Ubuntu 14.04 从零开始配置cuda8.0+cudnn5.1+Tensorflow+openCV+caffe+matlab

字数 923阅读 202

Ubuntu 14.04 从零开始配置cuda8.0+cudnn5.1+Tensorflow+openCV+caffe+matlab

*************显卡驱动安装(注意,安装Tenflow的时候会将这一步的驱动刷回375.66,所以直接安装375.66就好)

1.Delete the old driver(if have):

$ sudo apt-get purge nvidia*

2.Ban nouveau:

$ sudo gedit /etc/modprobe.d/blacklist-nouveau.conf

add the following content:

blacklist nouveau

options nouveau modeset=0

Update the system:

$ sudo update-initramfs -u

3.Enter text mode:

Ctrl + Alt + F1

4.Shut down GUI:

$ sudo service lightdm stop

5.Run installation file(*.run), the options doesn't mind:

$ sudo sh NVIDIA*

6.Restart GUI:

$ sudo service lightdm start

7.Check if GPU driver is correctly installed:

watch -n 0.5 nvidia-smi

*************安装cuda

cuda files download address:

https://developer.nvidia.com/cuda-80-ga2-download-archive

1. Do not re-install GPU driver

2. First install the following dependencies(not fully necessary, if not meet 5):

$ sudo apt-get install freeglut3-dev build-essential libx11-dev libxmu-dev libxi-dev libgl1-mesa-glx libglu1-mesa libglu1-mesa-dev

3. After install mesa(which appears above), the GPU driver need to be instantly re-installed before any reboot.

4. install cuda

$ sudo sh cuda_*

5. if error:/usr/bin/ld:cannot find -lglut

due to lack of freeglut3-dev freeglut3

solution: $ sudo apt-get install freeglut3 freeglut3-dev

6. add PATH:

export PATH=/usr/local/cuda-9.0/bin${PATH:+:${PATH}}

export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

7. Test if installation is correct:

$ cd Nvidia-*samples

$ make

after "make"

$ cd Samples/bin/x86_84/linux/release/

test if compiled file could run correctly,as example:

$ ./deviceQuery

8. cuda_8.0.61.2_linux is patch, if needed

********************************安装cuDNN5.1

1. unzip cuDNN

2. $ sudo cp cuDNN/cuda/include/cudnn.h /usr/local/cuda/include

$ sudo cp cuDNN/cuda/lib64/* /usr/local/cuda/lib64/

3. update soft link

$ cd /usr/local/cuda/lib64/

$ sudo rm -rf libcudnn.so libcudnn.so.5

$ sudo ln -s libcudnn.so.5.0.* libcudnn.so.5

$ sudo ln -s libcudnn.so.5 libcudnn.so

4. use cuDNN sample to verify installation

$ cd /home/jilei/Downloads/cudnn-sample-v5/mnistCUDNN

$ make

5. if error:lcudart.so.8 not found, relink it as step 3 in this part

*******************************安装 tensorflow1.2

1. following the instruction

2. if connot compile in bazel part, change bazel version to 0.52 by

$ sudo apt-get autoremove bazel

and install the binary file with parameter --prefix=/usr

Reference: https://github.com/tensorflow/tensorflow/issues/11859

3. check installation

do not import tensorflow with 'tensorflow' directory

********************************安装 openCV

1. install pre-requisit

$sudo apt-get update

$sudo apt-get upgrade

$sudo apt-get install build-essential

$sudo apt-get install cmake git libgtk2.0-dev pkg-config libavcodec-dev libavformat-dev libswscale-dev

$sudo apt-get install python-dev python-numpy libtbb2 libtbb-dev libjpeg-dev libpng-dev libtiff-dev libjasper-dev libdc1394-22-dev

2. Download OpenCV

3. make files

$ cmake .

$ sudo make #this step takes a long time, better do anything else

4. if error in 87%: "nppigraphcutstate has not been declared", you need to change the source code:

in graphcuts.cpp

#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)

modified to

#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER) || (CUDART_VERSION >= 8000)

this error is due to the incompabiltiy of cuda8.0

5. install opencv

$ sudo make install

6. set PATH

sudo /bin/bash -c 'echo "/usr/local/lib" > /etc/ld.so.conf.d/opencv.conf'

sudo ldconfig

7. cd opencv/opencv-3.0.0-beta/samples/

sudo cmake .

sudo make -j $(nproc)

8.如果出现奇怪的bug,可能是cmake 的版本不对 cmake 一定要3.2版本才行,2.8不行

********************************安装 caffe

1. pay enough attention to config file and there seems all right

2. follow the mnist instruction to test if installation is correct

********************************安装 matlab R2017

1. mkidr to mount iso

$ mkdir ~/matlabtemp

$ sudo mount -t auto -o loop Matlab_dvd1.iso  ~/matlabtemp

$ sudo ~/matlabtemp/./install

2. Do not install matlab server!!!!!

3. wait to mount 2nd iso

$ sudo mount -t auto -o loop Matlab_dvd1.iso  ~/matlabtemp

4. umount these iso(twice)

$ sudo umount ~/matlabtemp/

$ sudo umount ~/matlabtemp/

5. follow the crack to activate

6. clear unused dir

$ sudo rm -r ~/matlabtemp

7. if occur opengl error, type the following in matlab command:

$ opengl('save','software')

8. For convenience, remember to change shortcut from Emacs to Windows

9. if matlab cannot read video file(such as .mp4 or .avi):

install gstreamer1.0 as following:

$ sudo apt-add-repository ppa:mc3man/trusty-media

$ sudo apt-get update

$ sudo apt-get install gstreamer1.0

Remember gstreamer1.0 for R2017a

********************************tips

1. ls -il to see link state

********************************双系统公用数据盘时

1.如果Linux无法修改ntfs内文件和文件夹的权限

在/etc/fstab中加入如下句子

UUID=109E7E199E7DF814 /media/jilei/MyFile ntfs-3g defaults,umask=0000,noauto 0 0