Nvidia will still develop the Arm CPUs in a full spectrum
The highlight of last year’s GTC 2021 keynote was the launch of the first data center CPU called Grace. This is an Arm-based processor designed for AI and high-performance computing applications, and is expected to start shipping in 2023. The Alps supercomputer at the Swiss National Supercomputing Centre and the Los Alamos National Laboratory in the United States will be the first to use the Grace CPU.
On the Grace CPU, NVIDIA will use the next-generation Arm Neoverse core for the first time, and the fourth-generation NVIDIA NVLink will be used for internal communication. A bidirectional transfer bandwidth of up to 900 GB/s can be provided between the CPU and the GPU, and the transfer bandwidth between the CPU and the CPU is 600 GB/s. At the same time, LPDDR5x memory will be used, which can provide 500 GB/s bandwidth and has an ECC verification function. In the opinion of NVIDIA CEO Jensen Huang, the launch of Grace CPU is of great significance to NVIDIA, making it a company with three different types of chips (CPU/GPU/DPU) and the ability to re-architect the data center to advance AI.
In September 2020, Nvidia announced the purchase of Arm from SoftBank for $40 billion in cash and stock. After a lengthy tug-of-war with the relevant regulators, Nvidia’s ambitious plans took a hit and the acquisition ended in failure. However, Nvidia has obtained a license to use Arm-related IP for a period of 20 years.
Jensen Huang has confirmed on the earnings conference call that Nvidia will develop a series of CPUs based on the Arm architecture for different application areas, and Grace is just the beginning. Nvidia plans to bring the full range of its accelerated computing platform to Arm-based CPUs, which will eventually expand to robotics, self-driving cars, cloud computing, supercomputers, and more. Nvidia is currently working on several CPU projects around the Arm architecture.
Among them, the success of Arm-based data centers and supercomputing systems will motivate NVIDIA to accelerate the development of higher-end CPUs and will continue to promote the three-chip strategy of CPU, GPU, and DPU in the future.