Agentic AI Workflows Reveal Hidden CPU Bottlenecks, Prompt New Server Design
A Microsoft Azure production study and controlled experiments map how agentic AI workloads bounce between CPU and GPU, exposing fragmented execution and bursty demand that cripple conventional uniform servers. The authors propose a taxonomy, quantify latency and resource patterns, and introduce Agora—a prototype that harvests idle CPU cores and oversubscribes GPU memory to boost utilization while protecting tail latency.