Use conda create a new virtual environment
Input conda create -n intelligent-judge python = 3.6
Create a python version 3.6 name is intelligent-judge virtual environment
View card information
Enter lspci | grep -i vga graphics card to view information
It is clear that NVIDIA Quadro GP100
CUDA configuration and CUDNN
Was originally going to download cuda and cudnn the results of a look, on the server already has a ready-made
Entry
cat /usr/local/cuda/version.txt，
with
cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2
You can see cuda version is 9.0.176, cudnn is version 7.5.0
Therefore the corresponding version, decided to install tensorflow gpu-1.70version
Configuration package python in a virtual environment
Input source activate intelligent-judge activate the virtual environment
Input pip install tensorflow-gpu == 1.7.0
Successful installation
Test file from remote repository pull
Running problems
Tips can not find cuda
Adjusted in accordance with the following article
https://blog.csdn.net/qq_34374211/article/details/81018320
Successfully run the test file
Run training documents, but also out of the unitary moths
Check out the official API
The original is in the tf.random.normal tensorflow 1.7.0 should be replaced with tf.random_normal
After running change
He has begun training a pleasure