Nvidia's Digits is a tiny AI supercomputer for your desk

With all eyes on Nvidia's new graphics cards (and for good reason - these things are pretty powerful), the company released something that may end up being even bigger news: A personal AI supercomputer called Project Digits (we'll just call it Digits from now on because that's way cooler).

Digits is a tiny, Mac mini-like personal computer that should fit on pretty much every desk. You connect it to a keyboard and a monitor, plug it into power, and you're good to go.

But what's inside makes it pretty special. Digits is powered by Nvidia's new GB10 Grace Blackwell Superchip, which delivers up to 1 petaflop of AI performance, and Nvidia teamed up with MediaTek to make the chip more energy-efficient, meaning that running it requires the kind of power you get from a standard power outlet.

The GB10 is paired with 128GB of RAM and up to 4TB of NVMe storage. All of this will enable developers to run up to 200-billion-parameter large language models (LLMs). Using Nvidia ConnectX tech, they'll also be able to to pair two Digits computers to run up to 405-billion-parameter models.

Most of this stuff probably makes little sense to folks who aren't in the AI app development business. But for example, ChatGPT 3.5 had 175 billion parameters (ChatGPT 4 is much larger but we don't know the exact numbers), while Meta's most powerful LLM, Llama 3, has 405 billion parameters). This means you can run a very, very powerful LLM at home instead of relying on cloud infrastructure.

"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," Jensen Huang, founder and CEO of Nvidia, said in a written statement.

Developers who use Digits will be able to access Nvidia's library of AI software including development kits, orchestration tools, frameworks and models available in the Nvidia NGC catalog and on the Nviia Developer portal. They'll also have access to the Nvidia NeMo framework, and Nvidia RAPIDS libraries.

The biggest news here, perhaps, is the price. Digits will be available in May, starting at $3,000. This sort of money should make it available to a large number of smaller companies and researchers, who will use it to create and test AI apps.

If this still sounds too expensive, you can try Nvidia's Jetson. It's a $249 AI home computer that launched last December, and it can handle up to 8 billion parameters.

Mashable is on the ground live at CES 2025! We’re covering all the wildest and most important developments this week, so please keep checking back in with us. Want to submit a product you represent for our teams’ consideration as we identify the Best of CES? Here’s more info on how to do it.