Network Volumes

Keep your data safe and accessible across GPU sessions with Hyperbolic's persistent network storage volumes.

What are Network Volumes?

Network volumes are persistent storage drives that stay attached to your data even when your GPU instances shut down. Think of them as external hard drives for the cloud – perfect for storing datasets, model checkpoints, and any files you want to keep between training sessions.

Whether you're a student working on a semester project or a researcher managing large datasets, network volumes ensure your work is always there when you need it.

When to Use Network Volumes

Network volumes are perfect for:

Large datasets that take time to download and prepare
Model checkpoints you want to save during long training runs
Project files that need to persist between sessions
Shared storage for collaborative research projects
Backup storage for important experimental results

You can always skip adding a volume if you're just running quick experiments or testing code.

Getting Started

Creating Your First Volume

When launching a GPU instance, you'll see the volume attachment step in your deployment wizard:

Choose "Create New Network Volume" if this is your first time
Give your volume a name – something memorable like "mnist-dataset" or "thesis-project"
Select your storage size using the slider (256GB to 8TB available)
Click "Create Volume" and we'll set it up for you

The volume will be ready to use as soon as your GPU instance starts.

Using an Existing Volume

Already have volumes from previous sessions? Perfect! You can reattach them:

Choose "Attach Existing Network Volume"
Select from your available volumes – you'll see the name, size, and location
Check the hourly cost for each volume option
Click "Attach and Continue" to connect it to your new instance

Only volumes in the same region as your GPU will be shown for the best performance.

Skipping Volume Attachment

Running a quick test or don't need persistent storage? No problem:

Click "Skip" to continue without a volume
You can always add volumes to future deployments when you need them

Volume Specifications

Size Options

Minimum: 256GB – great for small datasets and code projects
Maximum: 8TB – perfect for large-scale research datasets
Increments: Choose any size that fits your needs

Regional Availability

Volumes are region-specific for optimal performance
Create volumes in the same region as your GPU instances
Your existing volumes will only appear when deploying in matching regions

Pricing

Network volumes are billed hourly at $0.0008/TB/hour for storage volumes. You'll see the exact hourly cost for each volume option before making your selection.

Mounting Your Volume

After attaching a volume to your GPU instance, you'll need to mount it manually to access your data. Here's how:

Step 1: Connect to Your Instance

SSH into your GPU instance using the connection details from your instance's Details panel:

ssh user@your-instance-ip

Step 2: Configure the File System

Add your storage volume configuration to /etc/fstab. You'll need your volume's virtual IP address, which you can find in:

The "Mounting your storage volume" dropdown in your instance Details panel
Your volume entry on the Storage page

# Replace <storage-vip> with your actual Storage virtual IP
echo "<storage-vip>:/data /data nfs rw,nconnect=16,nfsvers=3 0 0" | sudo tee -a /etc/fstab

Step 3: Mount the Volume

Run the mount command to connect your volume:

sudo mount -a

Step 4: Verify the Mount

Confirm your volume mounted successfully:

df -h

You should see your /data directory connected to your storage volume's virtual IP.

Working with Your Mounted Volume

Once mounted, you can use your volume like any other directory:

# Access your volume at /data
import os

# List files in your volume
print(os.listdir("/data"))

# Save files to your volume
with open("/data/results.txt", "w") as f:
    f.write("Training completed successfully!")

# Create project structure
os.makedirs("/data/my-project/models", exist_ok=True)
os.makedirs("/data/my-project/datasets", exist_ok=True)

Best Practices

Organize your data:

# Create a clear folder structure
mkdir -p /data/my-project/{datasets,models,outputs,logs}

Save checkpoints regularly:

# Save model checkpoints to your volume
import torch
torch.save(model.state_dict(), '/data/my-project/models/checkpoint_epoch_10.pth')

Keep important outputs:

# Save results and visualizations
import matplotlib.pyplot as plt

# Generate your plots
plt.savefig('/data/my-project/outputs/training_curve.png')

Volume Management

Lifecycle

Create: Set up new volumes during instance deployment
Attach: Connect existing volumes to new instances
Detach: Volumes persist when instances shut down
Reuse: Attach the same volume to multiple deployments over time

Data Persistence

Your data stays safe on the volume even when:

Your GPU instance stops running
You switch to a different GPU type
You deploy in the same region weeks later

Sharing and Collaboration

Each volume belongs to your account and can be attached to any of your GPU instances in the same region. Perfect for maintaining consistent environments across different experiments.

Common Use Cases

Machine Learning Workflows

Dataset Storage:

# Store your datasets on a volume for reuse
dataset_path = "/data/imagenet"

# Download once, use everywhere
if not os.path.exists(dataset_path):
    download_imagenet(dataset_path)

# Load data in any session
from torch.utils.data import DataLoader
from torchvision.datasets import ImageFolder
train_loader = DataLoader(ImageFolder(dataset_path))

Model Development:

# Keep model versions organized
model_dir = "/data/my-research/models"

# Save different experiments
import torch
torch.save(model, f"{model_dir}/experiment_v{version}.pth")

# Load previous work
previous_model = torch.load(f"{model_dir}/best_model.pth")

Research Projects

Long-term Studies:

Store experimental data across multiple sessions
Maintain version control for research code
Keep research papers and documentation organized

Collaborative Work:

Share consistent environments between team members
Maintain standardized dataset versions
Preserve experimental results for peer review

Troubleshooting

Volume Not Appearing

If your volume doesn't show up in the attachment list:

Check the Storage tab: Verify your volume exists in the Storage section
Verify ownership: Only your volumes are visible
Wait for creation: New volumes may take a moment to appear

Mounting Issues

If you can't mount your volume:

Check NFS installation: Ensure nfs-common is installed with sudo apt install nfs-common
Verify virtual IP: Double-check the virtual IP address in your Storage details
Check fstab syntax: Ensure the /etc/fstab entry follows the correct format
Try manual mount: Use sudo mount -t nfs <storage-vip>:/data /data -o rw,nconnect=16,nfsvers=3

Access Issues

If you can't access your volume files after mounting:

Check the mount path: Your data should be accessible at /data
Verify permissions: Ensure your user has read/write access to /data
Check mount status: Run df -h to confirm the volume is properly mounted
Restart if needed: Sometimes remounting helps: sudo umount /data && sudo mount -a

Performance Considerations

For best performance:

Use appropriate mount options: The recommended nconnect=16,nfsvers=3 options optimize throughput
Organize files efficiently to minimize directory traversal
Consider volume size for your workload requirements