Why is my GPU usages not going up with cuda? ( only around 10% usage)

Hi. I am very new to CUDA programming. And I have Quadro K5200 8GB.

I am playing around with OpenCV geometric image transformation function,
Mostly to warp perspective on my 1080P video input stream.

And for sure, comparing to non-CUDA version,
applying CUDA made a huge difference on processing time.

However, once I monitored my GPU and CPU usage of my CUDA versions, it was very interesting.
(I used third party GPU monitor. GPU Shark v0.7.5)

One 1080P input processing,
GPU usage 6%, GPU memory 400MB
CPU usage 6%, CPU memory 670MB

Two 1080P input processing,
GPU usage 10~13%, GPU memory 440MB
CPU usage 12%, CPU memory 978MB

Four 1080P input processing,
GPU usage 8~10%, GPU memory 450MB
CPU usage 21%, CPU memory 1.3GB

For sure, each cases got much faster with CUDA, but I am not sure

  1. Why did GPU usage not increase when jumping from Two to Four input.?
    (while my CPU usage increased significantly?).

  2. Why is GPU only 8~10 percent used? ( why not 80, 90… 99% ?? )

Am I doing something wrong? I feel like I am not fully using my GPU power?

Thank you