Intel previews new Xeon designs for AI agentic workloads
TL;DR. Intel, Foxconn, and others unveiled rack designs for Xeon processors optimized for AI agents and inference, packing thousands of CPU cores. - The designs support up to 36,864 cores using Intel's new Xeon 6 and Xeon 6+ processors in a 100kW rack. - These blueprints address the CPU-centric demands of AI agent harnesses, which connect models to tools. - Vector Core Compute and Together.AI will be early adopters of Intel's disaggregated inference platform.
- Intel introduced rack-scale reference designs with Foxconn for high-density CPU compute for AI agents.
- The designs integrate up to 36,864 E-cores from Intel's Xeon 6+ processors within a 100kW power envelope.
- This initiative specifically targets CPU-intensive AI agentic workloads, which manage AI model interactions.
- The disaggregated inference blueprint, a collaboration with SambaNova, also secured its first customer, Together.AI.