Qualcomm is expanding its Dragonwing lineup with two new processors aimed at IoT devices and industrial systems where cost, power consumption, and size drive the design. The Dragonwing Q-2390 and IQ-2390 combine a CPU, GPU, AI acceleration, image processing, and a RISC-V microcontroller to run real-time tasks without constantly relying on the cloud.
The new Qualcomm Dragonwing chips in 20 seconds
- The Q-2390 and IQ-2390 combine a quad-core CPU, an Adreno 704 GPU, AI, and a RISC-V microcontroller.
- Both reach 1.1 TOPS of AI performance.
- The Q-2390 targets retail, smart homes, robots, and enterprise terminals.
- The IQ-2390 adds features for automation, machine vision, and industrial control.
- Qualcomm expects evaluation kits to be available in early 2027.
The new chips were unveiled ahead of IFA 2026 and are already available to select customers through an early-access program. Modules based on both processors will be on display during the show, though Qualcomm is targeting early 2027 for evaluation-kit availability.
The move is especially interesting because Qualcomm isn’t trying to compete here with its higher-performance AI platforms. The new IQ2 line sits right at the bottom of its industrial catalog: 1.1 TOPS versus 40 TOPS for the IQ6 and IQ8 families, 100 TOPS for IQ9, or 350 TOPS for IQ10.
The goal here is different: bring local processing and AI to equipment where a much more powerful platform would also be more expensive, more complex, and harder to cool.
Arm CPU, Adreno GPU, Hexagon NPU, and a RISC-V microcontroller
The Dragonwing Q-2390 and IQ-2390 share much of their architecture.
The CPU uses four Arm cores — specifically one Cortex-A78 paired with three Cortex-A55 cores, running at up to 1.9 GHz. It’s joined by a Qualcomm Adreno 704 GPU at 1.1 GHz and a Hexagon V661 DSP.
AI acceleration is handled by a Qualcomm Hexagon NPU rated at 1.1 TOPS. That figure clearly places these processors in different territory than AI PC chips or the large accelerators used in data centers.
| Spec | Dragonwing Q-2390 | Dragonwing IQ-2390 |
|---|---|---|
| CPU | 1× Cortex-A78 + 3× Cortex-A55 | 1× Cortex-A78 + 3× Cortex-A55 |
| CPU clock | Up to 1.9 GHz | Up to 1.9 GHz |
| GPU | Adreno 704 | Adreno 704 |
| NPU | Hexagon, 1.1 TOPS | Hexagon, 1.1 TOPS |
| DSP | Hexagon V661 | Hexagon V661 |
| Memory | LPDDR4X, up to 8 GB | LPDDR4X |
| Real time | RISC-V MCU | RISC-V MCU |
| Ethernet | Dual Gigabit with TSN | Dual Gigabit with TSN |
| Operating systems | Android, Linux, and Zephyr | Android, Linux, and Zephyr |
| Focus | Commercial and consumer IoT | Industrial IoT |
One of the less common elements here is precisely that built-in RISC-V microcontroller.
Its job isn’t to replace the Arm cores running the main operating system. Qualcomm added it for deterministic, real-time tasks, letting certain control processes be kept separate from the applications running on Linux or Android.
That’s especially relevant on the IQ-2390.
An industrial system may need to run a graphical interface, process images, and run AI inference while also keeping certain control operations running with predictable response times. Integrating these resources into the same system-on-chip can save manufacturers from having to add certain external components.
Qualcomm also includes dual Gigabit Ethernet with Time-Sensitive Networking (TSN), a technology designed to provide deterministic timing characteristics over Ethernet networks. In industrial settings, it can be used for communications where it’s not enough for a packet to arrive quickly — it also has to arrive within known time margins.
Q-2390: AI for terminals, retail, appliances, and robots
Even though they share an architecture, Qualcomm clearly separates the markets for the two processors.
The Dragonwing Q-2390 belongs to the Q2 family and is designed for commercial, enterprise, and consumer devices.
