Skip to main content

Storage Volumes

Persistent block storage for GPU instances.

Overview​

Volumes provide network-attached storage that persists after instance termination.

Key characteristics:

  • Persistent (survives instance termination)
  • Billed hourly per GB
  • Must be in same cloud region as instance
  • 50 GB to 10,000 GB capacity
  • NVMe SSD-backed

Use for:

  • Datasets and training data
  • Model weights and checkpoints
  • Database storage
  • Application data

Don't use for:

  • Temporary files (use instance storage)
  • Small configuration files
  • OS and system files

Creating Volumes​

During Instance Launch​

  1. Expand "Attach Storage" section when configuring instance
  2. Toggle to "Create New Volume"
  3. Enter volume name
  4. Set size (50-10000 GB)
  5. Review additional cost
  6. Launch instance

Volume is created and automatically attached.

Standalone Creation​

  1. Go to Resources page, Volumes tab
  2. Click "Create Volume"
  3. Enter name
  4. Select region (must match future instance region)
  5. Choose size
  6. Click Create

Important: Volume begins billing immediately upon creation.

Regional Requirements​

Critical: Volumes must be in the exact same cloud region as the instance.

Example: Instance in oslo-norway-1 requires volume in oslo-norway-1.

Cross-region attachment is not supported. The console only shows compatible volumes during instance launch.

Attaching Volumes​

Volumes can only be attached during instance launch.

  1. Configure new instance
  2. In "Attach Storage", select "Use Existing Volume"
  3. Choose volume from dropdown (only shows compatible volumes)
  4. Launch instance

Compatible volumes must be:

  • In same cloud region
  • Status: active
  • Not attached to another instance

Default mount point: /mnt/volume

Verify mount:

df -h | grep /mnt/volume
ls -la /mnt/volume

Detaching Volumes​

Volumes detach automatically when instance is terminated.

Safe detach procedure:

  1. SSH into instance
  2. Stop applications using the volume
  3. Unmount volume:
umount /mnt/volume
  1. Verify unmounted:
df -h | grep /mnt/volume  # Should return nothing
  1. Terminate instance

Volume returns to active status and can be attached to a different instance.

Managing Data​

Upload Data​

Via SCP:

scp -i key.pem file.tar.gz root@IP:/mnt/volume/
scp -i key.pem -r /local/dir/ root@IP:/mnt/volume/

Via rsync:

rsync -avz -e "ssh -i key.pem" /local/data/ root@IP:/mnt/volume/data/

Download from internet:

cd /mnt/volume
wget https://example.com/dataset.tar.gz

From S3/cloud storage:

apt-get install -y awscli
aws s3 sync s3://bucket/data /mnt/volume/data

Download Data​

scp -i key.pem root@IP:/mnt/volume/results.tar.gz .
scp -i key.pem -r root@IP:/mnt/volume/results/ ./local/

Check Usage​

# Volume usage
df -h /mnt/volume

# Directory sizes
du -h --max-depth=1 /mnt/volume

# Find large files
du -ah /mnt/volume | sort -rh | head -20

Pricing​

Rate: Approximately €0.0001 per GB per hour (varies by region)

Examples:

  • 100 GB: €0.01/hour (€7.20/month)
  • 500 GB: €0.05/hour (€36/month)
  • 1 TB: €0.10/hour (€72/month)

Billing starts immediately on creation and continues until deletion, even when not attached.

Deleting Volumes​

  1. Ensure volume is detached (status: active)
  2. Click Delete on Volumes tab
  3. Confirm by typing volume name

All data is permanently deleted. This action cannot be undone.

Troubleshooting​

Volume not showing during instance launch:

  • Verify volume region matches instance region exactly
  • Check volume status is active (not attached or deleting)
  • Ensure not attached to another instance

Volume not mounted:

# List disks
lsblk

# Manual mount
mkdir -p /mnt/volume
mount /dev/sdb /mnt/volume

Cannot unmount (device busy):

# Find processes using volume
lsof /mnt/volume
fuser -m /mnt/volume

# Stop processes
fuser -km /mnt/volume

# Retry unmount
umount /mnt/volume

Volume full:

# Find large files
du -ah /mnt/volume | sort -rh | head -20

# Delete unnecessary files
rm -rf /mnt/volume/old-data/

# Compress data
tar czf archive.tar.gz large-directory/

Filesystem errors:

# Check and repair
fsck /dev/sdb
fsck -y /dev/sdb  # Auto-repair

Best Practices​

  • Name volumes descriptively (purpose, project, date)
  • Delete volumes when no longer needed
  • Backup important data to object storage (S3, GCS)
  • Always unmount before terminating instance
  • Use instance storage for temporary files
  • Create volumes only in regions you'll use