Skip to main content

Connect to a Linux GPU Node

Connecting to a Linux GPU node is the same SSH flow as a regular Linux node. The only addition is the post-login GPU check.

For the full SSH connect reference — finding the IP, choosing password vs key, host fingerprints, key rotation — follow Connect to a Linux node. This page only documents the GPU delta.


Connect over SSH

Use the SSH command exactly as you would for a regular Linux node:

ssh <username>@<gpu-node-ip>

Or with an SSH key:

chmod 600 ~/.ssh/<private-key-file>
ssh -i ~/.ssh/<private-key-file> <username>@<gpu-node-ip>

The default user, password retrieval, and security group rules are identical to a regular node. See Connect to a Linux node.


Verify the GPU with nvidia-smi

nvidia-smi is the canonical verification command. Run it as your first action after login:

nvidia-smi

A healthy node returns the card name, driver version, CUDA version, memory, and utilization. For example:

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 570.86.10 Driver Version: 570.86.10 CUDA Version: 12.8 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
|=========================================+========================+======================|
| 0 NVIDIA A100-SXM4-80GB On | 00000000:00:1E.0 Off | 0 |
| N/A 32C P0 63W / 400W | 0MiB / 81920MiB | 0% Default |
+-----------------------------------------+------------------------+----------------------+

What to read from this output:

FieldWhat it tells you
Driver VersionInstalled NVIDIA datacenter driver.
CUDA VersionThe CUDA runtime version the driver supports. Your workload's CUDA must be ≤ this.
NameThe detected GPU model. Confirm it matches the card you ordered.
Memory-UsageIdle memory should be near 0MiB on a fresh node.
GPU-UtilShould be 0% on idle.
Persistence-MOn keeps the driver resident across application starts. Recommended for long-running jobs.

Confirm CUDA with nvcc (Optional)

To confirm the CUDA toolkit (compiler and headers) is installed on the host:

nvcc --version

On container-based images, nvcc may only be available inside the container, not on the host. That is expected — the host only needs the driver.


Everything Else

Tmux/screen for long-running sessions, SSH key rotation, security group hardening, and Access Console recovery work the same as on any Linux node — see Connect to a Linux node.


ResourceUse it for
Connect to a Linux nodeCanonical SSH connect reference (non-GPU steps).
Create a GPU nodeLaunch a GPU node.
Choose a GPU cardPick the right card before you connect.
Connect to a Windows GPU nodeRDP into a Windows GPU node.
Troubleshoot GPU NodesFix nvidia-smi errors, missing cards, CUDA mismatch, and OOM.
SSH Key ManagementCreate and rotate SSH keys.
Last updated on May 26, 2026.