Articles » SSH guide

How to SSH Into a Vast.ai GPU Instance

SSH is the most reliable way to install packages, move code, inspect logs, and manage long-running jobs on a rented GPU.

Create a dedicated key

ssh-keygen -t ed25519 -C "vast-gpu" -f ~/.ssh/vast_gpu

Upload only vast_gpu.pub to your Vast.ai account. Never upload or share the private key.

Connect to the instance

Choose an SSH-enabled template. After the instance starts, copy the connection command shown by Vast. The platform often maps container port 22 to a different public port, so use the displayed value:

ssh -i ~/.ssh/vast_gpu -p MAPPED_PORT root@HOST_ADDRESS

Verify the current host fingerprint before accepting it, especially when an address may have been reused.

Confirm GPU and disk access

nvidia-smi
pwd
df -h

If nvidia-smi fails, fix GPU access before installing more software. Check free disk space before downloading model weights.

Copy project files

scp -i ~/.ssh/vast_gpu -P MAPPED_PORT train.py root@HOST_ADDRESS:/workspace/

For repeated transfers, use rsync with the same SSH key and port so unchanged files are skipped.

Forward a private service

ssh -i ~/.ssh/vast_gpu -p MAPPED_PORT -L 8888:127.0.0.1:8888 root@HOST_ADDRESS

Open http://127.0.0.1:8888 locally. Tunneling is safer than publishing an unauthenticated notebook or development API.

Common failures

• Permission denied: confirm the private key matches the public key on the account.
• Connection refused: verify the mapped port and instance state.
• Timeout: check host status, local firewall, and VPN rules.
• Host key changed: verify the new instance fingerprint before editing known_hosts.

Consult the official Vast.ai quickstart for the current console workflow. Next, learn how to preserve data between sessions.