Scale up, scale down: the Developer’s guide to Qualcomm IoT ecosystem

How do I get the highest possible performance for robotics, multi-camera installations and industrial deployment? Dragonwing IQ9 Series is here to help.

For industrial IoT, edge AI and robotics applications involving heavy workloads, the Dragonwing IQ9 series combines reliability, developer-friendly design and on-device AI of up to 100 dense TOPS.

The series of SoCs is optimized for industrial-grade edge AI, with the deterministic compute demanded in environments like AMRs and industrial automation. Besides Ubuntu, it runs upstream Linux with Yocto Support, helping developers, OEMs and ODMs get their IoT products to market with less effort and cost. This scale-up platform for high-performance and multi-system deployment is ideal for:

• Teams building industrial edge AI systems
  
  
• Multi-camera robotics and automation workloads
  
  
• Multimodal vision language models that can process both images and text
  
  
• Industrial automation architects and robotics leads
  
  
• Developers who need more I/O and higher throughput

The series includes the Dragonwing IQ-9075 for power efficiency with high performance in extreme environments. With dual neural processing units (NPUs), Dragonwing IQ-9075 is available in two configurations: a 100-TOPS variant requiring four power management integrated circuits (PMICs), and a 50-TOPS variant requiring two PMICs.

Designed to maximum specifications for IoT devices, the SoC ensures that your applications will have the memory, CPU/GPU/NPU cycles, display ports, video codecs, megapixels, physical interfaces and storage they need. Use it in industrial IoT applications such as the following:

• Edge AI boxes
• Industrial robots
• Drones
• Machine vision
• Industrial personal computers
• Autonomous mobile robots (AMRs)

The octa-core CPU of the SoC comprises 4 RTSS (real-time) cores for deterministic performance, and the unit is rated for temperatures between –40°C and +85°C. Besides high performance, it offers a wide variety of connections for peripheral devices. We provide a continuously expanding list of, for example, approved cameras and memory that you can use with it.

To simplify your design work and shorten your time to market, take advantage of the Dragonwing IQ‑9075 Module. It provides a pre‑integrated, industrial‑grade solution with the performance and flexibility of the Dragonwing IQ9 Series.

What is my entry point for compute-heavy machine vision and AI processing in industrial IoT applications? Dragonwing IQ8 series is the answer.

For less demanding IoT applications in less extreme environments, the Dragonwing IQ8 Series offers upstream Linux support for commercial products in industrial settings. The series offers heterogeneous compute on independent CPU, NPU, GPU, and DSP cores, plus a real-time, MCU-like subsystem.

Included in the Dragonwing IQ8 series is the Dragonwing IQ-8275. Like the Dragonwing IQ-9075, it features dual NPUs. Available configurations enable up to 40 TOPS for AI inference (requires four PMICs) and 20 TOPS (requires two PMICs).

The SoC is especially valuable in smart devices that depend on immersive multimedia, powerful on-device AI, and enterprise-grade security features at the edge. Its industrial IoT applications include the following:

• Factory automation
• Industrial gateways
• Edge AI boxes
• Autonomous mobile robots (AMRs)
• Industrial robots
• Drones
• Machine vision
• Industrial personal computers

How can I prototype and test extremely demanding, multi-camera applications for industrial IoT? Use Dragonwing IQ-9075 EVK.

Aligned with Dragonwing IQ9, the Dragonwing IQ-9075 EVK is your path from proof of concept to commercial product, especially on upstream Linux. The kit consists of a main board, supporting interfaces and optional mezzanine boards for additional features.

Built for mission-critical edge inference, the kit is designed for sustained, system-level efficiency, to handle high-throughput, multi-camera applications in automation and robotics. It maximizes AI performance on LLMs and VLMs in a small form factor. It is engineered for multi-sensor fusion and high-throughput AI, with up to seven concurrent camera feeds and industrial I/O (PCIe, TSN Ethernet, CAN-FD). It offers PCIe Gen 4 options and 2.5 Gb Ethernet on the core board, with additional Ethernet on mezzanines.

