Study on Intelligent Heterogeneous Computing Technology for Reliable-critical Application

Abstract

Intelligent functions need higher compute ability, and the deployment process is more complicated. In scenarios where power consumption, size, and weight are limited, neither CPU nor ASCI cannot meet computing power requirements, and complex computing requirements are difficult to implement with a single FPGA. Further, on strict process control and cold start time put forward higher requirements for the use of accelerated devices such as GPUs with operating system (OS). We developed a prototype of a heterogeneous computing platform including development methods. Comparing the performance indicators of FPGA solution, the average time to calculate a single image was increased from 91.5ms/picture to 13.7ms/picture, an increase in the ratio 668%.