Stalling time - GPU Load with multi-sources

hello, i have build a face detection model but when using 2 input sources i have noticed that for too many seconds GPU has a load 0% while pipeline is running. I can tell that GPU is not being well utilized and for other seconds GPU utilization does not exceed 20%. Is there a way to increase the GPU utilization?
When enabling the DLA to run the same model the GPU load is even higher compared to the model running on GPU only which weird specially that all layers are supported by DLA. Is there any justification for this?


Could you profile the application with nsys to locate the issue?

is there any other way to check? i can’t find the Nsight folder in opt/nvidia.
I’m using JP5.1


Please install it via the below command:

$ sudo apt-get install nsight-systems*


Thank you. After installing i have run ./nsys-ui and i was able to NVIDIA NSIGHT SYSTEMS install. But how to start profiling? i can only start new projects but how can i enable to start profiling my device.

Thank you