深度学习环境搭建——cuda ,cudnn

1，安装显卡参考流程：

https://www.cnblogs.com/mar-q/p/7482720.html

注意，由于yum会自动安装依赖，所以安装gcc会自动安装kernel head等的版本，改版本可能和内核版本不一致，因此在第一步安装gcc前最好先安装kernel-head和kernal-devel （查看当前内核： uname -r 3.10.0-862.el7.x86_64）

查看gpu驱动是否装成功：nvidia-smi

 

2，安装cuda参考流程：

https://docs.nvidia.com/cuda/cuda-installation-guide-linux/

注意下kernel系列的版本号和内核一致即可，然后直接安装，

1）安装后将命令加入配置文件中

2）测试是否安装成功

cd /usr/local/cuda-8.0/samples/1_Utilities/deviceQuery sudo make （这里第一次执行出错，提示没有g++，安装后解决） ./deviceQuery

安装cuda9.0，查询后的版本号为

CUDA Driver Version / Runtime Version 10.0 / 9.0

 

3，安装cudnn

拷贝过去cudnn->cuda (cuda9是cuda安装目录)

cp cuda/include/cudnn.h cuda9/include/

cp cuda /lib64/libcudnn* cuda9/lib64

chmod a+r cuda9/include/cudnn/h cuda9/lib64/libcudnn*

 

4，安装nvdia docker (docker安装成功后)

wget -P /tmp https://github.com/NVIDIA/nvidia-docker/releases/download/v1.0.1/nvidia-docker-1.0.1-1.x86_64.rpm sudo rpm -i /tmp/nvidia-docker*.rpm && rm /tmp/nvidia-docker*.rpm sudo systemctl start nvidia-docker

 

5，恢复镜像,

cat centos7-cuda1.tar | docker import - centos7-cuda1

(导出镜像docker export <CONTAINER ID >> my_container.tar)

 

6，启动镜像

导入的镜像docker run启动有点问题

docker run -it -d centos7-cuda /bin/bash

带挂载目录的启动：docker run -it -v /home/weihua:/home centos7-cuda /bin/bash

进入

docker exec -it CONTAINER_ID /bin/bash

关闭

 

7，安装opencv

官方指导：https://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_setup/py_setup_in_fedora/py_setup_in_fedora.html#install-opencv-python-in-fedora

或者参考https://blog.csdn.net/xbw12138/article/details/80025790

 

第一步非常重要配置epel源。

因为opencv暂时对cuda9的支持有问题，

错误：提示CMake Error: The following variables are used in this project, but they are set to NOTFOUND. Please set them or make sure they are set and tested correctly in the CMake files:

解决：https://stackoverflow.com/questions/46584000/cmake-error-variables-are-set-to-notfound

 

错误：make: command not found提示

解决：yum -y install gcc automake autoconf libtool make

 

错误：ModuleNotFoundError: No module named 'cv2'

解决：将opencv2.4.13/release/lib下的cv2.so复制到 python组件包的路径中：/usr/local/lib/python2.7/site-packages到python路径中

（vim ~/.bashrc 添加export PYTHONPATH=$PYTHONPATH:/opt/conda/lib/python3.6/site-packages source ~/.bashrc）

 

错误：ImportError: dynamic module does not define module export function (PyInit_cv2)错误

解决：pip install opencv-python

 

错误：提示yum被锁上

解决： rm -rf /var/run/yum.pid 

 

错误：找不到 kernal header

Failed to run `/usr/sbin/dkms build -m nvidia -v 410.73 -k 3.10.0-862.el7.x86_64`: Error! echo

Your kernel headers for kernel 3.10.0-862.el7.x86_64 cannot be found at

/lib/modules/3.10.0-862.el7.x86_64/build or /lib/modules/3.10.0-862.el7.x86_64/source.

问题1：提示yum被锁上

解决： rm -rf /var/run/yum.pid 

 

问题2：找不到 kernal header

Failed to run `/usr/sbin/dkms build -m nvidia -v 410.73 -k 3.10.0-862.el7.x86_64`: Error! echo

Your kernel headers for kernel 3.10.0-862.el7.x86_64 cannot be found at

/lib/modules/3.10.0-862.el7.x86_64/build or /lib/modules/3.10.0-862.el7.x86_64/source.

解决：

https://www.cnblogs.com/mylinux/p/5612168.html 解决方案2