When running the detectnet program or our own custom python code which is based on detectnet, we can see a high CPU usage ~90% while GPU usage keeps fluctuating between 40 to 80%. Does that means we are not fully using the GPU? How come CPU usages are high if all the inferencing is done on GPU?
Attached is the screenshot. I am using Jetson Nano in 5W mode.
Usually, the CPUs are occupied by some pre-processing and memory copy.
You might meet some data bottleneck so the GPU utilization cannot reach 99%.
(GPU is waiting for data input sometimes)