Alternate method to install the Intel Gaudi Software Stack and Driver

The accelerator role allows users to set up the Intel Gaudi Software Stack and Driver. This tools allow users to unlock the potential of installed Intel Gaudi accelerators.

Prerequisites

  • The Intel Gaudi local repositories must be configured using the local_repo.yml script.

  • The input/software_config.json must contain valid intelgaudi version. See input parameters for more information.

Note

Intel Gaudi platform is only supported on Ubuntu 22.04 clusters containing Intel Gaudi accelerators.

Playbook configurations

The following configurations takes place while running the accelerator.yml playbook:

  1. Servers with Intel Gaudi accelerators are identified and the latest drivers and software stack are downloaded and installed.

  2. Servers with no accelerator are skipped.

Executing the playbook

To install all the latest drivers and toolkits, run:

cd accelerator
ansible-playbook accelerator.yml -i inventory

Note

  • While executing the accelerator.yml playbook for Intel Gaudi nodes, a Cron job is run which brings up the Intel Gaudi scale-out network interfaces.

  • If a node contains an Intel Gaudi GPU with internet access during provisioning, then the user needs to install the Gaudi driver using the accelerator.yml playbook.

If you have any feedback about Omnia documentation, please reach out at omnia.readme@dell.com.