Nvidia Project Digits: A Linux-powered desktop for AI developers

What’s making the headlines at CES 2025 are AI-powered TVs, new smart home gadgets, and fresh laptop releases. For my money, though, the big news from CES is Nvidia Project Digits. This revolutionary desktop AI supercomputer is designed to bring unprecedented computing power to artificial intelligence (AI) developers, researchers, and students. And, by the way, it will be running Nvidia’s DGX OS, a customized Ubuntu Linux 22.04 distro.

As you might guess, DGX OS is a Linux distro designed with system-specific optimizations and configurations, drivers, and diagnostic and monitoring tools to provide a fully supported version of Linux for running AI, machine learning, and analytics applications on Nvidia DGX Supercomputers.

Also: The best tech we’ve seen at CES so far

On top of that, Nvidia will also provide AI software development kits; orchestration tools; Nvidia NGC catalog frameworks and models; the Nvidia NeMo framework for fine-tuning models; and Nvidia Rapids libraries for data science acceleration. Of course, as an Ubuntu Linux system at heart, you can run Ubuntu and Linux software. So, if you want to, you can play Doom on it. That’s probably the first I’ll do if I get my hands on one of these devices. 

Now, in a box that appears to be about the size of a Mac mini, Nvidia’s Project Digits PC promises to bring a petaflop of AI performance to your desktop. A petaflop, for those of you who don’t hang out in supercomputer land, is equal to 1,000,000,000,000,000 (one quadrillion) floating-point operations per second (flops).

Until 2008, when IBM rolled out its Roadrunner supercomputer, no one had ever hit that speed. Soon, you could have that level of computing performance in your home office. Even today, the latest Top 500 SuperComputer List has computers that squeak in with just over two petaflops. Mind you, these computers also need over a thousand cores to pull that trick off. Oh, and they all run Linux.

Project Digits will be almost Top 500-fast and is powered by Nvidia’s band’s new GB10 Grace Blackwell Superchip. With the Nvidia AI software stack preinstalled and 128GB of memory, developers can prototype, fine-tune, and infer large AI models of up to 200B parameters locally and then seamlessly deploy to the data center or cloud.

Also: The best robotics and AI tech of CES  

The Grace Blackwell Superchip is named after Grace Hopper, the famed computer scientist, United States Navy rear admiral, and the “Mother of Cobol”. In addition, the chip’s name honors David Harold Blackwell, an American mathematician who made significant contributions to game theory, probability theory, information theory, and statistics, and was the first Black scholar to be inducted into the National Academy of Sciences.

In this superpowered chip, you’ll find an ARM-based Grace CPU featuring 10 Cortex-X925 and 10 Cortex-A725 cores. That’s 20 cores in all. The CPU is backed up by a Blackwell GPU equipped with Nvidia’s latest CUDA and RT cores. The PC has 128 GBs of RAM and up to 4 TBs of flash storage to put all this chip firepower to good work.

Nvidia promises that AI developers will be able to work with models of up to 200 billion parameters on a single unit. For more demanding tasks, you can link two Project Digits machines together to handle models with up to 405 billion parameters.

Also: ZDNET joins CNET Group to award the Best of CES, and you can submit your entry now

Nvidia’s CEO, Jensen Huang, emphasized Project Digits’ transformative potential, stating: “Placing an AI supercomputer on the desks of every data scientist, AI researcher, and student empowers them to engage with and shape the age of AI.” 

With a starting price of $3,000, Project Digits aims to make high-performance AI development more accessible to a broader range of users, from small enterprises to schools. At that price point, I expect to see Project Digits desktops in homes. 

This Linux-powered system, set to launch in May 2025, promises to reshape the landscape of AI development by offering data center-level performance in a compact, desk-friendly form factor. I, for one, will be doing my darndest to get my hands on one.