GPU memory usage using Matlab deep learning

Hi, I am using Matlab for fine tuning pre-trained CNN. The code works without any problems. However, when I monitor my GPU memory usage lively (using both GPU-Z and nvidia-smi in command line), I notice it varies between 4-7 GB. My GPU memory is 8 GB (GTX1070). Have any idea what is the reason behind that? The total running time is around 30 minutes for my code. I also observe much less running time on Ubuntu operation system when I tried to run the similar code on Keras.

Answer 1

[Joss Knight]

While you are training a network it needs a lot of working memory. The deeper the network, the more memory it needs. Which network are you training?

[Amirreza Mahbod]

I am tying na d training different networks such as ResNet, VGG16 and GoogleNet