Deploy a GPU Instance

Step 1: Open the Deploy tab

Navigate to Dashboard → GPU Compute and click the Deploy tab. This loads the live GPU catalog — a real-time listing of every available GPU model, its VRAM, and current per-hour rate in DZD.

Use the search bar to filter by GPU name (e.g. "A100", "RTX 4090", "H100"). Cards that show Available can be deployed immediately. Unavailable GPUs will show a stock indicator — availability refreshes automatically every 15 seconds.

Step 2: Select a GPU

Click on a GPU card to select it. The page scrolls down to the deploy configuration panel. Here you can adjust:

• GPU count — deploy multiple GPUs of the same model on a single instance (the hourly rate scales proportionally).
• Pod name — a human-readable name for your instance. Auto-generated by default.

Step 3: Choose a template

Click Change Template to browse available container images. Templates are pre-built Docker images with common ML frameworks and tools pre-installed. The default template includes PyTorch, CUDA, and Python.

Each template card shows the framework, CUDA version, Python version, and base OS. Select the one that matches your workload, or use the default for a general-purpose ML environment.

Step 4: Configure storage

GPU instances support two types of storage:

• Container Disk — ephemeral fast storage for your OS and installed packages. Default is 20 GB. Data here is lost when the instance is terminated.
• Volume Disk — additional ephemeral storage mounted at /workspace. Default is 20 GB. Also lost on termination, but survives pause/resume cycles.
• Network Volume (optional) — persistent storage that survives across instance lifecycles. Toggle "Attach Network Volume" and select an existing volume from the dropdown, or create one from the Network Volumes tab first.

For a deeper comparison of storage options, see Network Volumes.

Step 5: Access options

Two access methods are enabled by default:

• SSH — requires an SSH public key saved in your Account Settings. If no key is saved, SSH will be unavailable even if the toggle is on.
• JupyterLab — a browser-based notebook IDE accessible via your instance's connection URL. Protected by an auto-generated password shown in the instance details panel after deployment.

You can toggle either off before deploying if you don't need them.

Step 6: Deploy

Review the configuration summary at the bottom of the panel, then click Deploy. Your account balance must be sufficient to cover at least one hour of the selected configuration.

The instance will appear in the My Instances tab with a PENDING status. Once provisioning completes (typically 15–60 seconds), the status changes to RUNNING and connection details become available.