1. 安装anaconda/miniconda

pip install anaconda/miniconda

2. 创建环境

conda create -n [envs_name]

3. 安装tensorflow(cuda+cudnn会被捆绑安装)

3.1 激活环境

source activate [envs_name]

3.2 在该环境下安装tensorflow-gpu版本

conda install tensorflow-gpu

4. 测试tensorflow

进入python环境： $ python

加载tensorflow:

>> import tensorflow as tf

>> hello = tf.constant('hello,world!')

>> sess = tf.Session()

>> print(sess.run(hello))

打印输出b'hello,world!',则表示成功。

注意：

如出现

ImportError: libnvidia-fatbinaryloader.so.375.51: cannot open shared object file: No such file or directory Failed to load the native TensorFlow runtime.

则可以通过下面两种方式设置环境变量解决

第一种，临时修改环境变量

export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/lib:/usr/local/cuda/lib64:/usr/lib/x86_64-linux-gnu:/usr/lib/nvidia-375 export LIBRARY_PATH=${LIBRARY_PATH}:/usr/local/cuda/lib64:/usr/lib/x86_64-linux-gnu:/usr/lib/nvidia-375

第二种，永久修改，每次打开终端都会被读取

打开~/.bashrc文件，将下面两行写入并保存

export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/lib:/usr/local/cuda/lib64:/usr/lib/x86_64-linux-gnu:/usr/lib/nvidia-375 export LIBRARY_PATH=${LIBRARY_PATH}:/usr/local/cuda/lib64:/usr/lib/x86_64-linux-gnu:/usr/lib/nvidia-375

参考： https://github.com/tensorflow/tensorflow/issues/9071