Ubuntu 16.04 安装 Tensorflow Gpu版本

原文地址：blogof33.com/post/2/

Tensorflow有Pip,Docker, Virtualenv, Anaconda 或 源码编译的方法安装 ,本文中采用Pip安装。

由于国内中文教程中有关Pip安装的不多，官方中文文档也有一些顺序上的错误，语义的模糊，按照上面的来很容易出错。因此我就在这里写一篇教程吧。

注：建议 Ubuntu 16.04版本

配置：

Linux Distribution: Ubuntu 16.04 64位

Cpu: Intel Core i5 6300HQ

Gpu:GTX 960M

Python:2.7

由于笔者是装的双系统，用UEFI引导的没有关Security Boot(安全启动)，因此出了一点问题。

1.开机进入Bios,关闭Security Boot

2.系统设置(setting)->软件和更新->附加驱动

• 先打开terminal输入：

  sudo apt-get update

• 而后将显卡驱动选择为NVIDIA的显卡驱动(更改之后须要等待一段时间。)

3.安装CUDA8

• 去cuda toolkit 8.0下载，如图，选择deb(local)安装，网上有些说神马必定要选runfile，其实deb也很简单。
• 注：cuda 7.5只有ubuntu15.04和14.04版本，因此要用cuda 8.0

• 下载成功之后按照图中的命令操做便可安装完成。 安装完成之后，cuda默认安装在了/usr/local/cuda-8.0/目录处，而后: vim ~/.profile

  注：不少其余发行版是打开.bash_profile，ubuntu没有这个，而是.profile

  设置环境变量(加在文件末尾，每次登录后自动生效)： export PATH="$PATH:/usr/local/cuda-8.0" export LD_LIBRARY_PATH="/usr/local/cuda-8.0/lib64" 保存之后，继续在terminal中输入:： source ~/.profile #使更改的环境变量当即生效 nvidia-smi #测试是否配置成功 结果出现如下相识输出说明配置成功：

  

4.下降gcc版本

由于ubuntu16.04的gcc编译器是5.4.0，然而cuda8.0不支持5.0以上的编译器，所以须要降级，把编译器版本降到4.9：

sudo apt-get install g++-4.9
  sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-4.9 20
  sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-5 10
  sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-4.9 20
  sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-5 10
  sudo update-alternatives --install /usr/bin/cc cc /usr/bin/gcc 30
  sudo update-alternatives --set cc /usr/bin/gcc
  sudo update-alternatives --install /usr/bin/c++ c++ /usr/bin/g++ 30
  sudo update-alternatives --set c++ /usr/bin/g++
  gcc --version

复制代码

若是版本显示出来是4.9就成功了

5.安装深度学习库cuDNN

必定要下载cuDNN5.1!!!!,至少如今下载6.0版本后面是有问题的，我以前下载的6.0版本，后面导入tensorflow时一直报错，在gaythub上看见有人说6.0“don not work”而5.1能够,不得已退回到5.1，而后便成功了。

• 下载以前要注册一个NVIDA DEVELOPER账号，而后填三个问题就能够啦。如图选择cuDNN v5.1 Library for Linux：

  

  ​

• 下载完成之后，解压并拷贝 CUDNN 文件到 Cuda Toolkit 8.0安装路径下. 假设 Cuda Toolkit 8.0 安装在 /usr/local/cud-8.0(默认路径), 执行如下命令(若/usr/local/cuda-8.0/include目录不存在，先建立一个include目录):

tar xvzf cudnn-8.0-linux-x64-v5.1.tgz
      sudo cp cuda/include/cudnn.h /usr/local/cuda-8.0/include
      sudo cp cuda/lib64/libcudnn* /usr/local/cuda-8.0/lib64
      sudo chmod a+r /usr/local/cuda-8.0/include/cudnn.h /usr/local/cuda-8.0/lib64/libcudnn*
复制代码

6.Install and upgrade PIP

• 首先要确保安装和更新pip：

sudo apt-get install python-pip python-dev
     pip install --upgrade pip
复制代码

7.安装 Tensorflow Gpu enable python 2.7 版本

pip install --upgrade tensorflow-gpu

测试一下：

$python 
       Python 2.7.12 (default, Nov 19 2016, 06:48:10) 
       [GCC 5.4.0 20160609] on linux2
       Type "help", "copyright", "credits" or "license" for more information.
       >>> import tensorflow as tf
       >>> hello = tf.constant('Hello, TensorFlow!')
       >>> sess = tf.Session()
       2017-08-25 14:34:54.825013: W tensorflow/core/platform/cpu_feature_guard.cc:45] The 
       TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations. 2017-08-25 14:34:54.825065: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your 
       machine and could speed up CPU computations.
       2017-08-25 14:34:54.825081: W tensorflow/core/platform/cpu_feature_guard.cc:45] The 
       TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations. 2017-08-25 14:34:54.825093: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your 
       machine and could speed up CPU computations.
       2017-08-25 14:34:54.825105: W tensorflow/core/platform/cpu_feature_guard.cc:45] The 
       TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations. 2017-08-25 14:34:55.071951: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:893] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero 2017-08-25 14:34:55.072542: I tensorflow/core/common_runtime/gpu/gpu_device.cc:940] Found device 0 with properties: name: GeForce GTX 960M major: 5 minor: 0 memoryClockRate (GHz) 1.176 pciBusID 0000:01:00.0 Total memory: 1.95GiB Free memory: 1.31GiB 2017-08-25 14:34:55.072632: I tensorflow/core/common_runtime/gpu/gpu_device.cc:961] DMA: 0 2017-08-25 14:34:55.072695: I tensorflow/core/common_runtime/gpu/gpu_device.cc:971] 0: Y 2017-08-25 14:34:55.072730: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1030] Creating TensorFlow device (/gpu:0) -> (device: 0, name: GeForce GTX 960M, pci bus id: 0000:01:00.0) >>> print(sess.run(hello)) Hello, TensorFlow! 查看一下GPU开启状况： - "/cpu:0": The CPU of your machine. - "/gpu:0": The GPU of your machine, if you have one. - "/gpu:1": The second GPU of your machine, etc. 复制代码

>>> import tensorflow as tf
    >>> a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a')
    >>> b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b')
    >>> c = tf.matmul(a, b)
    >>> sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))
    >>> print(sess.run(c))
    MatMul: (MatMul): /job:localhost/replica:0/task:0/gpu:0
    2017-08-09 09:47:39.461702: I tensorflow/core/common_runtime/simple_placer.cc:847] MatMul: (MatMul)/job:localhost/replica:0/task:0/gpu:0
    b: (Const): /job:localhost/replica:0/task:0/gpu:0
    2017-08-09 09:47:39.461942: I tensorflow/core/common_runtime/simple_placer.cc:847] b: (Const)/job:localhost/replica:0/task:0/gpu:0
    a: (Const): /job:localhost/replica:0/task:0/gpu:0
    2017-08-09 09:47:39.461976: I tensorflow/core/common_runtime/simple_placer.cc:847] a: (Const)/job:localhost/replica:0/task:0/gpu:0
    [[ 22.  28.]
     [ 49.  64.]] 
复制代码

至此，Tensorflow安装成功。