Tensorflow + Ubuntu 16.04 + GTK780 + GIGABYTE-uefi DualBIOS 配置

1.DEL 进入BIOS设置

进入BIOS功能，将启动优先权各项全部关闭，F10保存，进入U盘启动模式，安装Ubuntu16.04

1.1 EFI引导分区 200M

1.2 swap area(交换空间）大约与本机内存相等

1.3 剩余空间全部用作Ext4日志文件系统， 挂在点选“/”

1.4 Device foor boot loader installation 选择在EFI引导分区下

2.预检

# lspci | grep -i nvidia

01:00.0 VGA compatible controller: NVIDIA Corporation GK110 [Geforce GTX 780] (rev a1)

01:00.1 Audio device:NVIDIA Corporation GK110 HDMI Audio (rev a1)

#uname -m && cat /etc/*release

x86_64

DISTRIB_ID=Ubuntu

DISTRIB_RELEASE=16.04

DISTRIB_CODENAME=xenial
DISTRIB_DESCRIPTION="Ubuntu 16.04.1 LTS"

NAME="Ubuntu"

VERSION="16.04.1 LTS (Xenial Xerus)"

ID=ubuntu

ID_LIKE=debian

PRETTY_NAME="Ubuntu 16.04.1 LTS"

VERSION_ID="16.04"

HOME_URL="http://www.ubuntu.com/"
SUPPORT_URL="http://help.ubuntu.com/"

BUG_REPORT_URL="http://bugs.launchpad.net/ubuntu/"

UBUNTU_CODENAME=xenial

#gcc --version

gcc (Ubuntu 5.4.0-6ubuntu1~16.04.02) 5.4.0 20160609
Copyright (C) 2015 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO

warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

3.安装cuda

3.1 安装NVIDIA驱动

Nouveau是由第三方为NVIDIA显卡开发的一个开源3D驱动，没能得到NVIDIA的认可与支持，不过确让Linux更容易的应对各种复杂的NVIDIA显卡环境，让用户安装完系统即可进入桌面并且有不错的显示效果，故很多Linux发行版默认集成了Nouveau驱动，在遇到NVIDIA显卡时默认安装。企业版的Linux更是如此，几乎所有支持图形界面的企业Linux发行版都将Nouveau收入其中。

关闭Nouveau

#sudo gedit /etc/modprobe.d/blacklist.conf  在文件后面加入blacklist nouveau 

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

#dracut -v /boot/initramfs-$(uname -r).img $(uname -r) （sudo apt-get install dracut）

下载驱动程序NVIDIA-Linux-x86_64-367.44.run

# sudo service lightdm stop  关闭图像界面，进入命令行

# sudo ./NVIDIA-Linux-x86_64-367.44.run

You appear to be running an X server; please exit X ...

# ps -e | grep X

# sudo kill Xorg   kill掉X进程

# sudo ./NVIDIA-Linux-x86_64-367.44.run

# sudo service lightdm start   返回图像界面

进入图像界面后报错：System program problem detected (未解决)

# nvidia-smi  检查驱动

3.2 安装CUDA

# sudo ./ cuda_7.5.18_linux.run --override

# sudo gedit ~/.bashrc 修改环境变量

export PATH=$PATH:/usr/local/cuda-7.5/bin

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64

#source ~/.bashrc

# sudo gedit /etc/ld.so.conf    修改动态链接

/usr/local/cuda/lib64

# sudo ldconfig

#  nvcc -V  检查CUDA

# cd /usr/local/cuda-7.5/samples

# sudo make测试CUDA

报错： unsupported GNU version! gcc versions later than 4.9 are not supported

gcc版本过高，修改CUDA配置文档，没有选择去降低gcc版本

# cd /usr/local/cuda/include/

# sudo cp host_config.h host_config.h.bak

# sudo gedit host_config.h 

if _GNUC_>4 || (_GNUC_ == 4 && _GNUC_MINOR_ > 9)  将两个4改为5

# cd /usr/local/cuda-7.5/samples

# sudo make  测试CUDA，通过

3.3 安装CUDNN

下载cudnn-7.0-linux-x64-v4.0-prod.tgz

# sudo cpinclude/cudnn.h /usr/local/cuda/include

# sudo cp lib64/libcudnn.*  /usr/local/cuda/lib

4.安装Anaconda

下载Anaconda2-4.1.1-Linux-x86_64

# sudo bash  Anaconda2-4.1.1-Linux-x86_64

PREFIX=/usr/anaconda2

installing: hdf5-1.8.16-0 ...

installing: mkl-11.3.3-0 ...

export PATH="/root/anaconda/bin:$PATH"    anaconda 自动添加环境变量

5.安装TensorFlow

# cd /usr/bin

# sudo rm -rf python

# sudo ln -s /usr/anaconda2/bin/python2.7 python

# pip install https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow-0.5.0-cp27-none-linux_x86_64.whl

# python; import tensorflow

报错：ImportError: libcudart.so.7.0: cannot open shared object file: No such file or directory

网上查到

#sudo ln -s /usr/local/cuda/lib64/libcudart.so.7.5 /usr/local/cuda/lib64/libcudart.so.7.0

运行 python convolutional.py

依旧报错：ImportError: libcudart.so.7.0: cannot open shared object file: No such file or directory

安装失败，经人提醒，有可能是TensorFlow版本太低

# sudo pip install --upgrade https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.8.0-cp27-none-linux_x86_64.whl

报错：Cannot remove entries from nonexistent file /usr/anaconda2/lib/python2.7/site-packages/easy-install.pth

几经尝试，放弃安装tensorflow-0.8.0

# sudo pip install tensorflow-0.10.0rc0-cp27-none-linux_x86_64.whl

报错：Loaded CuDNN library: 4007 (compatibility version 4000) but source was compiled with library 5103 (compatibility version 5100) ...

运行，cudnn版本报错，版本过低，应该装v5

下载cudnn7.5-linux-x64-v5.1.tgz安装

测试cifar10下的 python cifar10_train.py 成功

总结：

软件环境：

操作系统：Ubuntu16.04

显卡型号：GTK780

主板型号：GIGABYTE-uefi DualBIOS

所需软件版本：

NVIDIA-Linux-x86_64-367.44.run

cuda_7.5.18_linux.run

cudnn7.5-linux-x64-v5.1.tgz

Anaconda2-4.1.1-Linux-x86_64

tensorflow-0.10.0rc0-cp27-none-linux_x86_64.whl