Evaluating the Energy Efficiency of Deep Convolutional Neural Networks on CPUs and GPUs

By Da Li, Xinbo Chen, Michela Becchi, Ziliang Zong. 2016

176 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Electrical and Electronic Engineering, Hardware and Architecture, Efficiency & Compression, ML Systems & Hardware

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

1. Anne-Laure Ligozat
2. Emma Strubell
3. Benjamin Kanding
4. Lasse F. Wolff Anthony
5. Raghavendra Selvan
6. Andrew McCallum
7. Ananya Ganesh
8. Kaushik Roy

In Electrical and Electronic Engineering

1. Nitin Rathi
2. Andrew McCallum
3. Anne-Laure Ligozat
4. Benjamin Kanding
5. Lasse F. Wolff Anthony
6. Kaushik Roy
7. Raghavendra Selvan
8. Yonghong Tian

In Hardware and Architecture

1. Kaushik Roy

In Efficiency & Compression

1. Andrew McCallum
2. Benjamin Kanding
3. Lasse F. Wolff Anthony
4. Kaushik Roy
5. Raghavendra Selvan
6. Nitin Rathi

In ML Systems & Hardware

1. Andrew McCallum
2. Kaushik Roy

Citations in the corpus

• Energy and Policy Considerations for Deep Learning in NLP (2019, background): “Li et al. (2016) present a detailed study of the energy use required for training and inference in popular convolutional”
• Unraveling the hidden environmental impacts of AI solutions for environment (2021, background): “Premises of such an approach were already present in [23] for CNN with less meaningful metrics (energy per image or power with no indications on the global duration).”
• DIET-SNN: A Low-Latency Spiking Neural Network With Direct Input Encoding and Leakage and Threshold Optimization (2021, background): “The skyrocketing performance and success of multilayer ANNs came at a significant power and energy cost [6].”
• Carbontracker: Tracking and Predicting the Carbon Footprint of Training Deep Learning Models (2020, background): “Power management techniques like dynamic voltage and frequency scaling (DVFS) can further help conserve energy consumption (Li et al., 2016) and for some models even reduce time to reach convergence (Tang et al., 2019).”