credits MediaTek for contributing to the optimization of the GB10 and putting their expertise on Arm-based SoC designs to good use.
Introducing the NVIDIA Project DIGITS personal AI supercomputer, powered by the GB10 Grace Blackwell Superchip: https://t.co/sHjW920BE4
Project DIGITS enables developers to prototype, fine-tune and inference models locally and seamlessly deploy at scale to the data center or… pic.twitter.com/UZwm1evrzU
— NVIDIA (@nvidia) January 8, 2025
Despite the AI supercomputer’s advanced processing power, Nvidia reports that the device will rely on standard electrical outlets to operate. This keeps energy efficiency in mind as conversations around AI and its impact on the planet continue to grow.
The Nvidia Project Digits systems will include 128GB of unified, coherent memory, and will be further supported by 4TB of NVMe storage to handle AI workloads. Those who find the system to still fall below their computing needs can consider investing in two systems that can be linked using the Nvidia ConnectX network for larger models with up to 405 billion parameters.
Final Thoughts on Project Digits
Memory bandwidth numbers and other key details are not available currently, which has users wondering what the company isn’t saying about their mini PC. The Apple Mac Mini and, to a much smaller extent, the upcoming Windows 365 Link, have already grabbed consumer attention with their promise of high processing capabilities.
This leaves us wondering whether the target audience of researchers and AI explorers will actually take to the device and how their numbers will add up in total sales once the product is officially released. With more experts in the field gaining access to advanced processing within their limited budget and space, we may just see the AI landscape pick up its pace exponentially in the coming years.