Performance Characterization of AI Workloads on CPU: A Methodological Framework
DOI:
https://doi.org/10.63282/3050-9246.IJETCSIT-V7I2P125Keywords:
CPU, Performance, AI, Optimization, Linux, ProfilingAbstract
This paper presents a systematic methodology for characterizing AI workload performance on CPU architectures through profiling, optimization, and analysis. We outline a complete framework encompassing workload selection, performance profiling using Linux perf tools, targeted optimization, multi-core scaling analysis, and system monitoring. This methodology provides a reusable framework for performance engineers across different architectures and deployment scenarios.
Downloads
References
[1] Brendan Gregg. “Systems Performance: Enterprise and the Cloud.” 2nd Edition, 2020.
[2] Linux perf Documentation. https://perf.wiki.kernel.org/
[3] PyTorch Performance Tuning Guide. https://pytorch.org/tutorials/recipes/recipes/tuning_guide.html
[4] TensorFlow Performance Guide. https://www.tensorflow.org/guide/performance
[5] Hennessy & Patterson. “Computer Architecture: A Quantitative Approach.” 6th Ed, 2017.
