Installed base package
yum -y install epel-release
yum -y install kernel-devel
yum -y install dkms
|
Edit the file /etc/default/grub 修改
" GRUB_CMDLINE_LINUX
"rd.driver.blacklist=nouveau nouveau.modeset=0
grub2-mkconfig -o /boot/grub2/grub.cfg
|
Restart
Install the graphics driver
Download graphics driver
下载完成是类似这样的文件 NVIDIA-Linux-x86_64-
410.72
.run 直接执行即可,
卸载 直接接参数 --uninstall
验证驱动是否正常
nvidia-smi
|
CUDA 9.0 installation
We use rpm mounted
https:
//developer.nvidia.com/cuda-91-download-archive?target_os=Linux&target_arch=x86_64&target_distro=CentOS&target_version=7&target_type=rpmlocal
|
The need to patch together also download
-rw-r--r--
1
root root
155023660
Jul
14
2018
cuda-repo-rhel7-
9
-
0
-
176
-local-patch-
4
-
1.0
-
1
.x86_64.rpm
-rw-r--r--
1
root root
1595730326
Sep
23
2017
cuda-repo-rhel7-
9
-
0
-local-
9.0
.
176
-
1
.x86_64.rpm
-rw-r--r--
1
root root
150679337
Dec
21
2017
cuda-repo-rhel7-
9
-
0
-local-cublas-performance-update-
1.0
-
1
.x86_64.rpm
-rw-r--r--
1
root root
149765568
Feb
15
2018
cuda-repo-rhel7-
9
-
0
-local-cublas-performance-update-
2
-
1.0
-
1
.x86_64.rpm
-rw-r--r--
1
root root
173037836
May
24
2018
cuda-repo-rhel7-
9
-
0
-local-cublas-performance-update-
3
-
1.0
-
1
.x86_64.rpm
|
During installation, there may be a software package written conflicts, such as ipa-client, salt-minion, etc., we can temporarily put the conflict have uninstalled, installed back in the back
Cause of the conflict is yum python modules are installed, the cause of pip python management problems, so there is a conflict
the installation of these rpm rpm -ivh not get away, it actually just unpacked in the / var directory
(paddle-venv) root
@algorithmgpu
-
11
-
123
:/var# ls -ld /var/cuda*
drwxr-xr-x
3
root root
138
Aug
6
11
:
48
/var/cuda-repo-
9
-
0
-
176
-local-patch-
4
/
drwxr-xr-x
3
root root
4096
Aug
6
11
:
56
/var/cuda-repo-
9
-
0
-local/
drwxr-xr-x
3
root root
138
Aug
6
11
:
48
/var/cuda-repo-
9
-
0
-local-cublas-performance-update/
drwxr-xr-x
3
root root
138
Aug
6
11
:
48
/var/cuda-repo-
9
-
0
-local-cublas-performance-update-
2
/
drwxr-xr-x
3
root root
182
Aug
6
11
:
48
/var/cuda-repo-
9
-
0
-local-cublas-performance-update-
3
/
|
We need access to these directories rpm installed inside, the installation is also the order of
- /var/cuda-repo-9-0-local/ rpm -ivh *
- Sequentially pressing the remaining patch / var / cuda-repo-9-0-local / rpm -Uvh * sequence
Such cuda installation is complete, the installation directory /usr/local/cuda-9.0
The following configuration environment variable
cat /etc/profile.d/cuda.sh
export LD_LIBRARY_PATH=/usr/local/cuda/lib64/:${LD_LIBRARY_PATH}
export PATH=/usr/local/cuda/bin/:${PATH}
|
Importing the environment variable source / etc / profile
Installation cudnn-9.0-linux-x64-v7.1
Download the above version
https:
//developer.nvidia.com/rdp/cudnn-download
|
installation
https:
//docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html
其实只是拷贝到对应目录
$ sudo cp cuda/include/cudnn.h /usr/local/cuda/include
$ sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64
$ sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*
|
Compile and install paddle
File
Note compile the sentence before you confirm installed in a virtual environment, compiler-dependent table of dependencies mentioned:
We use native compilation methods, non-docker
Shining documents do step by step on it
tips
make -j $ (nproc) command execution time may have half a day, depending on the network availability
Because you want to download a lot of warehouses on git
This last paddle directory has 8.4G