NVIDIA's Revolutionary Grace Processor Announcement
At the GTC 2021 conference, Nvidia announced its first processor called Grace, based on the ARM architecture.
We've been promised 10x the performance of "today's fastest servers" for AI and HPC.
For years, we hardly heard a word from Nvidia about processors, after in 2014 year its Project Denver processor and related Tegra K1 mobile processors appeared. But now the company is actively returning to chips with new Nvidia Grace, an Arm-based processor specifically designed for AI data centers.
It's a good time for Nvidia to play Arm: it is currently trying to buy Arm itself for $40 billion , positioning it as an attempt to "create the world's leading computer company for the age of artificial intelligence", and this chip may be the first point of evidence. Arm also has a moment in consumer computing where Apple M1 chips recently turned our concept of laptop performance upside down . Of course, this is also a big competition for Intel, whose shares fell after Nvidia's announcement .
Meet the NVIDIA Grace #CPU, leveraging the flexibility of @arm's data center architecture and designed from the ground up to accelerate the largest HPC and AI workloads. # GTC21 https://t.co/PHDaxrfzQv pic.twitter.com/uck0akde3a
- NVIDIA GTC (@NVIDIAGTC) April 12, 2021
The new Grace is named after computing pioneer Grace Hopper, and according to Nvidia it will arrive in 2023 to deliver "10x the performance of today's fastest servers on the most complex AI and HPC workloads" . This will certainly make it attractive to research firms building supercomputers, which the Swiss National Supercomputing Center (CSCS) and Los Alamos National Laboratory have already signed on to build in 2023.
Grace Next is already on the roadmap for 2025. Here is a slide from the Nvidia GTC 2021 presentation where it announced the news:
It's worth noting that Nvidia hasn't released many specs yet, but Nvidia says it has a fourth-generation NVLink with a record-breaking 900GB/s interconnect between CPU and GPU. “Critically, this is more than the CPU memory bandwidth, meaning that NVIDIA GPUs will have a coherent cache link to the CPU that can access system memory at full bandwidth, while also allowing the entire system to have a single shared memory. – address space,” writes anandtech.