Analysing Edge Computing Devices for the Deployment of Embedded AI.

Area of Science:

• Computer Science
• Artificial Intelligence
• Network Engineering

Background:

• The proliferation of Internet of Things (IoT) devices generates vast amounts of data.
• Traditional cloud computing faces limitations in latency, efficiency, and real-time response for IoT applications.
• Edge computing emerges as a solution to process data closer to the source, addressing cloud limitations.

Purpose of the Study:

• To analyze and compare the performance of various Edge Computing devices for deploying artificial intelligence algorithms.
• To evaluate the impact of artificial intelligence accelerators, specifically Tensor Processing Units (TPUs), on edge device performance.
• To guide the selection of optimal edge devices based on specific AI requirements.

Main Methods:

• Conducting a detailed experiment comparing multiple edge devices, AI models, and performance metrics.
• Deploying artificial intelligence algorithms on selected edge computing hardware.
• Observing and measuring the performance of artificial intelligence accelerators, such as TPUs.

Main Results:

• The Jetson Nano demonstrates strong performance when utilizing only its CPU.
• The integration of a Tensor Processing Unit (TPU) significantly enhances the performance of edge devices.
• Specific performance gains vary depending on the edge device, AI model, and accelerator used.

Conclusions:

• Edge computing, combined with AI, offers a powerful solution for real-time data processing and autonomous decision-making.
• While CPU-based processing on devices like the Jetson Nano is viable, AI accelerators like TPUs provide substantial performance improvements.
• The choice of edge device and the inclusion of AI accelerators are critical for meeting demanding AI application requirements at the network edge.