I’m trying out my Yolov4-trained models on the Jetson Nano, and the fps is not enough for me.Jetson is running in nano MAXN mode.although I ran the jetson_clocks command, there was no increase in fps either.Why might it not have had some effect?Cuda gives 6-7fps on the yolov4 model in jetson_clocks and Maxn mode.That’s not enough for me either.is it normal that jetson_clocks is of no use to me?Or is there any other contribution it provides besides fps?If you have any ideas about this topic and how I can improve the fps further.
Hi @furkant, what the jetson_clocks script does is disables the DVFS governor (Dynamic Voltage and Frequency Scaling) and maximizes the clock speeds to their maximums as defined by the active nvpmodel (which you already have as MAXN). If your application is already heavily utilizing the CPU/GPU, then the clocks have probably already spun-up, which is why you would see negligible impact.
What I would recommend is looking into running your YOLOv4 model with TensorRT like here: https://jkjung-avt.github.io/tensorrt-yolov4/
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.