Field-Programmable Gate Arrays (FPGAs) are well-suited for field deployment of deep learning models as they offer flexible reconfiguration, support for parallel processing, and efficient power consumption. In this paper, we present the first documented implementation of YOLOv8, a state-of-the-art object detection model, on a real MPSoC FPGA board, specifically the ZCU102. While previous works focused on older YOLO versions (e.g., YOLOv6, YOLOv7) and simulation-based evaluations, our approach, starting from a PyTorch-trained YOLOv8n model, proposes a full deployment of YOLOv8n, including model inspection, quantization, and compilation of the xmodel format required by the deep learning processing unit. The final implementation achieves real-time performance at 30 frames per second with only 5.28 W of power consumption, significantly outperforming previous simulation-based results reported in the literature.