CUDA Installation Guide on Linux systems
Applicable operating system
Fedora 7,8,9,10
Redhat Enterprise 3.x,4.x,5.x
SUSE Linux Enterprise Desktop 10-sp1,10.2,11.0
OpenSUSE 10.1,10.2,10.3,11.0,11.1
Ubuntu 7.04, 7.10.,8.04,8.10,9.04
--------------------------------------------------------------------------------
Download and operating system matching
Driver, SDK, Tookit
Address: http://www.nvidia.com/object/cuda_get.html
--------------------------------------------------------------------------------
Setup (TESLA is used with non-NVIDIA graphics cards without the need to install the video driver)
--------------------------------------------------------------------------------
Install in terminal (do not enter Xwindow)
Install cuda2.1 under Linux as5.2 for example
1. Install CUDA operation driver
Command line execution: SH nvidia-linux-x86_64-180.22-pkg2.run
Follow the prompts to return to perform each step of the installation process
For information on how to install NVIDIA's Linux drivers, refer to the
NVIDIA accelerated Linux Driver Set README and installation Guide
Http://us.download.nvidia.com/XFree86/Linux-x86/1.0-9755/README/index.html
After installation, [Nvidia-xconfig-query-gpu-info] can be executed in terminal to view the installed NVIDIA GPU
See below for implementation results
2. Install NVIDIA CUDA toolkit
Command line execution: SH cudatoolkit_2.1_linux64_rhel5.2.run
The installer will ask you to enter the installation path or accept the default value, and it is recommended to install as root and use the default path (/usr/local).
After that we will replace the actual installation path with <CUDA_INSTALL_PATH>
Increase CUDA binary (NVCC) and function Path (libcuda.so) to path and LD_LIBRARY_PATH environment variables
After installation, you can perform [Nvidia-smi] to view the installed CUDA GPU
[Nvidia-smi] is a new tool for Nvidia that allows us to confirm whether the GPU installed in the machine is functioning properly CUDA
See below for implementation results
3. Install NVIDIA CUDA SDK
Execution at the command line: Shcuda-sdk-linux-2.10.1215.2015-3233425.run
Setup will require you to enter the installation path or accept the default, and the default installation path is the user's home directory (/NVIDIA_CUDA_SDK).
After that we will take the <SDK_INSTALL_PATH> instead of the actual installation path. In the Bash_profile, add the following lines in the home directory.
Path= $PATH: <cuda_install_path>/bin
Ld_library_path= $LD _library_path:<cuda_install_path>/lib64
Export PATH
Export Ld_library_path
Note <CUDA_INSTALL_PATH> Replace with actual path installed in the system
Then enable the configuration
source. bash_profile
4. Building the SDK Project sample program
CD <SDK_INSTALL_PATH>
Build:
-Release enter "make".
-Debug Enter ' Make Dbg=1 '.
-Emurelease Enter "Make Emu=1".
-Emudebug Enter ' Make Emu=1 dbg=1 '. Make
Libcutil This common tool used in <SDK_INSTALL_PATH> execute make create sample programs
Libcutil is provided for ease of use and is not part of the CUDA.
Attention:
Some examples of OpenGL used in make are compile errors about GL, due to the fact that there is no library for OpenGL installed, and a separate library to install GL is required.
Other examples should be compiled correctly.
You can directly execute make to compile directly into each of the examples under/root/nvidia_cuda_sdk/projects:
such as: Matrix multiply
Cd/root/nvidia_cuda_sdk/projects/matrixmul
Make
5. Examples of Implementation
The Devicequery in the sample program is that we get GPU information that we can perform CUDA operations on this machine.
Building an example program
CD <sdk_install_path>/projects/devicequery
Make
Then execute the sample program in <sdk_install_path>/bin/linux32/release/devicequery
Devicequery execution results are shown below
and perform release, Debug, emurelease or Emudebug, etc.
Its directory is located in/bin/linux32/[release|debug|emurelease|emudebug]
--------------------------------------------------------------------------------
to create your own program
--------------------------------------------------------------------------------
Using the CUDA SDK makes it easy to create new CUDA programs.
To replicate and modify the project "template" provided by the CUDA SDK to meet your needs
Steps are as follows
1. Copy the entire "template" project (this myproject represents the project you want to create)
CD <sdk_install_path>/projects
Cp-r Template <myproject>
2. Change the file name of the project to the name of the file you want
MV Template.cu MYPROJECT.CU
MV Template_kernel.cu MYPROJECT_KERNEL.CU
MV Template_gold.cpp Myproject_gold.cpp
3. Change the file name of the project contents to the file name you want
Edit Makefile and original files
Replace all "template" with "MyProject".
4. Compile
Make
5. Execute the new program in the following location
.. /.. /bin/linux32/release/myproject
The execution result should be "Test passed"
6. Finally change the program code to meet your operational needs can be
For this part please refer to CUDA programming Guide