Set Up

   

Get started with Intel® Gaudi® accelerators by installing the latest software, accessing Intel® Tiber™ AI Cloud or Amazon EC2* DL1, and following our quick start guide for seamless setup and onboarding.

Software Installation or Running the Docker* image for Intel Gaudi AI Accelerators

In most cases where you are getting direct access to an Intel Gaudi Software node or virtual machine from a cloud service provider, the software should be pre-installed and users should simply load the latest Docker image to run their workloads.  Follow the steps in the section below to verify the software you have installed. If you are working on a new Intel Gaudi node or want to install all the software manually, please refer to the detailed installation guide.   We provide an automated installer script (habanalabs-installer.sh) to install or update the Intel Gaudi Software. You can setup a bare metal environment, where the software is loaded directly on the machine or in a virtual environment.   You can also use the script to setup the platform to use the Docker image, which requires the base system software installation of the Intel Gaudi Software driver, firmware and firmware tools.

To run the Docker image follow these steps:

  1. Run the hl-smi command to verify the current version of software that is running.
HL-SMI Version:       hl-1.18.0-XXXXXXX
Driver Version:       1.18.0-XXXXXX

  1. Check support documentation and confirm base installation

    • Refer to the support matrix in the documentation to see what Docker versions are compatible with the driver and software.
    • Confirm that the base system software is installed (see below) and the Docker container runtime is active by running Docker images command.  If the Docker runtime is not functional, please refer to the installation guide.

  2. Select the appropriate OS and PyTorch versions, and then Pull and Run the appropriate Docker image.  The example below shows how to Pull and Run the Docker using version 1.18.0 with the Ubuntu* 22.04 OS.

docker pull vault.habana.ai/gaudi-docker/1.18.0/ubuntu22.04/habanalabs/pytorch-installer-2.4.0:latest
docker run -it --runtime=habana -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice --net=host --ipc=host vault.habana.ai/gaudi-docker/1.18.0/ubuntu22.04/habanalabs/pytorch-installer-2.4.0:latest
  1. Check Installation: Follow these steps to confirm the current version installed on the platform.  For the full details of how to manage the Intel Gaudi Software on the platform please refer to the detailed installation guide.

The Intel Gaudi Software stack consists of two components: * The base system software: this contains the Intel Gaudi Firmware, Drivers, core software. * The platform software: these are the Intel Gaudi versions of PyTorch* and other packages.

The first step is to understand what version of firmware is loaded on the system, run the hl-smi command. Use the HL-SMI Version at the top of the output. For example, if the installed version is 1.18.0, the output should be as follows:

HL-SMI Version:       hl-1.18.0-XXXXXXX
Driver Version:       1.18.0-XXXXXX

To verify the base system software, run the following command: 

apt list --installed | grep habana

For a correct base system software installation you will see the following list:

habanalabs-container-runtime/focal,now 1.18.0-524 amd64 [installed]
habanalabs-dkms/focal,focal,now 1.18.0-524 all [installed]
habanalabs-firmware/focal,now 1.18.0-524 amd64 [installed]
habanalabs-firmware-odm/focal,now 1.18.0-524 amd64 [installed]
habanalabs-firmware-tools/focal,now 1.18.0-524 amd64 [installed]
habanalabs-graph/focal,now 1.18.0-524 amd64 [installed]
habanalabs-rdma-core/now 1.18.0-524 all [installed]+
habanalabs-qual/focal,now 1.18.0-524 amd64 [installed]
habanalabs-thunk/focal,focal,now 1.18.0-524 all [installed]

To verify the platform software, run the following command:

pip list | grep habana

For a correct platform software installation you will see the following list:

habana_gpu_migration        1.18.0.524
habana-media-loader         1.18.0.524
habana-pyhlml               1.18.0.524
habana-torch-dataloader     1.18.0.524
habana-torch-plugin         1.18.0.524
habana_transformer_engine   1.18.0.524
lightning-habana            1.6.0

You can then download the correct habanalabs-installer.sh install script (in this example change the path to 1.18.0), and follow the installation guide to install the base system software and/or the platform software.

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

 

 

 

Installation Guide

Intel® Tiber™ AI Cloud

Developers get access to Intel hardware, including the deep learning server that is powered by eight of the latest Intel® Gaudi® 2 AI accelerators.

To get access:

  1. Go to the Intel® Tiber™ AI Cloud and follow the steps in the get started section.

    • Register for an account and enroll in the program by entering your customer information and reviewing the terms and conditions.

    • Set up ssh access and your ssh keys to access the Intel Tiber AI Cloud.

  2. Get cloud credits by entering your payment information or redeeming a coupon.

  3. Sign in and access the Intel Gaudi platform instance in the hardware catalog.

 

 

 

Intel® Tiber® AI Cloud

 

Stay Informed


Register for the latest Intel Gaudi AI accelerator developer news, events, training, and updates.

Sign Up