Profiling Energy Consumption of Deep Neural Networks on NVIDIA Jetson Nano
Stephan Holly, Alexander Wendt, Martin Lechner
Abstract
Improving the capabilities of embedded devices and accelerators for Deep Neural Networks (DNN) leads to a shift from cloud to edge computing. Especially for battery-powered systems, intelligent energy management is critical. In this work, we provide a measurement base for power estimation on NVIDIA Jetson devices. We analyze the effects of different CPU and GPU settings on power consumption, latency, and energy for complete DNNs as well as for individual layers. Furthermore, we provide optimal settings for minimal power and energy consumption for an NVIDIA Jetson Nano.
Topics & Concepts
Profiling (computer programming)Computer scienceEnergy consumptionArtificial neural networkLatency (audio)Embedded systemCloud computingPower consumptionDeep neural networksEfficient energy useEdge computingPower managementPower (physics)Operating systemElectrical engineeringEngineeringArtificial intelligenceTelecommunicationsQuantum mechanicsPhysicsAdvanced Memory and Neural ComputingEnergy Harvesting in Wireless NetworksParallel Computing and Optimization Techniques