Skip to content

Run Your Own Docker Workload (Batch Mode)

This guide shows how to run your own Docker workload in batch mode. The workload runs until the run-script completes; there is no SSH endpoint.

Sign in to AAC if needed (see Sign in to AAC).

Permissions

You need Developer or Admin permissions to run your own Docker. Contact support to request them.

New workload

  1. Click Workloads in the top bar.
  2. Click New Workload at the top right.

New Workload button

Select team

If you are assigned to more than one team, a pop-up asks you to select one of the customer teams. If you have only one team, this step is skipped.

Note

In this example we selected AMD Internal.

Click Next at the top right.

Workload selected team

Configure your own container

The view has two tabs: Docker container (configure your own container) and Application list (choose a pre-configured application). For batch mode we use Docker container.

  • General information: SSH enabled (off for batch), Container url.
  • Repository Authorization: User, Password, Server (default Docker Hub: https://index.docker.io/v1/). Use these only for private images.

In batch mode SSH service is deactivated; the workload runs until the run-script ends. Enter the container URL; add credentials only if the image is private.

Configure own Docker container

Select input files

Upload any input files the application needs. Click Upload files, then drag files in or click Browse files. If you do not need to upload files, click Next.

Workload input files

Configure run-script

Set the scripts to run in your container:

  • Pre-run script: Runs before your container in a basic Ubuntu image.
  • Run-script: Command executed inside your container.
  • Post-run script: Runs after your container in a basic Ubuntu image.

Click Next at the top right.

Workload run-script

Request resources

Set the resources for your workload:

  • Number of GPUs: Default 1 GPU.
  • Maximum allowed runtime: How long the workload can run before the system stops it (default 1 hour). You cannot change this after launch.
  • Oversubscribe: Allow sharing resources on the same GPU. Disabled by default.
  • Telemetry: Enable performance metrics (GPU, CPU, memory, network, I/O). Disabled by default.

Click Next at the top right.

Select resources

Select compute resources

Choose the cluster and queue assigned to your team. Select an available queue (e.g. one with MI300X) by clicking it, then click Next.

Select compute queue

Review workload submission

Review the workload details. Use Change in any section to edit, then click Run Workload.

Review workload

Monitor workload

After submission, the workload moves through these states:

  • Created – Submitted to the system.
  • Sent – Sent to the selected queue.
  • Pending – Waiting in the queue.
  • Running – Executing.
  • Completed – Run-script finished successfully.
  • Failed – The workload did not complete successfully.
  • Cancelled – Stopped by you or the system (e.g. maximum runtime exceeded).

Logs

Use the SysLog, Stdout, and Stderr tabs to inspect logs and output.

Workload logs