Installing the NVIDIA HPC SDK

Procedure

1. Use PuTTY to log in to the server as the root user.
2. Obtain the NVHPC software package.

   wget https://developer.download.nvidia.com/hpc-sdk/21.9/nvhpc_2021_219_Linux_aarch64_cuda_11.4.tar.gz

3. Create an installation directory.

   mkdir -p /opt/compiler

4. Copy the NVHPC software package to the installation directory.

   cp –r nvhpc_2021_219_Linux_aarch64_cuda_11.4.tar.gz /opt/compiler

5. Go to the installation directory.

   cd /opt/compiler

6. Decompress the software package.

   tar -zxvf nvhpc_2021_219_Linux_aarch64_cuda_11.4.tar.gz

7. Go to the directory generated after the decompression.

   cd nvhpc_2021_219_Linux_aarch64_cuda_11.4/

8. Start the installation.

   ./install

   

   Press Enter to confirm the installation location. The default installation location is /opt/nvidia/hpc_sdk. Then press 1 to select single-node installation. The installation program automatically finds the previously installed CUDA Toolkit. After the installation is complete, press Enter.

9. Set the NVIDIA HPC SDK environment variables.

   module load /opt/nvidia/hpc_sdk/modulefiles/nvhpc/21.9

10. Perform the verification.

    pgcc --version

    If the command output contains "PGI Compilers and Tools", the installation is successful.