Understanding, Analyzing, and Optimizing Agentic AI: A CPU-Centric Perspective
The paper discusses the optimization of agentic AI execution from a CPU-centric perspective. It identifies system bottlenecks and proposes scheduling optimizations to improve CPU-GPU utilization. Experimental results show significant reductions in latency for both homogeneous and heterogeneous workloads.
- ▪Agentic AI converts LLM-based inference into autonomous problem-solving capabilities.
- ▪The paper characterizes and analyzes system bottlenecks introduced by agentic AI workloads.
- ▪Two scheduling optimizations, CPU-Aware Overlapped Micro-Batching and Mixed Agentic Scheduling, are proposed to enhance performance.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2511.00739 (cs) [Submitted on 1 Nov 2025 (v1), last revised 16 Apr 2026 (this version, v3)] Title:Towards Understanding, Analyzing, and Optimizing Agentic AI Execution: A CPU-Centric Perspective Authors:Ritik Raj, Souvik Kundu, Ishita Vohra, Hong Wang, Tushar Krishna View a PDF of the paper titled Towards Understanding, Analyzing, and Optimizing Agentic AI Execution: A CPU-Centric Perspective, by Ritik Raj and 4 other authors View PDF HTML (experimental) Abstract:Agentic AI serving converts monolithic LLM-based inference to autonomous problem-solvers that can plan, call tools, perform reasoning, and adapt on the fly.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv.org.
Discussion
0 commentsMore from arXiv.org
