Nvidia integrates Intel Sapphire Rapids CPU into DGX H100 systems for system-level compute optimization.

Nvidia has selected Intel's next-generation Xeon Scalable processor Sapphire Rapids as the CPU for its upcoming DGX H100 AI system, according to CEO Jensen Huang's confirmation at the BofA Securities 2022 Global Technology Conference on Tuesday. The DGX H100 will house eight H100 GPUs based on Nvidia's new Hopper architecture, connected via fourth-generation NVLink interconnect and capable of delivering 32 petaflops of AI performance using its FP8 format. The system is scheduled to arrive by the end of 2022. Huang stated: "We buy a lot of x86s. We have great partnerships with Intel and AMD. For the Hopper generation, I've selected Sapphire Rapids to be the CPU for Nvidia Hopper, and Sapphire Rapids has excellent single-threaded performance. And we're qualifying it for hyperscalers all over the world. We're qualifying it for datacenters all over the world. We're qualifying it for our own server, our own DGX. We're qualifying it for our own supercomputers."

The selection of Sapphire Rapids represents a reversal from Nvidia's previous choice of AMD's second-generation Epyc server CPU, codenamed Rome, for its DGX A100 system introduced in 2020. Industry publication ServeTheHome had reported in mid-April that Nvidia maintained motherboard designs for both Sapphire Rapids and AMD's upcoming Epyc CPU, codenamed Genoa, for the DGX H100, indicating the decision had not yet been finalized at that time. While Intel can claim this as a victory in its effort to regain technology leadership, the broader competitive landscape reveals this as a relatively small win in the larger battle between Nvidia, Intel, AMD and other companies over GPUs and accelerators—spurring Intel to invest heavily in its upcoming Ponte Vecchio GPU and AMD to enhance competitiveness with its latest Instinct GPUs.

Beyond the DGX H100 decision, Nvidia's strategic focus is shifting toward its own Arm-compatible CPU, Grace, designed to combine CPU and GPU in a single package to accelerate data flow for AI workloads and demanding applications. Nvidia plans to introduce the Grace Hopper Superchip next year alongside the 144-core, CPU-only Grace Superchip, with a new DGX system likely to follow. Huang emphasized that "Grace is going to be an amazing CPU" that will enable Nvidia to optimize components, systems, and software end-to-end. The architecture is designed for recommender systems and large language models used by hyperscale companies, though Huang indicated broader applications are planned. Huang stated: "Grace has the advantage that in every single application domain that we go into, we have the full stack, we have all of the ecosystem all lined up, whether it's data analytics, or machine learning, or cloud gaming, or Omniverse, [or] digital twin simulations. In all of the spaces that we're going to take Grace into, we own the whole stack, so we have the opportunity to create the market for it." Intel also plans to introduce its own CPU-GPU design for servers, the Falcon Shores XPU, in 2024.