This study focuses on people counting in a video stream captured from a fixed camera. Aforementioned counting process is implemented by graphical processing unit (GPU) programming real-timely. For this reason, two video streams with different resolution and two different NVIDIA graphic cards are used. For all combinations of these video streams and graphic cards, the number of people are obtained in the video streams and they are compared with regard to performances. Consequently, it is examined that real-time people counting process can be successfully implemented by compute unified device architecture (CUDA) programming on NVIDIA graphic cards.