Skip to main content
GPU instances are ephemeral — delete one and the data goes with it. Use storage volumes to persist code, datasets, and model weights across deploys. Pay for compute only when you need it.

​
1. Create a storage volume

runcrate volumes create --name dev-workspace --size 100
runcrate volumes regions     # Check available regions

​
2. Deploy with the volume attached

runcrate instances create --name devbox --gpu RTX4090 --template ubuntu-devbox --storage dev-workspace
runcrate instances status devbox
The volume is mounted at /workspace/.

​
3. Set up your environment

runcrate ssh devbox -- "cd /workspace && git clone https://github.com/you/your-project.git"
runcrate ssh devbox -- "cd /workspace/your-project && pip install -r requirements.txt"
Store model weights on the volume to avoid re-downloading: 
runcrate ssh devbox -- "mkdir -p /workspace/models"
runcrate ssh devbox -- "huggingface-cli download meta-llama/Llama-3.1-8B-Instruct \
  --local-dir /workspace/models/llama-8b"

​
4. Upload and download files

runcrate cp ./data.csv devbox:/workspace/your-project/
runcrate cp devbox:/workspace/your-project/output/ ./results/

​
5. Tear down (data persists)

runcrate instances delete devbox

​
6. Resume later

runcrate instances create --name devbox --gpu RTX4090 --template ubuntu-devbox --storage dev-workspace
Your code, models, and data are exactly where you left them at /workspace/.

​
Manage volumes

runcrate volumes list
runcrate volumes resize <volume-id> --size 200
runcrate volumes delete <volume-id>

​
Tips

  • Keep model weights on the volume — a 70B model takes 30+ minutes to download but loads from the volume in seconds.
  • The ubuntu-devbox template includes CUDA, Python, git, and common CLI tools.
  • Volumes are billed separately from instances — check runcrate billing usage for a breakdown.
  • Use runcrate billing balance to check credits before deploying.

​
Cleanup

runcrate instances delete devbox
runcrate volumes delete <volume-id>    # Only when you no longer need the data