The Detailed Data on the Neural Compute Stick Acceleration Performance

The four trends such as real-time, intelligence security and privacy have spawned the rise of edge computing and front-end intelligence. The Neural Computing stick(NCS) can move calculations from the cloud to the terminal, and provide dedicated deep neural network acceleration for terminal AI devices. However, there is currently no detailed data comparison of the NCS acceleration performance. This paper compares the computational speed of convolutional neural networks on the Raspberry Pi in the presence or absence of the NCS. In addition, we further compare the calculated speeds of the CPU, GPU, and NCS through the vehicle model. The result shows that the NCS can achieve 4 to 6 times acceleration in neural networks.