Connecting to a GPU server over SSH: keys, first login and files

SSH is the main way to work with a server: it gives you a terminal, and SFTP and scp move files. A key is safer than a password, so set it up before the first login.

1. Create a key

ssh-keygen -t ed25519 -C "gpu"
# public key: ~/.ssh/id_ed25519.pub

On Windows with OpenSSH the command is the same, in PowerShell. Never share the private key; only the public one is sent.

2. Add the public key to your account

Open the Keys section in your account, paste the contents of the .pub file and save. Then select the key in the rental form: keys are chosen when the server is launched, so add it before you rent.

3. Connect

The exact address, user and port are on the server card: copy the connection command from there.

ssh <user>@<address> -p <port>

4. Check the GPU

nvidia-smi

It shows the card model, memory and driver version. If there is no output, contact support with the command output.

Files

scp model.safetensors <user>@<address>:/workspace/ -P <port>
scp -P <port> <user>@<address>:/workspace/result.png ./

For large files, use rsync (it can resume) or download directly on the server with wget. Download results before you stop or delete the server.