Uncovering Detailed Power Characterizations of GPUs on Edge Platforms
GPUs in edge platforms operate on a tight energy budget.Although power characterization of server GPUs has been studied, there has been a lack of systematic study of the power and thermal characteristics of edge GPUs.In this paper, we provide a detailed characterization of the power, energy, and temperature of edge GPUs by designing micro-kernels for instruction-level power analysis and executing a real-world application for application-level power characterization.From the observations, we discuss the limitations of existing power simulators and also suggest power optimization methods.
Authors
- K.K. Chau
- Hyeran Jeon (ORCID: https://orcid.org/0000-0002-1767-8198)
- Mujahid Al Rafi
Institutions
- University of California, Merced (US)
Publication Details
- Published
- 2025-03-01
- DOI
- https://doi.org/10.1145/3725798.3725806
- Citations
- 2
- Primary Topic
- Parallel Computing and Optimization Techniques
- Type
- article
- Field-Weighted Citation Impact
- 4.54