Voltile GPU utilization in caffe training

I use GPU for caffe training. My caffe algorithm has printed output as

I0814 12:27:37.772343 22016 caffe.cpp:217] Using GPUs 0
I0814 12:27:37.793123 22016 caffe.cpp:222] GPU 0: Tesla V100-PCIE-16GB
I0814 12:27:38.285532 22016 solver.cpp:65] Initializing solver from parameters: 
train_net: "/home/user/Projects/caffe-ssd/Numberplate/RregularNet/Net_19/train_0.prototxt"
test_net: "/home/user/Projects/caffe-ssd/Numberplate/RregularNet/Net_19/test_0.prototxt"

That means GPU is used in training and my GPU is Tesla V100.

I checked my nvidia-smi and Voltile GPU Utilization is quite low. Maximum I have seen is 17% only in training Caffe.

Is that normal or something not right?

Nvidia-smi screen shot is shown below.

enter image description here