---
title: "Connect to a Windows GPU node"
description: "RDP into a Windows GPU node the same way as any Windows node, then confirm the NVIDIA driver loaded."
content_type: "how-to"
product: "gpu-node"
platform: "MyAccount"
url: "/myaccount/gpu-node/how-to/connect-to-a-windows-gpu-node/"
topic: ["create-and-connect"]
---

# Connect to a Windows GPU node

Connecting to a Windows GPU node is the same RDP flow as any Windows node. The only addition is
confirming the driver once you are in; for the full connect reference, RDP clients, retrieving
the Administrator password, security group rules, see
[Connect to a Windows node](/myaccount/node/how-to/connect-to-a-windows-node/).

## Connect over RDP

Open your RDP client and enter the node's public IP, `Administrator` as the user, and the
password from the secure password email.

## Check that the NVIDIA driver sees the card

After your first login, confirm the GPU is present and the driver loaded.

**Device Manager.** Right-click **Start > Device Manager**, expand **Display adapters**, and
confirm the GPU is listed, for example **NVIDIA A40** or **NVIDIA L40S**. Right-click it,
choose **Properties > Driver**, and confirm a driver version and date are present with a status
of "This device is working properly."

**`nvidia-smi` in PowerShell.** The Windows driver installer also installs `nvidia-smi.exe`:

```powershell
nvidia-smi
```

A healthy node prints the card, driver version, CUDA version, memory, and utilization, the same
fields as on a Linux GPU node. With nothing running, utilization reads near `0%`. If Device Manager shows **Microsoft Basic Display Adapter**
instead of the card, or `nvidia-smi` reports it cannot communicate with the driver, see
[Troubleshooting GPU Node](/myaccount/gpu-node/troubleshooting/gpu-node/).
Once the card checks out and your software is installed, save the node as an image. Every new
node then starts with the same setup; see
[Save and reuse a GPU image](/myaccount/gpu-node/how-to/save-and-reuse-a-gpu-image/).

## WDDM and TCC display modes

The NVIDIA driver on Windows runs a card in one of two modes. **WDDM**, short for Windows
Display Driver Model, lets the card draw a desktop, and RDP needs it. A Windows GPU node starts
in **WDDM**. **TCC**, short for Tesla Compute Cluster, turns the card into a compute-only device:
it can give a small speed-up for a job that needs no desktop, but the RDP desktop no longer
draws through that card. Stay on **WDDM** unless you run compute jobs only and never need the
desktop.

To switch, open PowerShell as an administrator and run one of these, then reboot the node:

```powershell
nvidia-smi -dm 0     # WDDM, display and compute
nvidia-smi -dm 1     # TCC, compute only; the RDP desktop no longer draws through this card
```

To check the current mode, run `nvidia-smi`: the mode shows next to the card's name.

Everything else, password changes, RDP client tuning, security group rules, Access Console
recovery, works the same as any Windows node; see
[Connect to a Windows node](/myaccount/node/how-to/connect-to-a-windows-node/).