The premise is correct cuda and cudnn, as well as the corresponding platform installation.
1、caffe
caffe there is a command-line argument: device_query can view the information of a specified GPU
Such as: the command line input caffe device_query -gpu 0
2, pytorch
Enter the following command directly in the python environment, it displays true it means you can call gpu
import torch
print (torch.cuda.is_available())
3、tensorflow
Enter the following command at the command line: If the available information is displayed GPU said GPU may be used to accelerate, if only the display cpu, can not call the GPU
import tensorflow
from tensorflow.python.client import device_lib
print(device_lib.list_local_devices())
Here is no case of GPU can be used:
------------- ----------------------------------- additional content ------------------------------
Finally the way, you can look at all those processes using the information in the call GPU, GPU view with a command:
nvidia-smi
If the GPU acceleration with tensorflow or pytorch platform, a red python process will occur in this region
If caffe platform will be displayed caffe process: