CentOS 7.2安装1080TI驱动+CUDA+Tensorflow记录

1 安装1080TI显卡驱动

1 .1安装内核源码包

yum -y install gcc kernel-devel "kernel-devel-uname-r == $(uname -r)"

1.2 禁用nouveau 模块

echo -e "blacklist nouveau\noptions nouveau modeset=0" > /etc/modprobe.d/blacklist.conf

mv /boot/initramfs-$(uname -r).img /boot/initramfs-$(uname -r).img.bak
dracut /boot/initramfs-$(uname -r).img $(uname -r)

1.3 下载对应显卡驱动及执行安装脚本

地址：https://www.geforce.cn/drivers/results/121083

 sh NVIDIA-Linux-x86_64-390.59.run 

1.4 测试

[root@103 home]# nvidia-smi
Sat Jul  7 03:34:44 2018       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 390.59                 Driver Version: 390.59                    |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 108...  Off  | 00000000:02:00.0 Off |                  N/A |
| 27%   48C    P0    59W / 250W |      0MiB / 11178MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
|   1  GeForce GTX 108...  Off  | 00000000:03:00.0 Off |                  N/A |
| 23%   40C    P0    52W / 250W |      0MiB / 11178MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
|   2  GeForce GTX 108...  Off  | 00000000:83:00.0 Off |                  N/A |
| 23%   27C    P0    56W / 250W |      0MiB / 11178MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

1.5遇到问题（内核源码包不对应）

执行安装驱动命令时提示：

error：unable to find the kernel source tree for the currently running kernel. please 
make sure you have installed the kernel source files for your kernel and that htey 
are properly configured; on red hat linux system, for example, be sure you have 
the 'kernel-source' or 'kernel-devel' RPM installed. if you know the correct kernel 
source files are installed ,you may specify the kernel source path with the '--
kernel-source-path' command line option.

按提示指定内核源码目录：

sh NVIDIA-Linux-x86_64-390.59.run --kernel-source-path=/usr/src/kernels/3.10.0-862.6.3.el7.x86_64/

接着提示出错：

ERROR: Unable to load the kernel module 'nvidia.ko'.  This happens most
       frequently when this kernel module was built against the wrong or
       improperly configured kernel sources, with a version of gcc that differs
       from the one used to build the target kernel, or if a driver such as
       rivafb/nvidiafb is present and prevents the NVIDIA kernel module from
       obtaining ownership of the NVIDIA graphics device(s).
       
       Please see the log entries 'Kernel module load error' and 'Kernel
       messages' at the end of the file '/var/log/nvidia-installer.log' for
       more information.

原因分析：

Loaded plugins: fastestmirror
Loading mirror speeds from cached hostfile
 * base: mirror.sunnyvision.com
 * elrepo: hkg.mirror.rackspace.com
 * extras: ftp.cuhk.edu.hk
 * updates: ftp.cuhk.edu.hk
Package gcc-4.8.5-28.el7_5.1.x86_64 already installed and latest version
No package kernel-devel-uname-r == 3.10.0-327.el7.x86_64 available.
Resolving Dependencies
--> Running transaction check
---> Package kernel-devel.x86_64 0:3.10.0-862.6.3.el7 will be installed
--> Finished Dependency Resolution

Dependencies Resolved

=====================================================================================================================================================================================
 Package                                     Arch                                  Version                                              Repository                              Size
=====================================================================================================================================================================================
Installing:
 kernel-devel                                x86_64                                3.10.0-862.6.3.el7                                   updates                                 16 M

Transaction Summary
=====================================================================================================================================================================================
Install  1 Package

Total download size: 16 M
Installed size: 37 M
Downloading packages:
Delta RPMs disabled because /usr/bin/applydeltarpm not installed.

因为使用时CentOS7.2，内核版本为:

Linux 103.215.190.172 3.10.0-327.el7.x86_64 #1 SMP Thu Nov 19 22:10:57 UTC 2015 x86_64 x86_64 x86_64 GNU/Linux

但是源已经更新了，没有对应的内核源码包，安装的是最新的

kernel-devel.x86_64 0:3.10.0-862.6.3.el7

执行安装驱动脚本时导致找不到内核源码目录，解决办法是卸载yum安装的内核源码包，手动下载对应的rpm源码，rpm安装

yum remove  kernel-devel 
rpm -ivh kernel-devel-3.10.0-327.el7.x86_64.rpm 

2 安装CUDA

2.1 安装 cuda(安装9.2发现tensorflow不支持，改回了9.0)

下载地址：https://developer.nvidia.com/cuda-downloads

选择runfile（rpm不知为何老是装不上）。

执行

chmod +x cuda_9.2.88_396.26_linux.run 
sh cuda_9.2.88_396.26_linux.run

2.2 测试

cd /usr/local/cuda-8.0/samples/1_Utilities/deviceQuery
sudo make
./deviceQuery

输出：

2.3 遇到问题

测试时出现：

[root@103 deviceQuery]# ./deviceQuery 
./deviceQuery Starting...

 CUDA Device Query (Runtime API) version (CUDART static linking)

cudaGetDeviceCount returned 38
-> no CUDA-capable device is detected
Result = FAIL

原因分析：

开始安装的是cuda8.0，估计版本比较旧（与驱动版本相比），后来安装了cuda9.0（最新9.2，但是tensorflow不支持），问题解决

3 安装cudnn

3.1 下载

注意一定要下载cuda对应版本的cudnn

下载地址：https://developer.nvidia.com/rdp/cudnn-download

3.2 拷贝文件到cuda

ktar xvf cudnn-9.2-linux-x64-v7.1.tar
sudo cp cuda/include/cudnn.h /usr/local/cuda/include
sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64/ -d
sudo chmod a+r /usr/local/cuda/include/cudnn.h
sudo chmod a+r /usr/local/cuda/lib64/libcudnn*

4 安装Tensorflow

4.1 安装带GPU加速

pip install --upgrade https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-1.8.0-cp27-none-linux_x86_64.whl 

4.2 测试

import tensorflow as tf
import numpy as np

# 使用 NumPy 生成假数据(phony data), 总共 100 个点.
x_data = np.float32(np.random.rand(2, 100)) # 随机输入
y_data = np.dot([0.100, 0.200], x_data) + 0.300

# 构造一个线性模型
# 
b = tf.Variable(tf.zeros([1]))
W = tf.Variable(tf.random_uniform([1, 2], -1.0, 1.0))
y = tf.matmul(W, x_data) + b

# 最小化方差
loss = tf.reduce_mean(tf.square(y - y_data))
optimizer = tf.train.GradientDescentOptimizer(0.5)
train = optimizer.minimize(loss)

# 初始化变量
init = tf.initialize_all_variables()

# 启动图 (graph)
sess = tf.Session()
sess.run(init)

# 拟合平面
for step in xrange(0, 201):
    sess.run(train)
    if step % 20 == 0:
        print step, sess.run(W), sess.run(b)

# 得到最佳拟合结果 W: [[0.100  0.200]], b: [0.300]

输出：

4.3 问题

4.3.2 cuda版本太新，安装cuda9.0

[root@103 home]# python test.py 
Traceback (most recent call last):
  File "test.py", line 1, in 
    import tensorflow as tf
  File "/usr/lib/python2.7/site-packages/tensorflow/__init__.py", line 24, in 
    from tensorflow.python import pywrap_tensorflow  # pylint: disable=unused-import
  File "/usr/lib/python2.7/site-packages/tensorflow/python/__init__.py", line 49, in 
    from tensorflow.python import pywrap_tensorflow
  File "/usr/lib/python2.7/site-packages/tensorflow/python/pywrap_tensorflow.py", line 74, in 
    raise ImportError(msg)
ImportError: Traceback (most recent call last):
  File "/usr/lib/python2.7/site-packages/tensorflow/python/pywrap_tensorflow.py", line 58, in 
    from tensorflow.python.pywrap_tensorflow_internal import *
  File "/usr/lib/python2.7/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 28, in 
    _pywrap_tensorflow_internal = swig_import_helper()
  File "/usr/lib/python2.7/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 24, in swig_import_helper
    _mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description)
ImportError: libcublas.so.9.0: cannot open shared object file: No such file or directory

4.3.2 numpy版本太新或太旧

卸载numpy

ValueError: numpy.dtype has the wrong size, try recompiling

再次安装：

 pip install --upgrade https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-1.8.0-cp27-none-linux_x86_64.whl --ignore-installed six