iQ Studio

Show Performance, Spark Imagination.

It helps users quickly understand, explore, and prototype ideas by showcasing the platform’s performance and capabilities—inspiring innovation through hands-on experience.

Note

iQ-Studio is suitable for a range of products developed based on the Innodisk Qualcomm Dragonwing SoC.

Pick Your Path

Three common starting points. Pick the one that matches your goal today; you can always come back for another.

🚀 Run a demo	🛠 Develop & customize	📊 Evaluate the platform
For FAEs and evaluators. See an AI inference demo running on the hardware in under ten minutes.	For application developers and ML engineers. Swap models, change inputs, or deploy your own trained model.	For PMs and integrators. Review measured performance on real workloads and how iQ-Studio fits the platform stack.
Step 1: Getting Started Step 2: 30-Second Demo Step 3: Applications	Step 1: Getting Started Step 2: SDKs to customize, or Model Deploy to bring your own model	Step 1: Benchmarks Step 2: Core Software Stack & Architecture

Getting Started

Follow these two steps in order. Once they are done, every demo and tutorial below will run.

Step 1. Boot your platform

Flash an image and boot the device. See the Starting Guides; for the Q911 series, start with the EXMP-Q911 Starting Guide.

Two image options are available (see OS Images for details):

• Innodisk BSP (Yocto), recommended. iQ-Studio is developed and validated on the Yocto-based BSP.
• Innodisk Ubuntu image. Earlier versions have been verified to run iQ-Studio; use it only if you need an Ubuntu environment.

Step 2. Install iQ Studio

On the platform, run:

git clone https://github.com/InnoIPA/iQ-Studio.git
cd iQ-Studio
./install.sh

Note: The latest iQ-Studio requires BSP 2.5.x (QLI 2.0). On BSP 2.3.x (QLI 1.8), run git checkout v0.0.10 in the iQ-Studio directory before ./install.sh. Tag v0.0.10 and earlier are the releases for these platforms.

30-Second Demo

With the platform booted and iQ-Studio installed, two commands are enough to see a vision-language model running on a live UVC Camera feed. For the detailed walkthrough, see iQS-VLM.

The demo consists of two separate services, OGenie and iqs-vlm-demo. Each runs in its own terminal, so please open two terminals and run one command in each.

Important: Make sure the OGenie server is up and running before you start iqs-vlm-demo.

In Terminal 1, launch the OGenie API server:

iqs-launcher --autotag iqs-ogenie

In Terminal 2, display VLM predictions on the monitor in real time:

iqs-launcher --autotag iqs-vlm-demo

For other applications and the launch commands behind them, see Applications.

Note: iqs-launcher runs every iQ-Studio application by pulling (online) or loading (offline) the matching Docker image or IPK package. See How to use iqs-launcher.

Explore Documentation & Resources

iQ Studio resources are grouped into categories based on functionality:

Categories	Description	Topic
Starting Guides	Quick-start guides for everything.	Starting Guides Overview Q911 Quick Start Guide Q911 Image Flashing Guide Qualcomm OTA Guide
AVL(Approved Vendor List)	Provides guidance on verifying that the driver starts correctly on the system and quickly demonstrating the validated results.	Approved Vendor List GMSL Camera MIPI Camera
Applications	Application-level examples focused on specific use cases and vertical scenarios.	Applications Overview iQS-VLM iQS-Streampipe YOLOv10n INT8 Inference on GPU and NPU
Model Deploy	End-to-end guides for turning trained AI models into target-ready artifacts, covering quantization, conversion, quality validation, and on-device inference.	Model Deploy Overview Model Deploy: How to Convert, Optimize, and Perform Inference with YOLO26 Models
SDKs	Documentation and examples on how to use the SDKs effectively.	SDKs Overview iQS-VLM: How to Interact with the OGenie Server through Open WebUI iQS-Streampipe: How to Change the Custom Model and Video Source iQS-OGenie: Run Your Own Demo with OGenie Server
Benchmarks	Reproducible performance tests for multi-stream YOLO, generic streaming pipelines, and perception models on Qualcomm QCS9075.	Benchmarks Overview InnoPPE Benchmark Multi-stream Inference Status Perception AI Benchmark

Core Software Stack & Architecture

This section is for readers who want to understand what runs on the board and how it is built. You do not need it to run the demos above.

