In the paper we present performance-energy trade-off investigation of training Deep Convolutional Neural Networks for image recognition. Several representative and widely adopted network models, such as Alexnet, VGG-19, Inception V3, Inception V4, Resnet50 and Resnet152 were tested using systems with Nvidia Quadro RTX 6000 as well as Nvidia V100 GPUs. Using GPU power capping we found other than default configurations minimizing three various metrics: energy (E), energy-delay product (EDP) as well as energy-delay sum (EDS) which resulted in considerable energy savings, with a low to medium performance loss for EDP and EDS. Specifically, for Quadro 6000 and minimization of E we obtained energy savings of 28.5%–32.5%, for EDP 25%–28% of energy was saved with average 4.5%–15.4% performance loss, for EDS (k = 2) 22%–27% of energy was saved with 4.5%–13.8% performance loss. For V100 we found average energy savings of 24%–33%, for EDP energy savings of 23%–27% with corresponding performance loss of 13%–21% and for EDS (k = 2) 23.5%–27.3% of energy was saved with performance loss of 4.5%–13.8%.
Authors
Additional information
- DOI
- Digital Object Identifier link open in new tab 10.1007/978-3-031-08751-6_48
- Category
- Aktywność konferencyjna
- Type
- publikacja w wydawnictwie zbiorowym recenzowanym (także w materiałach konferencyjnych)
- Language
- angielski
- Publication year
- 2022
