We’ve heard of Nvidia exiting on the CPU front for years, following the stunning reach of its Project Denver CPU and its Tegra K1 mobile processor connected in 2014 – but now, the company’s CPU is making a massive comeback . New nvidia grace, An arm-based processing chip designed specifically for AI data centers.
This is a good time for Nvidia to flex its arm: it is currently trying to buy the arm itself for $ 40 billion, making it one of the world’s premier computing companies, especially for the age of “A.I. Effort “, and may be a chip. First proof point. Arm is also having a moment in the consumer computing space, where recently Apple’s M1 chips have retained our concept of laptop performance. It’s also more competition for Intel, of course, Whose shares sank after the nvidia announcement.
The new Grace is named after the pioneering Grace Hopper’s computing, and it is coming to bring “10x the performance of today’s fastest servers on the most complex AI and high-performance computing workloads”, according to Nvidia in 2023 . This will be attractive to research firms producing supercomputers, among which, the Swiss National Supercomputing Center (CSCS) and Los Alamos National Laboratory Already signed to create
Even in 2023.
A Grace Next is already on the 2025 roadmap. Slide here from Nvidia’s GTC 2021 presentation where it announced the news:
I would recommend to read What our friends at anandtech say About where Grace might fit into the data center market and Nvidia’s ambitions, and it’s worth noting that Nvidia isn’t yet releasing as much as Specs – but Nvidia says it’s with a record Features fourth-gen HMLink with 900 Gb / s Interconnection between CPU and GPU. “Critically, this exceeds the CPU’s memory bandwidth, meaning NVIDIA’s GPU will have a consistent link to the CPU that can access system memory at full bandwidth, and the entire system with a single shared memory Allows. Address space, “writes Anandtek.