Ryan King & Chris Tobias | Red Hat Summit 2025
summary
Longstanding Red Hat–Intel work centers on concrete results: a RHEL 9.5 + Xeon stack sets a published world-record benchmark, Signal Six measures cost per token on Gaudi 3 as competitive, and the teams run hundreds of POCs measuring inference and edge use cases. They integrate vLLM (now part of Red Hat) and announce LLMD for distributed inference, expose CPU/GPU parity for inference, and leverage silicon features like TDX plus post-quantum plans to cut attack surface and power per watt.
AI-written summary · about
