2020/10/13 by Théo Benoit-Cattin, Benoit-Cattin, Théo, Delia Velasco-Montero +3 · 3 citations
Computer Science · Engineering · #Advanced Memory and Neural Computing #Advanced Neural Network Applications #CCD and CMOS Imaging Sensors #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences
paper · pdf · doi:10.48550/arxiv.2010.06291
openalex publication_date 2020/10/13 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Many application scenarios of edge visual inference, e.g., robotics or\nenvironmental monitoring, eventually require long periods of continuous\noperation. In such periods, the processor temperature plays a critical role to\nkeep a prescribed frame rate. Particularly, the heavy computational load of\nconvolutional neural networks (CNNs) may lead to thermal throttling and hence\nperformance degradation in few seconds. In this paper, we report and analyze\nthe long-term performance of 80 different cases resulting from running 5 CNN\nmodels on 4 software frameworks and 2 operating systems without and with active\ncooling. This comprehensive study was conducted on a low-cost edge platform,\nnamely Raspberry Pi 4B (RPi4B), under stable indoor conditions. The results\nshow that hysteresis-based active cooling prevented thermal throttling in all\ncases, thereby improving the throughput up to approximately 90% versus no\ncooling. Interestingly, the range of fan usage during active cooling varied\nfrom 33% to 65%. Given the impact of the fan on the power consumption of the\nsystem as a whole, these results stress the importance of a suitable selection\nof CNN model and software components. To assess the performance in outdoor\napplications, we integrated an external temperature sensor with the RPi4B and\nconducted a set of experiments with no active cooling in a wide interval of\nambient temperature, ranging from 22 \degC to 36 \degC. Variations up to\n27.7% were measured with respect to the maximum throughput achieved in that\ninterval. This demonstrates that ambient temperature is a critical parameter in\ncase active cooling cannot be applied.\n