Sample applications for the Dragonwing IQ-9075 EVK include multi-camera streaming/encoding, multi-channel video decode and compose (video wall), and multi-stream inference.

Choose the Dragonwing IQ-9075 EVK when you’re ready to produce your most demanding applications. It offers you the headroom, I/O density and performance envelope you need so you can build toward deployment. Its combination of AI processing, expansion capabilities and compact form factor makes it a versatile single-board computer, especially with desktop Ubuntu. It also is ideal for prototyping in robotics due to its size and flexibility.

How can I start developing and testing factory automation, multiple camera feeds and industrial IoT applications? Try Dragonwing IQ-8275 EVK.

Development of industrial IoT leads into autonomous mobile robots (AMRs), gateways and edge AI boxes. It demands a robust workbench for prototyping and testing applications. With the Dragonwing IQ-8275 EVK, you can be sure of what you’re moving into production.

The EVK lets you advance from proof of concept to commercial product with the features, performance and lower power consumption of Dragonwing IQ8. It is well suited to building proofs of concept of edge AI boxes that use small/visual/large language models (SLMs, VLMs, LLMs). Moreover, it’s designed for aggregating an existing network of cameras and sensors, such as those used in industrial robots like drones, AMRs, cobots and palletizers.

For computer vision and multi-sensor fusion applications, the kit is made to connect to a range of peripherals through low- and high-speed connectors including USB, PCIe, DSI, SDIO, OSPI and I2S. MIPI CSI inputs let you manage as many as 12 concurrent video streams at 1080p. Besides wired 2.5 Gbps Ethernet, you can connect through onboard Wi-Fi 6E and Bluetooth® 5.3 and extend the kit with a mezzanine board.

Dragonwing EVKs are designed to withstand the real-world conditions you throw at them. Test your fleet management applications on the controlled power consumption and heat dissipation of the EVKs.

How do I get a Linux box with strong performance that has an MCU and can run LLMs and VLMs? VENTUNO Q can help.

Robotics development takes you into physical AI. The transition from isolated machines to coordinated physical AI systems is another phase of robotics, anchored by the VENTUNO Q development board on Dragonwing IQ8.

VENTUNO Q is made for:

• Robotics developers
• Physical AI designers combining AI decision making with deterministic control
• Teams that need stronger connectivity, expansion and system integration than a starter or Raspberry Pi-style board can provide

Physical AI picks up where edge AI leaves off. It’s the artificial intelligence at work in advanced robotics, vehicles and drones: devices that move, manipulate and respond to the physical world. VENTUNO Q brings perception, decision and action onto a single board, eliminating the complexity, latency and cost of configuring multiple devices to perform the same work.

The board is equipped with an NPU that delivers up to 40 dense TOPS for advanced computer vision and inference on local LLMs and VLMs. It comes with 16 GB of RAM for loading larger models, processing high-resolution imaging and running intensive robotics algorithms smoothly. Its 64 GB of industrial-grade eMMC storage accommodates frameworks, operating system, models and data.

The board combines two processors in a dual-brain architecture that is accelerated for the perception, decision-making and real-time actuation demanded by physical AI. It combines a Dragonwing IQ8—the AI brain—and a microcontroller—the action brain. They communicate through remote procedure calls (RPCs) and divide advanced AI workloads. The AI brain delivers heterogeneous computing on NPU, CPU and GPU for AI inference. The action brain, a STMicroelectronics STM32H5F5 microcontroller, enables sub-millisecond response to guarantee stable, deterministic control for robotics, motion systems and industrial interfaces.

To keep up with the potential applications of physical AI, VENTUNO Q is equipped with a wide range of interfaces besides USB 3.0 and HDMI. It offers wireless networking with tri-band Wi-Fi 6 (2.4/5/6 GHz), Bluetooth 5.3 and wired networking with 2.5 Gb Ethernet. The board supplements industrial-grade eMMC with an M.2 connector for adding NVMe Gen 4 storage. Its expansion options include native support for Raspberry Pi Hat, UNO shields and carriers, and Arduino Modulino™ nodes.

