Ahead of the unveiling of the Core 200 Plus series processors, Intel shared details on why it thinks its Xeon 6 lineup is key for 5G and 6G expansion. Following that, today Intel announced at Nvidia’s GTC 2026 conference that its new Xeon 6 processors will serve as the host CPUs for Nvidia DGX Rubin NVL8 systems.
For anyone wondering, Nvidia Rubin, launched at CES 2026, is the company’s next‑generation AI platform built to handle large‑scale reasoning and agentic AI models. It focuses on eliminating communication and memory bottlenecks, and promises to deliver faster inference at lower cost per token compared to the Blackwell generation.
Rubin introduces a new Transformer Engine with adaptive compression for higher NVFP4 performance, a sixth‑generation NVLink interconnect that doubles bandwidth, and unifies up to 72 GPUs, and third‑generation confidential computing for hardware‑based security. Overall Rubin is speced to deliver 3.6 TB/s of bandwidth per GPU for a total of close to 260 TB/s of connectivity throughput.
Similar to how Intel previously highlighted the importance of traditional processors in AI, the company emphasized again how inference workloads are increasingly dependent not only on GPU throughput but also by CPU‑driven system performance. The company reasons that host processors have to manage memory, orchestrate tasks, and distribute workloads, functions that directly influence cluster efficiency and total cost of ownership (TCO) for enterprises.
In this regard, Intel says that Xeon 6 brings balanced performance, fast memory speeds, and a mature enterprise software ecosystem to Rubin NVL8 deployments, while also offering robust PCIe and I/O bandwidth for heterogeneous workloads. Features such as Priority Core Turbo and single threaded performance are said to help keep data flowing smoothly to GPUs.
The DGX Rubin NVL8 systems are being built upon the architectural foundation established with Intel Xeon 6776P in current Nvidia Blackwell‑based DGX B300 platforms. This is to ensure easy upgradability and continuity for organizations already invested in Xeon‑powered AI clusters. Intel also pointed to ecosystem support, including new compatibility with Nvidia Dynamo, enabling heterogeneous inference across CPUs and forthcoming GPUs.
Intel says security remains a priority as inference scales and confidential computing becomes critical across CPU‑GPU data paths. Here, Intel’s Trust Domain Extensions (TDX) is said to provide hardware‑based isolation and attestation to provide the necessary security.