GPU usage optimization for the image processing in Xavier NX

I’m using the Xavier NX som that flashed with l4t 32.6.1 for streaming the MIPI cameras. In this setup, I have observed the GPU usage is quite high than usual when starting the stream from cameras and loading the frames to the GL textures parallelly for the image processing.

So, primarily I want to confirm this observed GPU usage in attached tegrastats_log.txt is the expected and optimized value for the process given in the enclosed code snippet.

Also, since I have to perform some image processing additionally after loading the frames to GL textures, I am looking for the most optimized GPU usage. But just loading the camera frames to GL textures from PBO show the GPU usage than expected. Is there any other way to optimize this GPU usage in Xavier NX?

Here, I have enclosed the tegrastats log and code snippet which has the APIs that used to load the frames in GL textures from PBO.
tegrastats_log.txt (124.4 KB)
gl_load_code_snippet.txt (6.8 KB)

It is expected to have certain GPU usage for the GL function calls. Please run sudo jetson_clocks to get optimal performance.

And please run sudo tegrastats to get all information. Without sudo, the frequency of EMC and GPU is not printed.