With support for Ubuntu, Debian and Arduino App Lab, VENTUNO Q is the right platform for developers who are tired of choosing between AI and control. It lets them build systems that can do both well.

How do I get started with Qualcomm IoT SoCs and combine AI workloads with microcontroller units? UNO Q can help.

The UNO Q is the best entry point for beginner robotics and hands-on experimentation with embedded systems. It’s ideal for:

• Students and educators who want a practical entry point into robotics, sensors and embedded systems
• Beginning robotics projects where the goal is to understand cause and effect among sensing, local compute and physical behavior
• Teams that want an industry-aligned board without the need to jump immediately into a more complex Linux AI platform
• Beginning developers who want a small, low-cost, direct application board that bridges the world of MCUs and microprocessor units (MPUs)

UNO Q is a good, start-small board because it supports lightweight AI, Linux-capable workflows and real-time control in a form that keeps the learning curve manageable. More than a simple demo board, UNO Q is a robust, industry-standard entry point. UNO Q delivers enough compute to make robotics and AI tangible. The platform is designed for sensor data collection, real-time inference at the edge and local data processing for practical learning experiences.

Like VENTUNO Q, UNO Q is built on dual-brain architecture. The MPU is a Dragonwing QRB2210, the intelligent core with the kind of environment you would find on a higher-end single-board computer. The MCU is a STMicroelectronics STM32U585, an Arduino-standard microcontroller for real-time control and precise hardware interaction. From input like camera streams, API triggers, sensors and buttons, UNO Q can output AI inference, live video, I/O switching and signals to DC and stepper motors.

The board is well suited to applications that connect to the physical world; for example, by combining sensors, displays and robot add-ons into a weather station.

The board runs Debian or Qualcomm Linux.

Start with UNO Q and pre-built modules in Arduino App Lab called “Bricks”, if you want to build confidence with beginner robotics, embedded I/O and local AI workflows. Then, wade deeper into Linux applications for vision computing.

At all levels, a developer-friendly software stack

The Dragonwing IQ-8275 and Dragonwing IQ-9075 run both Ubuntu and Qualcomm Linux and come with a full software stack for developer-friendly multi-OS support.

The stack is a comprehensive package of software, tools, and documentation designed for Dragonwing SoCs. With its unified Linux distribution, you can build once, scale, and deploy across all supported IoT SoCs, including Dragonwing IQ-8275 and Dragonwing IQ-9075. The distribution is built on Yocto Project Long-Term Support (LTS), and it incorporates standard boot infrastructure, the latest LTS kernel, and drivers aligned with upstream Linux development.

This upstream-first, publicly developed software stack supports all processors, subsystems, and components within the platform, including the CPU, GPU, NPU, video processing unit (VPU), display processing unit (DPU), and PMICs. Qualcomm Linux is designed for production-grade deployment. It supports virtualization, over-the-air (OTA) updates, and long-term lifecycles in Dragonwing IoT SoCs.

On the UNO Q and VENTUNO Q, you have access to Arduino App Lab for quick app development on Ubuntu.

Navigating the ecosystem: Developer FAQ

1. What is the difference between Dragonwing IQ8 (IQ-8275) and Dragonwing IQ9 (IQ-9075)?

The main differences are in the level of performance and in the number of connectors. For example, Dragonwing IQ9 delivers up to 100 TOPS of AI inference, handles up to 16 cameras at 12 megapixels, and accommodates up to 12 displays. Because of its suitability to demanding environments, it is also available as the pre‑integrated, industrial‑grade Dragonwing IQ‑9075 Module. Dragonwing IQ8 delivers up to 40 TOPS of AI inference, handles up to 12 cameras at 8 megapixels, and accommodates up to 5 displays.

2. What is the out-of-the-box experience (OOBE) for the Dragonwing IQ-8275 Evaluation Kit and the Dragonwing IQ-9075 Evaluation Kit? How easy is it to go from prototype to product?

