Skip to Main Content
Most factory automation and motion control systems rely on a cyclic approach, where a controller device periodically inquires and sends commands to decentralized I/O peripherals. From a general viewpoint, the shorter the period of the cycle the better the accuracy of the control system. However, reducing the cycle duration implies that more complex (and expensive) solutions have to be adopted. In this paper, an inexpensive computing platform for control applications is described, which consists of a conventional PC equipped with a CUDA-based GPU and an Ethernet interface, and its performance is evaluated. We show that noticeable improvements can be actually obtained, as long as control algorithms can be parallelized effectively.