Driver and Software Installation

The following sections describe how to obtain and install Intel® Gaudi® software and drivers.

Installation Options

The following lists the available options for driver and software installation:

  • Install driver and software: Installs all the packages automatically using habanalabs-installer.sh script. This is the recommended installation method.

  • Upgrade driver and software: Enables upgrading an existing installation to the latest version.

  • Custom driver and software installation: Allows installing each package manually for a fine-grained control over the installation process.

Note

  • Make sure to review the currently supported versions and operating systems listed in the Support Matrix.

  • Driver and software installation is not required if you are using the Intel Gaudi Base Operator for Kubernetes or OpenShift.

  • Installing the package with internet connection available allows the network to download and install the required dependencies for the Intel Gaudi software package (apt get, yum, dnf install or pip install etc.).

Install Driver and Software

  1. Install the driver and software using habanalabs-installer.sh script. For further details on the package installers included, see Intel Gaudi Software Installers table.

    wget -nv https://vault.habana.ai/artifactory/gaudi-installer/1.18.0/habanalabs-installer.sh
chmod +x habanalabs-installer.sh
./habanalabs-installer.sh install --type base

    For further instructions on how to control the script attributes, refer to the help guide by running the following command:

    ./habanalabs-installer.sh --help

Note

  • The installation sets the number of huge pages automatically.

  • habanalabs-container-runtime and habanalabs-qual-workloads are not automatically installed with the habanalabs-installer.sh. Make sure to install them as shown in the steps below. Additionally, habanatools is not automatically installed. If you are using TPC and writing your own kernels, refer to TPC Tools Installation Guide to install habanatools package.

  1. Install optional packages:

    Install habanalabs-container-runtime. package. This package is required for running workloads in containers. Both Docker and Kubernetes are supported:

     sudo apt install -y habanalabs-container-runtime

    Install habanalabs-qual-workloads package. This package is required for running ResNet-50 training stress test plugin:

     sudo apt install -y habanalabs-qual-workloads

    Install ethtool if you are running a multi-server scale-out and need to bring up the accelerator interfaces:

     sudo apt install -y ethtool

    Install habanalabs-container-runtime. package. This package is required for running workloads in containers. Both Docker and Kubernetes are supported:

     sudo yum install -y habanalabs-container-runtime

    Install habanalabs-qual-workloads package. This package is required for running ResNet-50 training stress test plugin:

     sudo yum install -y habanalabs-qual-workloads

    Install ethtool if you are running a multi-server scale-out and need to bring up the accelerator interfaces:

     sudo yum install -y ethtool

    Install habanalabs-container-runtime. package. This package is required for running workloads in containers. Both Docker and Kubernetes are supported:

     sudo dnf install -y habanalabs-container-runtime

    Install habanalabs-qual-workloads package. This package is required for running ResNet-50 training stress test plugin:

     sudo dnf install -y habanalabs-qual-workloads

    Install ethtool if you are running a multi-server scale-out and need to bring up the accelerator interfaces:

     sudo dnf install -y ethtool

    Install habanalabs-container-runtime. package. This package is required for running workloads in containers. Both Docker and Kubernetes are supported:

     sudo zypper install -y habanalabs-container-runtime

    Install habanalabs-qual-workloads package. This package is required for running ResNet-50 training stress test plugin:

     sudo zypper install -y habanalabs-qual-workloads

    Install ethtool if you are running a multi-server scale-out and need to bring up the accelerator interfaces:

     sudo zypper install -y ethtool

Bring up Accelerator Interfaces

If you are running a multi-server scale-out and have the accelerator interfaces physically connected, make sure the network interfaces are brought up by running the command below. These interfaces need to be brought up every time the kernel module is loaded or unloaded and reloaded. A reference on how to bring up the interfaces is provided in the manage_network_ifs.sh. Note that the script can be found at /opt/habanalabs/qual/[gaudi3,gaudi2,gaudi1]/bin/.

# manage_network_ifs.sh requires ethtool
/opt/habanalabs/qual/[gaudi3,gaudi2,gaudi1]/bin/manage_network_ifs.sh --up

Upgrade Driver and Software

  1. Upgrade the software, run the following command:

     wget -nv https://vault.habana.ai/artifactory/gaudi-installer/1.18.0/habanalabs-installer.sh
 chmod +x habanalabs-installer.sh
 ./habanalabs-installer.sh upgrade --type base

  2. Perform Steps 2, 3 and 4 from the previous section Install Driver and Software to complete the upgrade.

Custom Driver and Software Installation

To install each package individually, refer to Custom Driver and Software Installation.

Note

While you can install each package manually, using the habanalabs-installer.sh script is the recommended method for installation. For further details, see Driver and Software Installation.