The EVKs come pre-loaded with Qualcomm Linux, so you’ll have a working, single-board computer as soon as you plug it in and turn it on. We’ve made it easy for you to switch to Ubuntu if you prefer. For the work of building prototypes, you’ll find design files, schematics, layout files, and mechanical 3D models for each kit. You can use the EVK as is, attaching it to your project, plugging in accessories such as cameras or low-speed peripherals and getting straight to work on prototypes. As for hardware, you can start by either creating your own mezzanine boards or using the EVK as a reference for new devices.

3. What is the difference between the evaluation kits and a typical single-board computer?

Both the Dragonwing IQ-8275 Evaluation Kit and the Dragonwing IQ-9075 Evaluation Kit are built with a wide variety of expansion interfaces including connectors for Ethernet, USB, cameras, and CEM PCIe for attaching any off-the-shelf PCIe card. They have low-speed headers and high-speed board-to-board connectors for additional daughter or mezzanine boards. You can install expansion mezzanines for even more Ethernet, USB, display and PCIe connections. Besides connectors, the kits have speaker headers, an onboard microphone, a real-time clock, onboard temperature and IMU sensors, and customizable features like a software programmable push button and a red/green/blue LED.

4. How do I know whether my AI workload needs 100 tera operations per second (TOPS), or whether 12 TOPS is enough?

If you're running single-model inference (like barcode scanning or wake word detection), 8 to 12 TOPS is likely sufficient. If you're managing real-time vision, sensor fusion or multiple concurrent models (e.g., in robotics), you’ll want something like the Dragonwing IQ‑9075 EVK, with up to 100 TOPS.

5. What are the biggest bottlenecks or developer problems that UNO Q and VENTUNO Q solve?

The UNO Q solves the common design problem of locating both an SoC running Linux and an MCU performing low-level processing right on the same board. It bridges a quad-core Qualcomm processor performing AI inference at the edge to a real-time STM32 co-processor that surpasses the capabilities of traditional 32-bit microcontrollers.

From the Arduino App Lab you can use Bricks to stack Arduino sketches, Python scripts, and pre-trained AI models into a single cohesive application. The result is that you can do more for less: you get a Linux environment for running Python-based AI tasks while simultaneously maintaining the millisecond-accurate hardware control essential for robotics and local automation. VENTUNO Q adds higher performance for developing and running autonomous agents that can perceive the physical world, execute tasks using local large language models (LLMs), and take independent actions.

Combined with Arduino App Lab, the board also helps developers to focus on developing agentic AI instead of spending time getting the processor and microcontroller to communicate with each other.

6. How do UNO Q and VENTUNO Q fit into the broader ecosystem? In what ways—both hardware and software—do they reduce friction for developers?

In hardware and connectivity, UNO Q takes advantage of Modulino via Qwiic connectors for no-solder testing and development, including edge AI motion, audio and vision. In software, you can use your Arduino Cloud credentials to log into the Edge Impulse developer portal, fine-tune models in your browser, then transfer them to your UNO Q using Arduino App Lab.

VENTUNO Q gives you access to industrial-grade hardware interfaces, such as triple MIPI-CSI and CAN bus. It greatly extends your AI software development options to Qualcomm® AI Hub and Hugging Face.

7. If I build a prototype on UNO Q, can I scale it to production later?

Yes. That’s the value of the Qualcomm IoT ecosystem. You can start with a board like UNO Q and move your project to hardware like VENTUNO Q or Dragonwing IQ‑9075 EVK. Thanks to consistent BSPs, SDKs and AI toolchain support, you can count on minimal changes to your code or model.

Conclusion

• Opt for Dragonwing IQ9 when high performance, I/O and deployment scale become the priority.
• Choose Dragonwing IQ8 and its upstream Linux support for commercial products.
• Select the Dragonwing IQ-8275 Evaluation Kit or Dragonwing IQ-9075 Evaluation Kit when you’re ready to go to production.
• Step into VENTUNO Q when your system needs physical AI.
• Start on UNO Q to learn the fundamentals.

We’d be glad to see what you build with these boards! Join us on Qualcomm Developer Discord to show off your project