The disk of a GPU instance is deleted when the deployment stops, and a volume is how you keep data between deployments. A volume is a block device that lives in one provider region, is attached when you launch an instance there, and keeps its data after that instance is gone. Volumes exist on Verda and Hyperstack only.
QuantaCloud, Massed Compute and Vast.ai have no volumes. On those providers, copy data off the instance before you stop it.
Create a volume
Open Create Volume in the sidebar.
Select the cloud provider
Pick Verda or Hyperstack. Each card lists the provider’s volume types, size range and regions.
Name it
A random name such as swift-falcon-234 is filled in. Change it to something you will recognize.
Choose the region and type
The region must be the one where you plan to launch, because a volume can only be attached in its own region and cannot be moved. Verda volumes are in Finland. The volume types on offer depend on the provider and region.
Set the size
The default is 50 GB, and the allowed range for the provider is shown next to the field.
Check the cost and create
The form shows the hourly rate, including any platform fee, and an estimated runtime: how long your current balance would pay for this volume alone. Click Create Volume.
Attach it when you launch
On the deploy page of a Verda or Hyperstack offer, pick the volume from the volume list. Only volumes that are Available and in the same region as the offer are listed. A volume cannot be attached to an instance that is already running, and it cannot be detached from one: it stays attached until the deployment stops.
Through the API, pass the volume’s ID in volume_ids when you create the deployment. See Volumes in the API reference.
Nothing formats or mounts a volume for you. After the instance is running, find the new disk and mount it yourself. Device names differ between providers, so look for the disk that matches the volume’s size and has no mount point:
On a new, empty volume only, create a file system. This erases everything on the device, so never run it on a volume that already holds data:
Then mount it, on this and every later deployment:
Replace /dev/vdb with the device name lsblk showed. To check whether a volume already has a file system before formatting, run sudo blkid /dev/vdb: if it prints TYPE="ext4", skip mkfs and mount it.
Statuses
The Volumes page and the API report these statuses:
When the deployment stops
Stopping a deployment deletes the instance and its disk, then releases the volume, which goes back to Available with its data intact. Launch another instance in the same region with the volume attached to pick up where you left off. If no GPU is in stock in that region, you have to wait for stock there or copy the data out. The stop instance guide lists what to save from the instance disk before you stop.
Delete a volume
On the Volumes page, open the menu on a volume and choose Delete. Only Available or Error volumes can be deleted, and the dashboard asks you to confirm first. Deletion is permanent: there are no snapshots or backups to restore from, so copy anything you need off first (moving data to and from a cloud GPU).
A deleted volume stops being billed straight away. If the provider cannot remove it yet, for example during its maintenance, the volume shows as Deleting until the provider confirms. You do not need to do anything.
Billing
A volume is billed every hour from the moment it is created until it is deleted, whether it is attached or not.
If your balance cannot pay for the next volume-hour, the volume is deleted permanently, and no email is sent. A volume attached to a running deployment is kept while that deployment runs, and the charge is tried again; once the deployment has stopped, the next unpaid volume-hour deletes it. Keep a balance, turn on auto top-up, and keep a copy of anything you cannot lose somewhere else. 
