Network Volumes

What are Network Volumes?

A Network Volume is a persistent storage device that exists independently of any compute instance. Unlike ephemeral instance storage (Container Disk, Volume Disk), a Network Volume retains all data even after every instance attached to it has been terminated.

Network Volumes are shared across instance types — the same volume can be attached to a GPU instance for training, then later attached to a CPU instance for post-processing. The only restriction is that the instance must be in the same data center region as the volume.

Volumes are billed continuously while they exist, regardless of whether they are attached to a running instance. Check the Create Volume modal in your dashboard for current per-GB rates.

Data center availability

Network Volumes are currently available in the following data centers:

Instances you deploy must be in the same data center to attach a volume. Plan your region choice accordingly — volumes cannot be moved between data centers after creation.

Creating a Network Volume

1. Navigate to Dashboard → GPU Compute → Network Volumes and click Create Volume.
2. The Create Volume modal opens. Select a data center from the grid. Each card shows the data center ID and any capability badges (⚡ for High-Performance, 🌐 for global networking).
3. Enter a name (e.g., my-model-data) — this identifies the volume in your dashboard and in the attach dropdown when deploying instances.
4. Set the size in GB using the stepper control. Minimum is 10 GB, maximum is 1000 GB. Use the +/− buttons or type a value directly.
5. If the selected data center supports it, optionally enable High-Performance Storage (NVMe-backed, labeled "⚡ 3x Speed"). In high-performance-only data centers, this toggle is locked on automatically.
6. Review the total monthly cost shown at the bottom of the modal. This cost applies for as long as the volume exists, whether or not an instance is attached.
7. Click Create. The volume provisions in a few seconds and appears in your volume list.

Attaching to a GPU instance

Volumes are attached at deploy time, not while an instance is running.

1. Start deploying a GPU instance from Dashboard → GPU Compute → Deploy.
2. In the storage configuration section, you'll see Container Disk and Volume Disk sliders, plus an Attach Network Volume toggle.
3. Toggle Attach Network Volume on. The Volume Disk slider disappears — the Network Volume replaces the ephemeral Volume Disk at /workspace.
4. Select your volume from the dropdown. Only volumes in the same data center region as your selected GPU appear. If no volumes show up, you need to create one in the matching region first.
5. Complete deployment as normal. Your volume is mounted at /workspace and any existing data on it is immediately available.

Attaching to a CPU instance

CPU instances support Network Volumes identically — toggle Attach Network Volume during deploy, select the volume, and deploy. The volume mounts at /workspace.

The key difference is that CPU instances have no Volume Disk — only a Container Disk for ephemeral storage. Without a Network Volume, CPU instances have no /workspace partition. Attaching a Network Volume adds this persistent workspace.

Storage model: GPU vs CPU

What happens when your balance runs out

The billing system runs periodic checks on all active instances. If your balance is depleted, an automated safety mechanism activates in two stages:

1. Kill switch (balance ≤ 0): Any RUNNING instance is immediately auto-paused. Your data is preserved — Container Disk, Volume Disk, and Network Volume all remain intact. A reduced idle rate continues to accrue.
2. Backstop termination (balance further depleted): If your balance continues to drop while paused (from the idle rate and ongoing volume charges) and crosses a backstop threshold, the instance is auto-terminated. All ephemeral storage (Container Disk, Volume Disk) is permanently deleted.

Migrating data between instances

Since all instances support SSH, you can use standard tools like scp and rsync to move data — no platform-specific CLI tools required.

Upload files to an instance

scp -P <port> -i ~/.ssh/your_key ./local-file.tar.gz root@<host>:/workspace/

Download files from an instance

scp -P <port> -i ~/.ssh/your_key root@<host>:/workspace/data.tar.gz ./local-backup/

Transfer between two instances (via local staging)

To move data from Instance A to Instance B, pull from A to your local machine, then push to B:

rsync -avz -e "ssh -p <port_A> -i ~/.ssh/your_key" \
  root@<host_A>:/workspace/models/ ./staging/

rsync -avz -e "ssh -p <port_B> -i ~/.ssh/your_key" \
  ./staging/ root@<host_B>:/workspace/models/

Replace <host_A>, <port_A>, etc. with the connection details from each instance's details panel. For large transfers, consider compressing first with tar -czf before transferring.

Deleting a Network Volume

To delete a volume, navigate to the Network Volumes tab and click the delete icon on the volume card. A confirmation modal requires you to type the volume name before proceeding.

Deleting a volume stops its monthly billing immediately. If an instance currently has this volume mounted, the instance continues to run but the /workspace mount may become unavailable.

FAQ