The examples Qualcomm gives include point-of-sale terminals, kiosks, access control systems, smart appliances, connected agriculture, home robots, fitness equipment, and enterprise terminals.
The platform integrates application processing, graphics, vision, local inference, and various input and output interfaces.
It also offers Wi-Fi and Bluetooth through compatible components, dual Ethernet, and a Q-2390M variant that adds built-in LTE Cat 4 connectivity. Qualcomm is thus targeting devices that can operate outside a conventional Wi-Fi or Ethernet network.
Local AI can be used, for example, to analyze camera data without constantly sending it to a data center.
That can reduce latency and network traffic, as well as enable certain functions when a connection to external services isn’t available.
Even so, the available 1.1 TOPS make clear what kind of AI this product is aimed at. It’s not a platform designed to run locally the large generative models that currently draw most of the market’s attention.
Its natural territory is much smaller models for classification, detection, recognition, sensor analysis, and computer vision.
IQ-2390 brings the same architecture to factories and energy systems
The Dragonwing IQ-2390 opens the new IQ2 family and toughens the platform for industrial scenarios.
Qualcomm specifies an operating range of -30°C to +115°C and adds ECC error-correcting memory, industrial-grade packaging, and features intended for installations exposed to vibration and shock.
The company envisions its use in human-machine interfaces (HMI), programmable logic controllers (PLC), CNC machines, industrial gateways, machine-vision systems, building automation, HVAC, EV chargers, and energy storage and management systems.
Here, integration can carry considerable weight.
A manufacturer could combine the device interface, camera processing, AI inference, deterministic Ethernet communications, and certain real-time control tasks into a single SoC.
Qualcomm is also trying to address a concern that matters a great deal in the industrial market: hardware lifespan.
The IQ-2390’s official page states product support for more than ten years, and the Dragonwing IQ2 family currently sits within the longevity program through 2036, subject to possible changes.
A cycle that long matters less for a consumer device likely to be replaced within a few years, but it can be decisive for industrial machinery designed to stay in operation for a decade or more.
The software also reflects that dual focus.
Qualcomm supports Android, Ubuntu, Yocto-based Linux, and Zephyr, allowing full operating systems to be combined with a lightweight real-time operating system (RTOS) for embedded devices.
The company already has several partners working around the new platforms. Fibocom is developing the SC236 module based on the Q-2390, while SECO is preparing Compact Vision 5 and an SBC board with the IQ-2390. Engicam is likewise working on a system-on-module (SOM) built around the industrial processor.
What’s interesting about these Dragonwing chips lies precisely in that less visible side of AI’s expansion. Large models need massive data centers, but a growing share of simple decisions can run directly where the data is generated.
An industrial camera doesn’t necessarily need to send every image to the cloud to detect an anomaly. An access-control system can handle certain tasks locally. An energy controller can combine sensors, communications, and small models without depending on a remote server for every decision.
Qualcomm wants to bring that capability to equipment where price and power consumption have so far made it hard to add dedicated AI hardware.
The Q-2390 and IQ-2390 are still in early access, so it will take the first commercial products to see how these possibilities translate into real devices.
Frequently asked questions
How much AI performance do the Qualcomm Q-2390 and IQ-2390 have?
Both include a Qualcomm Hexagon NPU rated at 1.1 TOPS. They’re aimed at lightweight edge inference, not large generative models.
What’s the difference between the Q-2390 and the IQ-2390?
They share most of the same architecture, but the Q-2390 is aimed primarily at commercial and consumer IoT. The IQ-2390 adds features specific to industrial environments, including an operating temperature range of -30°C to +115°C and ECC memory protection.
Why do they include a RISC-V microcontroller?
The RISC-V MCU is intended for deterministic, real-time tasks. It complements the main Arm CPU and allows certain control processes to be kept separate from the workloads running on the operating system.
When will they be available?
Qualcomm currently has an early-access program running for the Q-2390 and IQ-2390. Evaluation kits are expected in early 2027.
Source: Qualcomm
