Specifications
SoC
Qualcomm QCS6490
CPU
1x Kryo Gold Plus @ 2.7 GHz + 3x Kryo Gold @ 2.4 GHz + 4x Kryo Silver @ 1.9 GHz (8 cores)
GPU
Qualcomm Adreno 643 — OpenGL ES 3.2, Vulkan 1.3, OpenCL 2.2 and DirectX 12
AI Accelerator
Hexagon DSP plus Hexagon Tensor Accelerator, up to 12 TOPS combined
Video Processing
Up to 4K@60fps decode (H.264/H.265/VP9) and 4K@30fps encode (H.264/H.265)
Memory Options
4 GB / 6 GB / 8 GB / 12 GB / 16 GB LPDDR5
Boot Storage
32 MB QSPI NOR flash
Expandable Storage
microSD, eMMC module, UFS module or M.2 M-Key 2230 NVMe SSD
Networking
Gigabit Ethernet with PoE support via an external PoE HAT, Wi-Fi 6 and Bluetooth 5.4
USB
1x USB 3.1 Type-A (host/OTG) plus 3x USB 2.0 Type-A
Display
1x HDMI up to 4K@60Hz and 1x 4-lane MIPI DSI
Camera Inputs
1x 4-lane MIPI CSI plus 2x 2-lane MIPI CSI — three camera connectors
Expansion
40-pin GPIO header (UART, SPI, I2C) and M.2 M-Key 2230
Power Input
USB Type-C 12 V, external 12 V pin header, or PoE with a HAT
Dimensions
85 mm x 56 mm credit-card form factor
Operating Systems
Ubuntu, Qualcomm Linux (Yocto-based), Android 15/16 and Windows on ARM
Measured Power Draw
About 8 W typical, peaking near 20 W under simultaneous GPU and NPU load
List Pricing
From about $59.50 for the 4 GB model; higher memory configurations list roughly $72 - $124
Overview
The Radxa Dragon Q6A is a Raspberry Pi-class single board computer built around the Qualcomm QCS6490. Its defining feature for edge AI work is the Hexagon DSP and Hexagon Tensor Accelerator delivering up to 12 TOPS of on-device compute — enough to run quantised Llama 3.2 and similar small models locally without a discrete accelerator.
The I/O mix is unusually generous for the form factor. Three MIPI CSI camera connectors (one 4-lane, two 2-lane) make it a practical vision board, and storage is flexible: microSD, eMMC module, UFS module or an M.2 M-Key 2230 NVMe SSD. Gigabit Ethernet with PoE support through an external HAT, Wi-Fi 6, Bluetooth 5.4, USB 3.1, HDMI 4K@60 and a 40-pin GPIO header cover the rest.
Power behaviour is well suited to embedded deployment: roughly 8 W typical draw, peaking near 20 W when GPU and NPU run together. The board supports Ubuntu, Qualcomm Linux, Android and Windows on ARM, and pricing starts around $59.50 for 4 GB of LPDDR5, rising to roughly $124 for the 16 GB configuration. QS Compute supplies ARM edge AI boards, industrial single board computers and embedded AI modules — contact us for volume pricing.
Key Benefits
12 TOPS on-device: Hexagon NPU acceleration at a single-board price point. Three camera ports: tri-MIPI CSI for stereo and multi-view vision. Flexible storage: eMMC, UFS or M.2 2230 NVMe plus microSD. Low power: about 8 W typical draw, PoE-capable.
Applications
Edge computer vision and camera analytics, robotics and AMR perception, smart retail and access control, industrial monitoring gateways, on-device LLM experimentation, and cost-sensitive AI prototyping.
Request a Quote — RADXA DRAGON Q6A — QUALCOMM QCS6490 EDGE AI SINGLE BOARD COMPUTER
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →