NVIDIA’s first attempt at making ARM-based processors didn’t go as well as the company planned. Having left the years of failure behind, the company is now making a second attempt at making ARM-based processors. However, instead of fighting back against established names like Intel and AMD in the consumer electronics segment, the company is taking a safer route by targeting large-scale neural network systems for AI data centers – a domain where the GPUs are already doing the heavy lifting. Named NVIDIA Grace, the result of NVIDIA-ARM’s recent joint effort is expected to be commercially available in 2023.
NVIDIA bought ARM from SoftBank for $ 40 billion in September last year. Back to NVIDIA Grace, the company says it will offer Up to 30x increase in total bandwidth compared to today’s fastest servers and “Up to 10x the performance of today’s fastest servers on the most complex AI and high-performance computing workloads.” NVIDIA hasn’t named Intel or any of its x86-based Xeon processors here – or AMD’s EPYC range for that matter – but the goals are obvious.
Meet NVIDIA Grace – the building block of the modern data center. It is powerful @Poor-based #CENTRAL PROCESSOR allows us to have the largest HPC and accelerate #AI Workloads. Watch the introduction: https://t.co/z51GV8rkhg # GTC21 pic.twitter.com/EySjYKPsDj
– NVIDIA (@nvidia) April 12, 2021
NVIDIA has covered the underlying architecture and core specifications in detail, with the exception of the key areas where the Grace processor will be used. Natural language processing, recommendation systems and AI supercomputing. For now, the company is only realizing that its latest offering will combine energy-efficient cores with a low-power subsystem. Specifically, NVIDIA Grace is touted as the first server CPU to use LPDDR5x memory, which the company claims offers twice the bandwidth and 10 times more energy efficiency compared to the DDR4 standard.
The latest NVIDIA offering uses next-generation Arm Neoverse cores designed to increase both performance and energy efficiency. And when paired with NVIDIA’s own GPUs, Grace is said to use the fourth generation NVIDIA NVLink interface to provide a bidirectional bandwidth of 900 GB / s. CSCS and Los Alamos National Laboratory have confirmed plans to manufacture supercomputers powered by NVIDIA Grace supercomputers, expected to go online in 2023.