How To Create a GPU Container

Log in (Optional)

If not already authenticated:

npx cosmicac login

This opens your browser for authentication. If the browser doesn't open automatically, copy the URL from the terminal and paste it into your browser. Complete the login to continue.

Initialize the job configuration

Run the interactive setup to create a job.config.json file:

npx cosmicac jobs init

Follow the prompts to configure your job:

• Project name — A descriptive name for your job
• Tags — Comma-separated labels to organize your job
• Type — Select GPU_CONTAINER
• GPU type — Hardware configuration
• GPU count — Number of GPUs to allocate
• Country code — Region where your container runs (e.g., US, IN)
• Container image — Base image (e.g., ubuntu:24.04)
• Cost limit — Maximum spend threshold (USD)
• Alerts — Notifications to enable (e.g., Cost Exceeded, Errors)

To create the config in a specific directory:

npx cosmicac jobs init --dir ./my-project

Review the configuration

The generated job.config.json contains your job settings:

{
  "name": "train-image-model",
  "type": "GPU_CONTAINER",
  "tags": [
    "training",
    "image-recognition"
  ],
  "gpu": {
    "type": "GH100_H100_SXM5_80GB",
    "count": 4
  },
  "location": "IN",
  "params": {
    "cpu_limit": "4",
    "memory_limit": "8Gi",
    "cpu_request": "2",
    "memory_request": "4Gi"
  }
}

Adjust settings as needed. See GPU Types for available hardware options.

Create the container

npx cosmicac jobs create

Verify the job was created

npx cosmicac jobs list

Your container is ready when the status shows Running.