Benchmark Analysis of Jetson TX2, Jetson Nano and Raspberry PI using Deep-CNN

Hardware, low power consumption, high accuracy and performance are crucial factors for deep learning applications. High level graphics processing units (GPU) are commonly used in high performance deep learning applications. However, it is a lot in terms of cost and power consumption to build a high-performance platform. In this study, performances of single-board computers in NVIDIA Jetson Nano, NVIDIA Jetson TX2 and Raspberry PI4 through CNN algorithm created by using fashion product images dataset are compared. 2D CNN model has been developed so as to classify 13 different fashion products in tests. Data set is comprised of 45K pictures. Parameters for performance analysis has been defined as consumption (GPU, CPU, RAM, Power), accuracy and cost. Data set is divided into parts of 5K, 10K, 20K, 30K and 45K in training and test of the model in order to expand on the differences of single-board computers. Eventually, performance of the embedded system boards in different data set in CNN algorithm is analyzed. It is, thus, aimed to attain high accuracy preference by minimum hardware requirements in deep learning applications.