At the NVIDIA GTC 2026 conference in San Jose,Intel officially announced a major cooperation, and its Xeon 6 (Xeon 6) processor will become the host CPU of NVIDIA's next-generation flagship AI server system DGX Rubin NVL8.

This cooperation takes the cooperation between the two companies on the x86 architecture a step further. Previously, on the platform based on DGX B300 Blackwell, the two parties have established x86 architecture cooperation using the Xeon 6776P processor. This also means that Intel has successfully "embraced" NVIDIA and gained a firm foothold in the field of AI servers.

It is reported that,DGX Rubin NVL8 is NVIDIA's new generation flagship AI server system, focusing on emerging application scenarios such as agent AI and inference systems. In this system, the host CPU plays a crucial role and is mainly responsible for task orchestration, memory management, scheduling, and data transmission to the GPU accelerator.

As AI inference workloads gradually transform into agent AI and inference systems, these tasks have increasingly higher requirements for single-core performance and memory bandwidth.

Intel said that the Xeon 6 processor can just meet these needs through comprehensive improvements in memory capacity, bandwidth and I/O performance. The processor is built on a platform that can support up to 8TB of system memory, which Intel says is particularly critical to supporting large language models as key-value caches continue to expand.

With the help of MRDIMM technology, its memory bandwidth has been increased by 2.3 times compared with the previous generation, which can significantly increase the speed of data transmission to the GPU accelerator.

At the same time, the PCIe 5.0 channel enables high-bandwidth accelerator connections, and Intel's "Priority Core Turbo" function can also focus powerful single-thread performance on orchestration, scheduling, and data transmission tasks, ensuring that the GPU maintains high utilization even if the workload becomes more complex.

In terms of security and compatibility, the Xeon 6 processor can provide security protection for the entire data path from the CPU to the GPU through Intel Trusted Domain Extensions (TDX). Among them, TDX technology will add hardware-based isolation and authentication functions through encrypted rebound buffers, which can just meet the needs for end-to-end confidential computing when AI inference is deployed in multiple scenarios such as data centers, clouds, and edges. In addition, Xeon 6 also adds support for the NVIDIA Dynamo inference orchestration framework. With this framework, CPU and GPU resources in the same cluster can be scheduled heterogeneously.

"In this new era, the role of the host CPU is crucial." Jeff McVeigh, Intel's vice president and general manager of data center strategic projects, said that it directly determines the orchestration efficiency, memory access speed, model security and throughput of the GPU-accelerated system.

Intel also mentioned that the Xeon processor’s mature x86 software ecosystem, rich enterprise deployment experience, and good compatibility with existing AI software stacks are the key reasons why NVIDIA chose it.

This cooperation between the two parties follows the same architectural foundation as DGX B300, allowing the two generations of Blackwell and Rubin products to achieve platform-level continuity, and is expected to promote the large-scale implementation of AI inference technology in data centers, clouds, edges and other scenarios.