User Guidelines

DELL GB10 Guidelines (Testing Phase) View

Please be aware that these devices are not approved for sensitive or very sensitive data. Any application involving sensitive or very sensitive data will be automatically rejected. These devices are only approved for use with Public or Restricted data.

Powering the Device

The device will have a small green light on when it is powered on

If you are having problems powering the device, please notify the team via the AI Co-Lab Forum Google Space, including the MU number and/or the DELL GB10 device number.
The DGX dashboard requires a login – you must use the same credentials used to log onto the systems

Logging on/off

Check your allocation: Make sure you are using the specific Spark station assigned to you during your booking (for example: Spark 5 Station).
Username: Use your Monash username (a mix of letters and numbers, e.g., abcd0001).
Password: Use your standard Monash Single Sign-On (SSO) password.

Data Management & Storage

Because this is a shared testing environment, data discipline is critical.

Where to save: S-Drive: Use this for permanent storage, large datasets, and code repositories.
Local Device: Use only for temporary scratch space during active compute jobs.
The Golden Rule: ALWAYS back up/save your data to the S-Drive after each session. Local data on the device may be cleared or overwritten.
It is recommended to use the same device each time to save time when logging in.

Your S-Drive should mount automatically upon login. If it is missing, please notify the team via the AI Co-Lab Forum Google Space.

If you don’t have an S-Drive share to use as storage, you can request one here.

These devices require periodic data wipes to maintain security and operational efficiency. Users will be notified in advance of a scheduled wipe to minimise disruption to their ongoing work.

Troubleshooting & Assistance

If you require support when using the device, use the following resources for support & troubleshooting:

Quick Start View

Environment Setup Recipes

Depending on your project workflow, choose one of the following methods to set up your Python environment.

Option A: Miniforge (Recommended for Conda users)

Ideal if you need flexible package management with access to the conda-forge channel.

Download the miniforge3 installer following the instructions on the github page
Run the installer in a bash terminal: bash Miniforge3-....sh
Install the environment into your home directory (not your S drive)
Create your environment.yml according to these instructions, storing the file in your S drive
Activate your environment: source /path/to/your/miniforge3/bin/activate
Create your environment: mamba create –file /path/to/your/environment.yml

Option B: Standard Pip Virtual Environment

A lightweight, native Python approach.

Create your requirements.txt according to these instructions, storing the file in your S drive
Navigate to your home directory or project folder
Create your Python virtual environment: python3 -m venv my_env
Activate it: source my_env/bin/activate
Install your requirements: python3 -m pip install -r /path/to/your/requirements.txt

Option C: Default NVIDIA Jupyter Environment

If you just want to experiment with NVIDIA’s preselected libraries quickly, simply launch the default Jupyter instance provided on the system - follow NVIDIA’s instructions here

Connecting your Python environment to Visual Studio Code (VS Code)

Open your project folder on the DGX.
Download your preferred Python extensions, including the Microsoft Python and Jupyter packages. You can access extensions from the marketplace via the 4 squares icon in the activity bar on the left side of the window.
Select your Python Interpreter: Press Ctrl+Shift+P, type Python: Select Interpreter, and point it to the Miniforge or Pip virtual environment you created in Step 3 using ‘Enter interpreter path…’. You will need to provide the full path to the Python executable in the bin folder.
Your Python environment will now be available for use with the debugger or Jupyter notebooks.

Rootless Docker and NVIDIA containers

The DELL GB10s are set up with Docker rootless mode. This means that you will not need to use sudo to access the Docker daemon and run containers.

To follow the DELL GB10 tutorial, use the Docker instructions in the ‘Set Up Manually’ tab. The first step, ‘Configure Docker permissions,’ has been done for you.

To install a non-NVIDIA container, simply pull and run the container as normal without using the sudo command: docker run hello-world

Note that the NVIDIA Sync instructions should be ignored as remote access is not enabled for you on these devices.

Monitoring Resource Usage

The DELL GB10 utilises shared memory. There are no dedicated GPUs, so standard GPU monitoring tools, such as nvidia-smi, will not work.

Monitor Resources: Open your browser and navigate to the DGX dashboard at localhost:11000.
Blocked Packages? If network restrictions block a package download, do not attempt to bypass it. Contact the team via our AI Co-Lab Google Space.

Known Issues View

Community of Practice

The Google Space will be the primary channel for announcements, peer collaboration, and real-time support from eSolutions and eResearch during your booked sessions.

FAQs

You can check how much disk space is currently available on the system by opening your terminal and running the disk free command:
Bash
df -h
(The -h flag makes the output human-readable, showing sizes in GB or TB). If the drive is full, please archive old data to the S-Drive or contact support.
Alternatively, run the installed Disk Usage Analytics tool. The local disk will be represented as ‘/’
Since this is a shared environment, you may occasionally see remnants of previous sessions. Do not modify or delete another researcher's data. Please flag it in the Gmail Community channel so an administrator can safely move or clear it.
If you need to install system packages, go to the AI Co-Lab Google forum and let us know what you need access to – we can load this for you.

We are unable to provide sudo access during the testing phase. Please use Miniforge, Pip virtual environments, or rootless Docker to manage your dependencies in userspace.
1. The initial step would be to contact the support team via Google Space.
2. Alternatively, you can lodge a request via the general IT form, noting ‘ACL Request for AI Co Lab’ in the summary box, followed by the URL you need to access and the reason for the access.
Since these devices are in a shared lab, we will likely be wiping them periodically, but we have not yet determined how frequently we will do that. Any data or code you want persisted, you should be able to access the S-Drive and an AD home directory from the GB10 devices. This is the safest place for storing between GB10 sessions, whether or not you get the same device or a different one. All users will be notified before data is wiped via our user group management
The SSH service is not enabled by design. The workstations are intended to be used while seated and logged into one directly.
The SSH NVIDIA sync app will not work.

Monash Research Infrastructure