ubuntu 18.04 安装tensorflow-gpu-1.13.1

0. 本机环境

Ubuntu	18.04
显卡	GeForce GT 730M
Compute Capability	3.0
CUDA核心	384
NVCUDA.DLL	NVIDIA CUDA 9.2.156 driver

1. 下载安装包

1.1 CUDA

版本	下载地址	下载文件
CUDA10.0	https://developer.nvidia.com/cuda-10.0-download-archive	cuda_10.0.130.1_linux.run cuda_10.0.130_410.48_linux.run

1.2 cuDNN

版本	下载地址	下载文件
cuDNN v7.6.5 (November 5th, 2019), for CUDA 10.0	https://developer.nvidia.com/rdp/cudnn-download	libcudnn7_7.6.5.32-1+cuda10.0_amd64.deb libcudnn7-dev_7.6.5.32-1+cuda10.0_amd64.deb libcudnn7-doc_7.6.5.32-1+cuda10.0_amd64.deb

2. 安装Nvidia驱动

ubuntu 18.04 可以在Software & Update里面选择安装最新的nvidia驱动，安装后会自动禁用自带nouveau模块。

3. 安装CUDA 10工具包

针对CUDA9.0，必须将GCC降级为gcc5

$ sudo apt install gcc-5 g++-5
$ sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-5 50
$ sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-5 50

$ sudo sh cuda_10.0.130.1_linux.run
$ sudo sh cuda_10.0.130_410.48_linux.run

4. 设置环境变量

$ sudo vi /etc/ld.so.conf.d/cuda.conf

加入两行:

/usr/local/cuda/lib64
/usr/local/cuda/extras/CUPTI/lib64

$ sudo vi /etc/profile

加入两行:

export CUDA_HOME=/usr/local/cuda/bin
export PATH=$PATH:$CUDA_HOME

5. 测试CUDA 10安装

export PATH=$PATH:/usr/local/cuda/bin

$ cd ~/NVIDIA_CUDA-10.0_Samples/1_Utilities/deviceQuery
$ make
$ ./deviceQuery
# Result = PASS 成功

$ cd ../bandwidthTest
$ make
$ ./bandwidthTest
#Result = PASS 成功

6. 安装cuDNN 7

$ sudo dpkg -i libcudnn7_7.6.5.32-1+cuda10.0_amd64.deb
$ sudo dpkg -i libcudnn7-dev_7.6.5.32-1+cuda10.0_amd64.deb
$ sudo dpkg -i libcudnn7-doc_7.6.5.32-1+cuda10.0_amd64.deb

7. 测试cuDNN安装

$ cp -r /usr/src/cudnn_samples_v7/ $HOME/Work
$ cd  $HOME/Work/cudnn_samples_v7/mnistCUDNN
$ make clean && make

$ ./mnistCUDNN
#Test passed! 成功

8. 安装tensorflow-gpu

在https://tensorflow.google.cn/install/source#linux查看tensorflow对应CUDA及cuDNN版本：

tensorflow_gpu版本	Python 版本	编译器	编译工具	cuDNN	CUDA
tensorflow_gpu-1.13.1	2.7、3.3-3.6	GCC 4.8	Bazel 0.19.2	7.4	10.0
tensorflow_gpu-1.12.0	2.7、3.3-3.6	GCC 4.8	Bazel 0.15.0	7	9

$ sudo apt-get install python3-pip python3-dev
$ sudo pip3 install tensorflow-gpu==1.13.1 -i https://pypi.tuna.tsinghua.edu.cn/simple

9. 测试tenorflow-gpu安装

$ vi test.py

import tensorflow as tf

hello = tf.constant('Hello, TensorFlow!')
sess = tf.Session()
print(sess.run(hello))

$ python3 test.py

10. 卸载

10.1 卸载tensorflow

sudo pip3 uninstall protobuf
sudo pip3 uninstall tensorflow-gpu

10.2 卸载cuDNN

sudo apt-get remove --purge libcudnn7_7.6.5.32-1+cuda10.0_amd64.deb libcudnn7-dev_7.6.5.32-1+cuda10.0_amd64.deb libcudnn7-doc_7.6.5.32-1+cuda10.0_amd64.deb

10.3 卸载CUDA

sudo /usr/local/cuda-10.0/bin/uninstall_cuda_10.0.pl
sudo rm -rf /usr/local/cuda-10.0/

参考：
https://blog.csdn.net/wshixinshouaaa/article/details/83272282
https://www.jianshu.com/p/5ce71fe54d95