GPUtw Docs

Getting Started

Welcome to GPUtw. Follow this path to launch, connect, save work, and shut down your first GPU instance.

Open dashboard

Welcome to GPUtw

GPUtw provides dedicated GPU containers with dashboard, SSH, Web UI, Vault storage, port management, and scoped API keys. The fastest path is to add credits, choose a GPU and template, wait for the instance to reach RUNNING, then connect through Web UI or SSH.

Tip

If you only want a notebook or web app, start with a PyTorch/Jupyter, ComfyUI, or Open WebUI template and use the Web UI button. If you want a clean shell, use Ubuntu or CUDA and connect over SSH.

Before you deploy

Prepare these items before launching the first instance:

  • Account: register, sign in, and verify your email
  • Credits: add enough prepaid balance for at least one hour of committed usage
  • SSH key: optional for browser-only work, required for SSH access; add your public .pub key in Dashboard → SSH Keys
  • Work location: plan to keep datasets, checkpoints, and outputs under /vault if they should persist

Choose a GPU and template

Open Dashboard → Deploy and work from top to bottom: choose a GPU model, select an available machine, choose a template, review the hourly rate, then deploy.

  • GPU model: pick based on VRAM, architecture, and price
  • Machine: choose an online node with enough available capacity
  • Template: pick the software stack closest to your workload
  • Ports: templates with browser apps usually expose a Web UI port such as 8080

Connect to the instance

After deploy, the instance moves through provisioning states and eventually reaches RUNNING. Use the connection method that matches your workflow:

  • Web UI: open Jupyter, ComfyUI, or other browser tools from the Instances page
  • SSH: copy the SSH command from the instance row after adding your public key
  • API: create a scoped API key when automation needs to list, create, stop, or manage instances
Example
ssh pod-<instance-id>@ssh.gputw.ai -p 2222

Save work and expose services

Use /vault for persistent artifacts. Use HTTP port access modes when sharing browser services, and raw TCP/UDP only for services that already have their own authentication.

  • Private: best for personal notebooks and dashboards
  • Unlisted + password: best for sharing a link with a collaborator
  • Raw TCP/UDP: best for non-HTTP services such as custom protocol servers

Stop or delete when finished

Stop the instance when you may return later. Delete it when you no longer need the instance record or workspace. Billing stops when the instance is no longer running.

Warning

Long-running instances continue to consume credits while RUNNING. Check the Instances page before leaving a workload idle.

Where to go next