Nvidia Announces New Chip, “Grace,” for AI Supercomputing

On Monday, Nvidia announced its first-ever data center CPU, which engineers spent a combined total of 10,000 years—yes, you read that right—working toward.

The new chip is called Grace, in honor of programming pioneer Grace Hopper, who helped create the first all-electronic digital computer in the 1950s.

Why this is big: Like its namesake, “Grace” is all about high-performance computing—built to power tasks like natural language processing, recommender systems, and AI supercomputing on an unprecedented scale. And the organizations building supercomputers are taking notice.

Los Alamos National Laboratory is first in line to receive Grace, and the Swiss National Supercomputing Centre will use it to help power its new supercomputer, “Alps.”

• Billed as 10x faster than today’s fastest supercomputer, Alps will be released in 2023 and focus on weather and climate simulation, quantum physics, and quantum chemistry.

Zoom out: Until this week, Nvidia’s highest-performing chip worked in tandem with Intel processors. Now that the company’s traded those out for Arm-based ones, it’s taking on the current leaders in data center chipmaking—Intel and Santa Clara-based AMD—for the first time.