Skip to main content
Qualcomm® Device Cloud (QDC) gives you remote access to real Qualcomm devices, so you can try an Edge Impulse model on Qualcomm hardware before you buy a board. QDC is free to use: you request minutes for each device type you want, and QDC only counts them while a session runs. This guide uses the Qualcomm Dragonwing™ RB3 Gen 2 Development Kit as the example. You’ll start a QDC session, install the Edge Impulse Linux CLI on the device, connect it to Edge Impulse Studio, and run inference on a static image. The same steps work on the other Linux IoT devices in QDC, with the differences listed in Available devices.
The QDC confirmation page after minutes are provisioned for Dragonwing RB3 Gen 2 devices, with a Start Interactive Session button

QDC provisions free minutes for each device type you request.

​
Prerequisites

​
Available devices

These devices were listed in the QDC device catalog in October 2026. The catalog changes as devices are added, so check it for the current list. Early access devices may need you to request access first.

​
IoT devices (Linux)

You can install the Edge Impulse Linux CLI on these devices and follow the rest of this guide.

​
Mobile and compute devices

QDC also hosts Snapdragon® mobile reference designs running Android, and compute reference designs running Windows. The Linux CLI steps in this guide don’t apply to them.
  • Mobile (Android): Snapdragon 8 Elite Gen 5, 8 Elite, 8 Gen 3, 8 Gen 2, 8 Gen 1, 7+ Gen 3, and 4 Gen 2. To run Edge Impulse models on Android, see the Android tutorials.
  • Compute (Windows): Snapdragon X2 Elite, X Elite, and X Plus.

​
1. Start an interactive session

On the QDC home page, select Try Now on the RB3 Gen 2 banner, or go to Devices and select Qualcomm Dragonwing™ RB3 Gen 2. If the device doesn’t offer an interactive session, request minutes for it first.
The QDC home page with a Try Now button on the Advanced on-device AI with Qualcomm Dragonwing RB3 Gen 2 banner
Then create the session:
  1. Select Interactive session.
  2. On the Device Configuration page, select a build and make sure Wi-Fi is enabled. The CLI needs internet access to install packages and reach Edge Impulse Studio.
  3. On the Test Configuration page, enter a session name and select Interactive shell + SSH.
  4. Under Package Upload, upload a zip file that contains your test image. You can also upload files later from the session page.
  5. Select Create Session. QDC prepares the device and opens a console when the session starts.
The QDC Test Configuration step with example.zip uploaded under Package Upload for an RB3 Gen 2 session

​
2. Install the Edge Impulse Linux CLI

QDC resets the device after every session, so install the CLI at the start of each one. On the Qualcomm Linux devices, run the Edge Impulse setup script. It installs Node.js, the Edge Impulse Linux CLI, and SoX into ~/edge-impulse-tools and adds them to your ~/.profile:
The script has been tested on Qualcomm Linux 1.3 through 2.0, and prints a warning if it detects another version.
The QDC console on an RB3 Gen 2 as the setup script installs Node.js and the Edge Impulse Linux CLI
Load the updated profile and check the installation:
The QDC console after the setup script finishes, with the source ~/.profile command highlighted
On the Ubuntu devices, install Node.js and the CLI with apt and npm instead, as described in the Edge Impulse Linux CLI installation steps.

​
3. Connect the device to Edge Impulse Studio

When prompted, log in to Edge Impulse and select your project. The --disable-camera option skips the camera prompt, because this guide uses a static image. To log in to a different account or project later, run edge-impulse-linux --clean. The device then appears on your project’s Devices page.
The Devices page in Edge Impulse Studio listing a QDC device with the type QUALCOMM_LINUX

​
4. Run inference on a static image

QDC copies files from Package Upload to /data/local/tmp/QDC_files/ on the device. Find your test image there and copy it to your home directory:
Run the model with the image as a simulated camera:
The runner downloads your model and runs inference on the image repeatedly, printing the predicted labels and confidence scores each time. Press Ctrl+C to stop it.

​
Next steps