Computing Platform
Albert Chun Chen Liu, Oscar Ming Kin Law, Iain Law · 2022
This chapter discusses different artificial intelligence computing platforms, including Intel central processing unit (CPU), Nvidia graphics processing unit (GPU), Google tensor processing unit (TPU) – cloud servers, and Kneron neural processing unit – Edge artificial intelligence (AI) system. Traditionally, the CPU is targeted for general-purpose computation; it is evolved from single instruction single data architecture to single instruction multiple data. GPU consists of massively parallel Processing Elements designed for computer graphics and image processing. In recent years, it is also targeted for artificial intelligent development, especially for neural network training. Google successfully deployed TPU to support the growing cloud server artificial intelligence demand in 2013. Due to the surge of data, Edge AI is emerged to address the cloud limitations, data latency, network bandwidth, and power dissipation Edge AI is located closed to the mobile device to offer real-time processing, which reduces the data transfer and resolves the bandwidth bottleneck.