Software Stack

What runs on the platform, from hardware up to the applications:

• Hardware & Firmware: Qualcomm Dragonwing QCS9075 SoC and low-level firmware.
• Kernel Space: Powered by Qualcomm Linux, integrated with our custom Inno DTB/drivers and Yocto environments.
• User Space: Seamlessly supports 3rd-party LLM SDKs, device management (iCAP), and inno AVL. At the very top sits the iQS-App layer (VLM, Streampipe, YOLO, OGenie).

How the Software Is Built: From Upstream to iQ-Studio

Where each piece of that stack comes from. Four layers build on one another, each on top of the layer below:

Layer	What it contributes	Source
Upstream	Linux kernel LTS, Yocto Project releases, and open source components	kernel.org and the Yocto Project
Qualcomm Linux (QLI)	Qualcomm Linux kernel, firmware, and SoC drivers, packaged as the QLI distribution	Qualcomm Linux
Innodisk BSP	The inno meta-iq Yocto layer, adding the inno DTB and downstream drivers on top of QLI for EXMP-Q911 and the QCS9075 iQ-9075 EVK	InnoIPA/meta-iQ__manifest
iQ-Studio	Application demos, AVL checks, benchmarks, and iqs-launcher compatibility handling	This repository

The Innodisk BSP is the layer that turns a generic QLI distribution into a validated system image for Innodisk hardware. It is published as a reproducible Yocto manifest, so the exact BSP that iQ-Studio targets can be rebuilt from source.

BSP & QLI Releases

Each QLI release line is tied to a specific Linux kernel and Yocto Project version:

Linux Kernel	Yocto Project	Qualcomm Linux (QLI) Release
6.6 LTS	4.0 Kirkstone	QLI 1.x
6.6 LTS	5.0 Scarthgap	QLI 1.x
6.18 LTS	Wrynose (Master)	QLI 2.x

Note: Latest iQ-Studio targets QLI 2.x (BSP 2.5.x). QLI 1.x platforms (BSP 2.3.x) use tag v0.0.10 or earlier.

Note: For the full upstream timeline, including the Kirkstone track and Qualcomm's Mainline development branch, see the Qualcomm Linux Roadmap.

Each Innodisk BSP release carries one QLI release and lands one quarter after it. That quarter is used for board bring-up, driver enablement, and full validation on Innodisk hardware.

The BSP version number reads as vMAJOR.MINOR.PATCH:

Field	Tracks
MAJOR	Hardware change
MINOR	QLI version carried by the release
PATCH	Patch level

Note: Versions and dates beyond the current release are planned schedules and may change together with the Qualcomm Linux Roadmap.

OS Images

Besides the Yocto-based Innodisk BSP, Innodisk also provides integrated Ubuntu images for development.

Note: To keep all IO functions working correctly, use the Innodisk-provided Ubuntu image. See InnoIPA/iQ-ubuntu__manifest for the available image versions. The official Ubuntu image from Qualcomm can boot on the platform but does not guarantee full IO support.

In the News

Media coverage, deployments, and ecosystem case studies are collected in In the News.

Related Repositories

iQ-Studio integrates with several sibling repositories. Each owns a specific layer of the platform stack.

• InnoIPA/meta-iQ__manifest — Yocto-based BSP baseline for the platform.
• InnoIPA/iQ-Cam__manifest — Camera driver patches consumed by the AVL tutorials.
• InnoIPA/iQ-ubuntu__manifest — Reference for Innodisk-provided Ubuntu image versions.
• InnoIPA/iQ-Foundry — End-to-end model quantization and on-device validation for custom models.

Changelog

Please refer to the Changelog for all updates.

License

This project is licensed under the MIT License. See LICENSE for details.