Saturate your Tensor Cores: Intel at NVIDIA GTC 2026

Author:

Jeff McVeigh, Vice President & General Manager, Data Center Strategic Programs, Intel

The Mission-critical role of CPUs for AI infrastructure at scale 

Imagine spending your entire startup seed funding on GPU hardware only to find it sitting idle 90% of the time. It sounds like an agentic-workload cautionary tale, but it’s a reality for many companies. The individual hardware isn’t the problem. It’s a balanced system design issue. If the CPU infrastructure can’t provide the data throughput that an AI model demands or can’t keep up executing actions in the agentic logic loop, the GPU infrastructure won’t run at full capacity.

CPU+GPU coordination took center stage at NVIDIA GTC 2026 when we announced that Intel® Xeon® 6 processors have been selected as the host CPU for NVIDIA DGX Rubin NVL8 systems. Here’s a recap of what else we shared and learned at the San Jose, California, event, and how Intel is helping teams find infrastructure balance and maximize GPU utilization.

Engineered for AI

At NVIDIA GTC, one theme stood out: CPUs are mission-critical to AI infrastructure.

In agentic AI, generating an answer is only the first step. A separate CPU infrastructure is required to execute the actions generated by the GPUs. Consider a vibe coding environment, for example. GPU systems generate new code based on a developer’s query, then CPU systems compile the code, run validation tests, and manage CI/CD systems before feeding the results back to the GPUs for the next iteration.

If the CPU-only infrastructure lacks the necessary tool support or the ability to execute the logic loop quickly, it often becomes the main system bottleneck. That’s why demand for CPUs is rising and why balanced CPU+GPU system design matters.

This is where Intel Xeon 6 processors make all the difference for systems that include GPUs. They’re engineered for modern AI systems and serve as the foundation for agentic AI, building on decades of x86-optimized software for sandbox execution of code compilation, database queries, web search, and rich tool/API calls.

As the host CPU, Intel Xeon 6 processors deliver robust PCIe and IO capability and the high memory bandwidth needed to help ensure GPUs are fully utilized. They handle orchestration, decision logic, tool invocation, and data security while efficiently running smaller models and coordinating GPUs for large‑scale execution. This results in AI that’s not only powerful, but deployable, governable, highly secure, and reliable in the real world.

Winning Together

A fundamental shift is underway in AI system design. Delivering AI at scale requires deep collaboration between CPU and GPU processors, and Intel and NVIDIA are working together to deliver exactly that.

Through strategic co-engineering, NVIDIA’s accelerated computing platforms are paired with Intel’s x86 CPUs and ecosystem. This helps ensure AI systems can move data efficiently, orchestrate workloads across massive GPU clusters, and operate reliably in enterprise environments.

Together, we are enabling production-ready AI architectures across NVIDIA Blackwell- and Rubin-based systems, helping organizations move beyond experimentation to scalable deployments that balance performance, efficiency, and security.

Looking ahead, our collaboration continues to evolve with custom x86 variants, deeper CPU-GPU integration, and next-generation interconnects designed to further optimize performance and efficiency across the AI stack.

On the Show Floor

Attendees saw Intel and NVIDIA collaboration in action at GTC with live demonstrations, including:

• A preview of Intel® Trust Domain Extensions Connect (Intel® TDX Connect), a hardware-based security engine that enables encrypted, direct communications between virtual machines.
• An enterprise retrieval-augmented generation (RAG) workload using Intel® Trust Domain Extensions (Intel® TDX) with highly secure offloading to an NVIDIA H100 GPU—illustrating how safeguarded workloads can extend beyond the CPU.
• An up-close view of next-generation AI systems, including the Dell PowerEdge R770 and NVIDIA DGX B300.
• Newly announced support for NVIDIA Dynamo on Intel Xeon CPUs to enable heterogeneous inference.

Why CPU Architecture Matters for GPU Workloads

GTC conversations highlighted a simple engineering reality: the realized performance of GPUs ultimately depends on the infrastructure that feeds them. As a key part of that infrastructure, the host CPU needs to keep up to support multiple-GPU training clusters, real-time inference pipelines, and complex agentic and reinforcement learning workloads. Intel Xeon 6 processors are uniquely suited for these requirements based on:

• Fast single-thread performance
  Priority Core Turbo Technology (PCT Technology) enables dynamic frequency boost for critical GPU-feeding tasks while maintaining background operations at base frequency for optimal efficiency.(1)
• Superior I/O and memory bandwidth
  Industry-leading support for up to 192 PCIe 5.0 lanes per 2S server and MRDIMM technology.(2) These enable faster data movement to increase GPU utilization.
• Accelerated AI inference
  The only server-class CPU with Intel® Advanced Matrix Extensions (Intel® AMX) that enables accelerated matrix multiplication for data preparation and CPU-based inference.

Whether you’re a startup building the next AI breakthrough or scaling AI across a large enterprise, success isn’t just about buying fast GPUs. It’s about building a balanced CPU+GPU infrastructure that enables full GPU utilization, faster iteration, and more efficient use of your entire compute resources.

Get the Technical Deep Dive

Learn more about the architecture details and performance data behind Intel Xeon 6 processors:

• Unleash the Power of Intel Xeon 6 Processors as AI Host CPUs with Priority Core Turbo
• Accelerating Innovation: 5 Reasons for Choosing Intel Xeon 6 Processors as Host CPUs to Drive AI Success

Watch this video for more from NVIDIA GTC 2026: 

(1) Available in select Intel® processor SKUs.
(2) See Intel® Xeon® 6 processor section at www.intel.com/performanceindex for details.

Notices and Disclaimers

Performance varies by use, configuration, and other factors. Learn more on the Performance Index site.
Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available ​updates. See backup for configuration details. No product or component can be absolutely secure.
Your costs and results may vary.
Intel technologies may require enabled hardware, software, or service activation.
© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.