SmartCap: Coordinated CPU–GPU Power Capping for Performance-Assurance Energy Efficiency
Open Access
- 5 July 2026
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 175-187
Abstract
Large-scale systems face strict power limits when running heterogeneous CPU–GPU applications that exhibit varying responses to power caps. Existing power capping methods either target GPUs or CPUs alone or require extensive per-application profiling, making them impractical for production environments. We present SmartCap, an online framework that coordinates CPU–GPU power capping to reduce energy consumption while bounding performance degradation under a user specified threshold. SmartCap combines an offline phase, which builds a performance model and initializes a performance–power matrix, with an online phase that employs lightweight profiling and collaborative filtering to infer performance across the CPU–GPU cap space. Performance evaluation on three heterogeneous systems with a diverse workload suite shows that SmartCap achieves 95% mean prediction accuracy and as much as 20% energy savings, with negligible runtime overhead for most applications, demonstrating its practicality for performance-aware energy optimization in production environments.Keywords
All Related Versions
This publication has 47 references indexed in Scilit:
- Toward an End-to-End Auto-tuning Framework in HPC PowerStack2020
- Indicator-Directed Dynamic Power Management for Iterative Workloads on GPU-Accelerated Systems2020
- A Data-Driven Frequency Scaling Approach for Deadline-aware Energy Efficient Scheduling on Graphics Processing Units (GPUs)2020
- DVFS-aware application classification to improve GPGPUs energy efficiencyParallel Computing, 2019
- Predicting Performance Using Collaborative Filtering2018
- Design and Implementation of Movie Recommendation System Based on Knn Collaborative Filtering AlgorithmITM Web of Conferences, 2017
- A survey and measurement study of GPU DVFS on energy conservationDigital Communications and Networks, 2017
- Mutual Information Analysis: How, When and Why?2009
- Large-Scale Parallel Collaborative Filtering for the Netflix Prize2008
- GROMACS: Fast, flexible, and freeJournal of Computational Chemistry, 2005