Install CUDA

Compute Unified Device Architecture (CUDA) is a general-purpose parallel computing architecture created by NVIDIA. This architecture enables GPUs to perform complicated calculations. It contains the CUDA Instruction Set Architecture (ISA) and a GPU internal parallel computing engine. Developers can use C language, C++, or Fortran to write programs. The programs can run on CUDA-enabled processors at ultra-high performance.

GPU Elastic Physical Compute (GEPC) uses NVIDIA graphics cards and requires the CUDA development and running environment. This topic describes how to install the CUDA Toolkit on Ubuntu 16.0.4.

Install the dependency libraries

Log in to the GEPC instance, download the dependency libraries, and execute the following command:

sudo apt-get install libGLU* libXi* libXmu*  -y

Download the CUDA installation package and install CUDA

1. Log in to the GEPC instance and then access the CUDA driver.

2. Select the installation package for your operating system. In this example, the operating system is 64-bit Ubuntu 16.0.4.

   

   Note:

   • It is recommended that you select runfile (local) for Installer Type.
   • network: Network installer, which requires re-downloading each component with each new installation. The package is small.
   • local: Local installer, which includes all components. The package is large.

3. Click Download and select the destination storage directory:

   

4. Change to the directory where the CUDA installation package is saved and execute the following command:

   sudo sh cuda_9.1.85_387.26_linux.run  

5. Select accept -yes -enter as prompted.

   Note:

   • If the command displays the following information, the CUDA installation package contains the driver, toolkit, and samples but only the driver is installed.

     Driver: Installed require reboot    
     Toolkit: install skip  
     Samples: install skip  

     Execute the following command to reinstall CUDA:

     sudo sh cuda_9.1.85_387.26_linux.run

   • If the command displays the following information, CUDA is installed successfully:

     Driver: Installed   
     Toolkit: Installed in /usr/local/cuda   
     Samples: Installed in /home/XX   

6. In the /usr/local/cuda/samples/1_Utilities/deviceQuery directory, execute the make command to compile the deviceQuery program. If CUDA has been installed correctly, the following information is displayed when you run deviceQuery: