Machine learning stepped pit of Ubuntu systems install CUDA

Reference article: https://blog.csdn.net/u012235003/article/details/54575758

Ready to work

• Cuda official website to download the installation file
• Enter text line mode
  # install CUDA
• Close graphical interface

sudo service lightdm stop

• Run cuda file

First to cuda executable permissions:

sudo chmod a+x filename

• Run cuda file

sudo ./filename --no-opengl-libs

• Option to select the installation process
  read the statement, keep pressing D can then accept
  the option install nvidia accelerated Graphics Driver, enter n, as previously has been the nvidia driver install
  options install the OpenGL libraries, if dual graphics cards (set was + alone significantly) selected n, if only significant independence can choose y, if y dual graphics options, then there will be black or circulatory problems logged, if coupled with the above parameters will not have this option
  back can choose yes, we can choose the last option n, that is not copied sample, because there are sample installation directory

• Restart the graphical interface

sudo service lightdm start

• Environment Configuration
  After installation, restart open input terminal

export PATH="/usr/local/cuda-9.2/bin:$PATH" 
export LD_LIBRARY_PATH="/usr/local/cuda-9.2/lib64:$LD_LIBRARY_PATH"

These two lines of command is to export cuda bin and lib files to the file system environment.
If the installed version is not the same, replace cuda-9.2 path

Detecting whether the installation is successful

• Check the path
  if there are multiple files at the beginning of the nvidia, if not, incorrect installation, please refer to the solution under an article under ~ / dev

• Check the CUDA Toolkit is successfully installed
  terminal and type:

  nvcc -V

  Outputs CUDA version information

• Examples of samples compiled
  into the Samples directory, then the directory input terminal make, waiting for ten minutes.

• After compilation the test
  can then find Sample inside surface bin / x86_64 / linux / release / directory, that directory and switch to run the program deviceQuery

  sudo ./deviceQuery

  See the output, focusing on the last line, Pass is indicated by the test

  Run bandwidthTest program

  sudo ./bandwidthTest

  See the output displayed by the test results are expressed as PASS

• If the above step indicates successful installation CUDA