Using Analytical Performance/Power Model and Fine-Grained DVFS to Enhance AI Accelerator Energy Efficiency
Recent advancements in deep learning have significantly increased AI processors' energy consumption, which is becoming a critical factor limiting AI development. Dynamic Voltage and Frequency Scaling (DVFS) stands as a key method in power optimization. However, due to the latency of DVFS control in AI processors, previous works typically apply DVFS control at the granularity of a program's entire duration or sub-phases, rather than at the level of AI operators.
Authors
- Wanchun Dou (ORCID: https://orcid.org/0000-0003-4833-2023)
- Jian He
- Jingyi Zhang (ORCID: https://orcid.org/0000-0003-1909-6332)
- Xiaoliang Wang (ORCID: https://orcid.org/0000-0002-3410-8621)
- Yan-Lin Liu
- Guihai Chen (ORCID: https://orcid.org/0000-0002-6934-1685)
- Chen Tian (ORCID: https://orcid.org/0000-0003-2710-7628)
- Yijia Zhang (ORCID: https://orcid.org/0000-0001-6925-7777)
- Xiaoxin Xu
- Bingqiang Wang (ORCID: https://orcid.org/0009-0004-4964-3258)
- Zibo Wang (ORCID: https://orcid.org/0009-0009-0499-5328)
- Zhiheng Hu
- F. L. Wei
Institutions
- Huawei Technologies (China) (CN)
- Peng Cheng Laboratory (CN)
- Nanjing University (CN)
Publication Details
- Published
- 2025-02-06
- DOI
- https://doi.org/10.1145/3669940.3707231
- Citations
- 4
- Primary Topic
- Radiation Effects in Electronics
- Type
- article
- Field-Weighted Citation Impact
- 2.92