Connecting

Open a GPU from your list, then the Connect tab. Everything below is there, with copy buttons.

SSH

The Connect tab shows a command such as:

ssh -p 30568 [email protected]

and the root password. Paste the command into any terminal and enter the password when asked.

Use a key instead of a password

Add your public key under Settings → SSH keys before you launch, or tick it in the launcher. GPUs launched with a key accept it without a password:

ssh -p 30568 -i ~/.ssh/id_ed25519 [email protected]

Keys are added when a GPU is created, so add a key first, then launch.

VS Code and Cursor

The Connect tab gives you a block for your SSH config. Add it to ~/.ssh/config:

Host flint-my-gpu
  HostName 203.0.113.7
  Port 30568
  User root

Then in VS Code or Cursor run Remote-SSH: Connect to Host… and pick flint-my-gpu, or use Open in VS Code on the Connect tab. Open the /workspace folder.

The host name and port belong to that GPU. If you terminate it and launch another, update the block.

Browser terminal

Terminal (or press T with the GPU selected in the list) opens a full shell in a dialog. It's the same machine as SSH, so nothing to install.

JupyterLab

Jupyter opens JupyterLab in a new tab, already signed in. Keep your notebooks in /workspace (or a shared volume) so they survive pause and resume. PyTorch is installed and the notebook sees your GPU:

import torch
print(torch.cuda.is_available(), torch.cuda.get_device_name(0))

Check your GPU

nvidia-smi

On a slice, the memory total shown is your slice's limit.

Troubleshooting

• "Connection refused" right after launch: wait until the GPU shows Running; SSH starts last.
• Password rejected: copy it from the Connect tab again. Each GPU has its own password.
• Host key warning after relaunching: a new GPU can reuse an old address and port. Remove the old entry with ssh-keygen -R "[203.0.113.7]:30568".