At CES 2025 in Las Vegas, Nvidia unveiled Challenge Digits, a “private AI supercomputer” that gives get entry to to the corporate’s Grace Blackwell {hardware} platform in a compact shape issue.
“[Project Digits] runs all the Nvidia AI stack — all of Nvidia tool runs in this,” Nvidia CEO Jensen Huang mentioned onstage all through a press convention on Monday. “It’s a cloud computing platform that sits to your table … It’s even a workstation, if you happen to adore it to be.”
Designed for AI researchers, knowledge scientists, and scholars, Challenge Digits packs Nvidia’s new GB10 Grace Blackwell Superchip, which delivers as much as a petaflop of computing efficiency for prototyping, fine-tuning, and working AI fashions.
Nvidia claims a unmarried Challenge Digits unit can run fashions as much as 200 billion parameters in measurement. Parameters kind of correspond to a type’s problem-solving talents, and fashions with extra parameters typically carry out higher than the ones with fewer parameters.
The GB10, which used to be advanced in partnership with MediaTek, options an Nvidia Blackwell GPU hooked up to a 20-core Nvidia Grace CPU. Throughout the Challenge Digits enclosure, the chips are hooked as much as a 128GB pool of reminiscence and as much as 4TB of flash garage.
Symbol Credit:Nvidia
Nvidia says that two Challenge Digits machines will also be related in combination to run as much as 405-billion-parameter fashions, if a role requires it. Challenge Digits can ship a standalone enjoy, as alluded to previous, or hook up with a number one Home windows or Mac PC.
However it’s no longer reasonable. Challenge Digits machines, which run Nvidia’s Linux-based DGX OS, will likely be to be had beginning in Might from “best companions” for $3,000, the corporate mentioned.
So it received’t be simply any individual who can manage to pay for a Challenge Digits unit of their very own. Huang thinks that there’s a marketplace, alternatively.
“With Challenge Digits, the Grace Blackwell Superchip involves thousands and thousands of builders,” he mentioned in a observation. “Hanging an AI supercomputer at the desks of each knowledge scientist, AI researcher and pupil empowers them to interact and form the age of AI.”
