This topic describes how to manually install a GPU driver on a gn4, gn5, gn5i, gn6i, or gn6v instance. If you do not choose to automatically install a GPU driver when you create an instance, you must manually install the GPU driver.

Background information

You can install a GRID driver on a vgn5i instance. For more information, see Install NVIDIA GRID drivers on vgn6i or vgn5i instances (Linux) and Install NVIDIA GRID drivers on vgn6i or vgn5i instances (Windows).

Procedure

  1. Go to the NVIDIA website.
  2. Search for the suitable driver manually.
    1. Select a product type, series, and family based on the GPUs that your instance type is equipped with.
      The following table lists the information of GPUs that various instance types are equipped with.
      Parameter gn4 gn5 gn5i gn6v gn6i
      Product type Tesla Tesla Tesla Tesla Tesla
      Product series M-Class P-Series P-Series V-Series T-Series
      Product family M40 Tesla P100 Tesla P4 Tesla V100 Tesla T4
    2. Select an operating system version based on the image of the instance.
      If your instance uses the Debian operating system, select Linux 64-bit. If the drop-down list does not contain the operating system that you instance uses, click Select All Operating Systems in the lower part of the drop-down list.
    3. Select a CUDA Toolkit version.
    4. Select a language.
    5. Click Search.
  3. Confirm that the driver information is correct, and click Download.
    If your instance uses a Linux operating system, do not close the download page. When you install the GPU driver, you may need to refer to the installation steps in Additional Information.
  4. Install the GPU driver.
    • Install the GPU driver on a Linux instance:
      1. Download and install the kernel-devel and kernel-header packages based on your kernel version.
        Note The mismatch between kernel and kernel-devel versions results in a driver compilation error when the driver is installed from the .rpm file. You can run the rpm –qa | grep kernel command on an instance to check whether the versions match. Make sure that the versions match and reinstall the driver.
      2. Run the sudo rpm -qa | grep $(uname -r) command to check whether the kernel-devel and kernel-header packages are downloaded and installed.
        Take CentOS 7.3 as an example. If the following version information is displayed, the kernel-devel and kernel-header packages are installed:
        kernel-3.10.0-514.26.2.el7.x86_64
        kernel-headers-3.10.0-514.26.2.el7.x86_64
        kernel-tools-libs-3.10.0-514.26.2.el7.x86_64
        python-perf-3.10.0-514.26.2.el7.x86_64
        kernel-tools-3.10.0-514.26.2.el7.x86_64
      3. Perform the following steps in Additional Information on the download page to install the GPU driver.
        Take Linux 64-bit Ubuntu 14.04 as an example. The steps in Additional Information are shown in the following figure.
    • Install the GPU driver on a Windows instance:
      Double-click the package and follow the prompts to complete the installation.
      Note On Windows instances where the installed GPU drivers have taken effect, Windows Remote Desktop Protocol (RDP) may not support DirectX-based and OpenGL-based applications. In this case, you must install the Virtual Network Computing (VNC) service and client or use other protocols that support these applications, such as PC over IP (PCoIP) and XenDesktop HDX 3D.