当前位置：网站首页>NVIDIA three piece environment configuration

NVIDIA three piece environment configuration

2022-06-27 20:30:00 【51CTO】

For the environment of deep learning , We need to straighten out the relationship between .

Like PyTorch frame , The higher the version , The higher the demand for graphics card computing power , The computing power of the graphics card is up ,CUDA The version of should also follow up ,CUDA The version goes up , The driver version should also keep up with ...

Personally, I usually follow the driver of higher version + Low version of the CUDA+ And CUDA Corresponding cudnn+ And CUDA Corresponding PyTorch Configure the environment in this order .

CUDA Correspondence with the drive

But there will be metaphysical problems , Self exploration ！

Multi Baidu , Will be solved .

drive

Encounter problems that can't be solved , The easiest way is to clean it all up , Restarting . If the new computer is configured for the first time, you can skip the clearing operation .

       
       #  uninstall CUDA, Correspond to your version number 
       
       sudo /usr/local/cuda-11.1/bin/cuda-uninstaller
       
       sudo 
       
       rm 
       
       -rf /usr/local/cuda-11.1/
       
       #  Unload the drive 
       
       sudo /usr/bin/nvidia-uninstall
       
       sudo apt-get 
       
       --purge remove nvidia*
       
       sudo apt-get purge nvidia*
       
       sudo apt-get purge libnvidia*
       
       sudo apt-get autoremove
       
       #  Make sure nothing is output 
       
       sudo dpkg 
       
       --list | 
       
       grep mvidia*
       
       #  Make sure the open source driver is disabled nouveau, Theoretically normal installation NVIDIA The official driver will be automatically disabled 
       
lsmod | 
       
       grep nouveau
       
       #  No output means that the disabling is successful 
      
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.

Actually CUDA Will bring its own drive , But usually the driver is installed first , Then go install CUDA, In the installation CUDA Cancel the installation in the process of CUDA Built in drive .

Of course, it can also be directly driven and CUDA With the installation

       
       #  Download the corresponding driver for your card 
       
 Download address ：https://www.nvidia.cn/Download/index.aspx
       
       ?lang
       
       =cn
       
       #  After downloading, go to the directory where the file is located , It's a run Postfix file , Plus Execution Authority 
       
       sudo 
       
       chmod a
       
       +x NVIDIA-Linux-x86_64-470.82.00.run
       
       sudo ./NVIDIA-Linux-x86_64-470.82.00.run
       
       #  After that, restart 
       
       sudo reboot
       
       #  Test it , See if you output information 
       
nvidia-smi
      
1.
2.
3.
4.
5.
6.
7.
8.
9.

CUDA

       
       #  download CUDA, download run file , Don't deb file ,deb The file will replace the installed driver during installation 
       
 Download address ：https://developer.nvidia.com/cuda-toolkit-archive
       
       #  After downloading, go to the directory where the file is located , It's a run Postfix file , Plus Execution Authority 
       
       sudo 
       
       chmod a
       
       +x cuda_11.1.0_455.23.05_linux.run
       
       sudo ./cuda_11.1.0_455.23.05_linux.run
       
       #  If the driver has been installed , Remember to install CUDA Cancel the installation of the driver 
       
       #  Check the environment variables after the installation , Theoretically, it will automatically add... At the end 
       
       sudo 
       
       vim ~/.bashrc
       
       #  without , Remember to add , Corresponding to good version 
       
       export 
       
       PATH
       
       =/usr/local/cuda-11.0/bin
       
       ${PATH:+:
       
       ${PATH}
       
       }
       
       export 
       
       LD_LIBRARY_PATH
       
       =/usr/local/cuda-11.0/lib64
       
       ${LD_LIBRARY_PATH:+:
       
       ${LD_LIBRARY_PATH}
       
       }
       
       #  Update environment variables 
       
       source ~/.bashrc
       
       #  Test version 
       
nvcc 
       
       -V
       
       # Compile and test the device  deviceQuery：
       
       cd /usr/local/cuda/samples/1_Utilities/deviceQuery
       
       sudo 
       
       make
       
./deviceQuery
       
       # Compile and test bandwidth  bandwidthTest：
       
       cd ../bandwidthTest
       
       sudo 
       
       make
       
./bandwidthTest
      
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.

NVIDIA Three piece environment configuration _CUDA

cudnn

      
      
       
       #  Download the corresponding CUDA Version of cudnn, You need to register in advance 
       
       
 Download address ：https://developer.nvidia.com/rdp/cudnn-archive
       
       

       
       #  Download the following file ：
      
      
      
      
       
       1.
       
       2.
       
       3.

NVIDIA Three piece environment configuration _cudnn_02

       
       #  After decompressing , File replacement 
       
       sudo 
       
       cp cuda/include/cudnn.h /usr/local/cuda/include
       
       sudo 
       
       cp cuda/lib64/libcudnn* /usr/local/cuda/lib64
       
       #  Change permissions 
       
       sudo 
       
       chmod a
       
       +r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*
       
       #  install deb package 
       
       sudo dpkg 
       
       -i libcudnn8_8.0.5.39-1
       
       +cuda11.0_amd64.deb
       
       sudo dpkg 
       
       -i libcudnn8-dev_8.0.5.39-1
       
       +cuda11.0_amd64.deb
       
       sudo dpkg 
       
       -i libcudnn8-samples_8.0.5.39-1
       
       +cuda11.0_amd64.deb
       
       #  Test after completion 
       
       cd /usr/src/cudnn_samples_v8/mnistCUDNN
       
       sudo 
       
       make
       
./mnistCUDNN
       
       #  Pictured , success 
      
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.

NVIDIA Three piece environment configuration _cudnn_03

Problem summary

function sudo dpkg --list | grep mvidia* And the output

NVIDIA's related packages have not been deleted yet , Compare the output of the remaining packages , Delete one by one .

       
       #  repeat 
       
       sudo apt-get 
       
       --purge remove xxx
       
       #  Finally, make sure to run the following command to ensure no output 
       
       sudo dpkg 
       
       --list | 
       
       grep mvidia*
      
1.
2.
3.
4.

function lsmod | grep nouveau There is no output , But also prompt to disable

Theoretically, the driver will disable itself nouveau Open source driver , But the metaphysics problem , You'll see .

       
       #  Conduct manual blasting 
       
       sudo 
       
       vim /etc/modprobe.d/blacklist-nouveau.conf
       
       #  Add the following 
       
blacklist nouveau options nouveau 
       
       modeset
       
       =
       
       0
       
       #  Execute this command to update 
       
       sudo update-initramfs 
       
       -u
       
       #  Re install NVIDIA Just drive , Restart immediately after installation 
      
1.
2.
3.
4.
5.
6.
7.

Ps： The screen may be blank after the driver is installed again after the update , Wait a moment . Then force the restart .
If you cannot access the graphical interface ,Ctrl+Alt+F1 Switch to character interface ,Ctrl+Alt+F7 Switch to the GUI