Evaluation of High Density GPUs as Sustainable Smart City Infrastructure

Abstract

Internet of things (IoT) is driving the big data revolution in smart cities. In order to make informed, accurate and real-time decisions, smart cities have to invest in powerful computing infrastructure with the minimal total cost of ownership. Smart city infrastructure will need to process data from various scientific and engineering domains like weather variability, traffic management, disease control etc in real-time while keeping the operational costs to minimum. In this paper we build a case for using General Purpose GPUs (GPGPU) as an alternate to the traditional CPU based computing. Utilising the GPUs in development of smart city infrastructure is an attractive alternate as it provides an efficient computing capacity when compared with traditional CPU only solutions. However, we find that naive deployment of applications on high-density GPUs results in lower scalability and performance. We show that designing a NUMA and GPU affinity aware parallel execution model can lead to substantial speed-ups. Our results show that smart cities can save over 45% in infrastructure power and over 90% in data centre space if high-density GPU solutions are used.