Abstract: Recent years saw an increasing success in the application of deep learning methods across various domains and for tackling different problems, ranging from image recognition and classification to text processing and speech recognition. In this paper we propose, discuss, and validate a black box approach to model the execution time for training convolutional neural networks (CNNs), with a particular focus on deployments on general purpose graphics processing units (GPGPUs). We demonstrate that our approach is generally applicable to a
variety of CNN models and different types of GPGPUs with high accuracy. The proposed method can support with great precision (within 5% average percentage error) the management of production environments.