Implementing AI on Power11: Introducing the IBM Spyre Accelerator

A major highlight is IBM SpyreTM PCIe Gen5 x16 AI Accelerator Adapter (IBM Spyre Accelerator), which is a new reduced-power Peripheral Component Interconnect Express (PCIe) accelerator that is designed for scalable AI inference. With a highly parallel architecture and support for modern AI numerics, Spyre enables efficient running of large language models (LLMs), embedding, retrieval-augmented generation (RAG), and other enterprise AI workloads. Integrated with Red Hat AI and vLLM, it delivers on-premises AI with predictable performance and data-sovereign control.

The publication also addresses end-to-end lifecycle requirements, such as security, governance, machine learning operations (MLOps), and industry use cases, and shows how Power11 and Spyre support regulated, data-intensive environments. Combined with IBM's broader software and open-source ecosystem, the platform provides a complete blueprint for deploying enterprise-grade AI on IBM Power.