Nvidia is a company that is almost synonymous with the world of GPUs. However, at the Nvidia GPU Technology Conference today, the firm unveiled its first data center CPU. Dubbed Grace, Nvidia's first foray into the realm of central processing units is based on the famed Arm architecture and promises to deliver 10x the performance of today's fastest servers.
Named after Grace Hopper, a computing pioneer who invented one of the first linkers and laid the foundations of COBOL, Nvidia's Grace will cater directly to the use cases that prioritize high-performance computing. These applications range from natural language processing, computer vision, to protein folding and quantum chemistry.
The motivation behind Grace was to produce a CPU that can tightly couple with today's GPUs so that system bottlenecks are removed. To achieve this, Nvidia chose the arm architecture and LPDDR5x memory subsystem with the 4th-gen NVIDIA NVLink interconnect technology. Taken together, Nvidia claims that this ecosystem will provide 10x better energy efficiency compared to traditional DDR4 memory and a record 900 GB/s connection between Grace and Nvidia GPUs on the system that will in turn lead to 30x higher aggregate bandwidth compared to today’s leading servers.
Following today's announcement, Nvidia is now a three-chip company (meaning it specializes in CPU, GPU, and DPU). Jensen Huang, the Founder and CEO of Nvidia, commented on the momentous occasion:
“Leading-edge AI and data science are pushing today’s computer architecture beyond its limits – processing unthinkable amounts of data. Using licensed Arm IP, Nvidia has designed Grace as a CPU specifically for giant-scale AI and HPC. Coupled with the GPU and DPU, Grace gives us the third foundational technology for computing, and the ability to re-architect the data center to advance AI. NVIDIA is now a three-chip company.”
While Simon Segars, CEO of Arm pointed towards how the Arm architecture could be the driving force behind data center hardware in the future:
“As the world’s most widely licensed processor architecture, Arm drives innovation in incredible new ways every day. Nvidia's introduction of the Grace data center CPU illustrates clearly how Arm’s licensing model enables an important invention, one that will further support the incredible work of AI researchers and scientists everywhere.”
As expected, Grace will be fully supported by the Nvidia HPC software development kit and the complete fleet of CUDA and CUDA-X libraries. Though you will have to wait a bit for it the CPU itself. Expected availability is two years away at the start of 2023. The Swiss National Supercomputing Centre (CSCS) and the U.S. Department of Energy’s Los Alamos National Laboratory have already announced plans to power their data centers using Grace. Further details can be found here.