将Keras（Tensorflow后端）与移动GPU（笔记本电脑）结合使用

Question

If I only use the Tensorflow code, the GPU usage rate is more than 80% and temperature rises very much. But if I use Kers, the usage rate drops to 15%. Also, using Keras does not reach the maximum clock of the GPU.

I tried with 980m, 1070 (laptop) and 960m, but the same result was obtained. ANN and CNN all have the same result.

Why is my GPU utilisation so low when using Kers on a mobile (laptop) GPU? What can I do to ensure I am getting maximum performance when using Keras on a laptop?

Answer 1

Maybe upgrade to Tensorflow 1.3. In TF 1.3 Keras is already included and there is no need to additionaly install keras.

In order to use the keras version included in TF do the following: Use for example

from tensorflow.contrib.keras.python.keras.models import Model

or

from tensorflow.contrib.keras.python.keras.layers import Input, Conv2D, Dropout

instead of

from keras.models import Model
from keras.layers import Input, Conv2D, Dropout

Maybe that solves the problem!