Skip to main content Skip to main navigation


Fast and Accurate: Machine Learning Techniques for Performance Estimation of CNNs for GPGPUs

Christopher Metz; Mehran Goli; Rolf Drechsler
In: 5th Workshop on Parallel AI and Systems for the Edge (PAISE). Workshop on Parallel AI and Systems for the Edge (PAISE-2023), located at IPDPS 2023, April 19 - May 19, St. Petersburg, FL, USA, 2023.


High performance and on-time calculations of Machine Learning (ML) algorithms are essential for emerging technologies such as autonomous driving, Internet of Things (IoT) or edge computing. One of the major algorithms used in such systems is Convolutional Neural Networks (CNNs), which require high computational resources. That leads designers to leverage ML accelerators like GPGPUs to meet design constraints. However, selecting the most appropriate accelerator requires Design Space Exploration (DSE), which is usually time-consuming and needs high manual effort. In this paper, we present a novel automated approach, enabling designers to fast and accurately estimate the performance of CNNs for GPGPUs in the early stage of the design process. The proposed approach uses static analysis for feature extraction and Decision Tree regression analysis for the performance estimation model. Experimental results demonstrate that our approach can predict CNNs performance with an absolute percentage error of 5.73% compared to the actual hardware.