Radxa Dragon Q8B and Q5E: Qualcomm SBCs Built to Punch Above Their Price

Alright, gather round the workbench, because the Radxa Dragon lineup just got the kind of spec sheet that makes me start sketching builds on napkins. At the Radxa + Qualcomm developer day on May 30, 2026, Radxa expanded its Qualcomm-powered Dragon family of single-board computers, and the headliners — the Dragon Q8B and Dragon Q5E — are exactly the sort of price-to-performance hardware I love benchmarking in my head before the review units even ship.

Let me break down the numbers, because numbers are the whole point.

Dragon Q8B: a desktop-class Qualcomm SBC in a 100x75mm footprint

The Dragon Q8B is the flagship, and it is genuinely a lot of board. It runs the Snapdragon 8cx Gen 3 — an octa-core SoC paired with Adreno 690 graphics — and scales to a frankly generous 32GB of RAM. For an Arm SBC, that headroom changes what you can realistically run: think containerized homelab stacks, multiple VMs, or a self-hosted dev box that doesn't tap out the moment you open a few browser tabs and a build pipeline.

The I/O is where my enthusiast brain lit up. You get dual DisplayPort 1.4b output piped over USB-C, so a clean two-monitor desktop is on the table without a dongle drawer. Networking is dual 2.5GbE — two fast pipes, which is exactly what you want for a router, firewall, or a board that lives between your switch and your storage. And the storage story is the real treat: dual M.2 NVMe sockets on a 100x75mm board. Two NVMe slots means you can run a fast boot drive plus a roomy data drive, or mirror them for a tidy little redundant setup.

What you can build with it

This is the kind of board I'd reach for to build a compact 2.5GbE-connected NAS front-end, a low-power virtualization host, or a silent desktop replacement. The combination of 32GB RAM, dual NVMe, and dual DisplayPort over USB-C in this footprint is a sweet spot that didn't really exist at SBC scale before.

Dragon Q5E: an efficient AI-capable SBC

If the Q8B is the muscle, the Dragon Q5E is the clever, efficient sibling. It's built around the Dragonwing QCS6690 SoC with 6 TOPS of on-device AI, up to 16GB of LPDDR5, and — like its bigger sibling — dual 2.5GbE. It also adds a 32MP camera interface, which immediately reframes it as an edge-AI and vision platform.

6 TOPS isn't going to train large models, and I want to be clear that's analysis on my part, not a claim from Radxa — but it's a sensible budget for real inference workloads: object detection on a camera feed, smart-home automation, or a lightweight vision pipeline running locally without shipping frames to the cloud. Pair the camera interface with that NPU and dual fast NICs, and you've got a tidy edge node.

The roadmap: DragonStation, DragonBay, and 22 systems for 2026

Radxa didn't stop at two boards. The team teased a DragonStation with six M.2 NVMe bays and 10GbE — that is a serious all-flash storage proposition — plus a DragonBay NAS. Both are previews rather than shipping products, so file them under "exciting, but wait for hands-on numbers."

The bigger signal is the roadmap: 22 Qualcomm-based systems planned across SBCs, NAS, and robotics for 2026. That's a real platform commitment, and platform commitment is what turns a fun board into an ecosystem you can build long-term projects around.

Pricing and the bottom line

Pricing wasn't fully nailed down, but it was described as pleasantly surprising — and for a Qualcomm SBC with dual NVMe, dual 2.5GbE, and up to 32GB RAM, "pleasantly surprising" is the magic phrase for price-to-performance hunters like me. I'll reserve hard judgment for confirmed MSRPs and real benchmarks, but on paper the Dragon Q8B and Q5E look like two of the most build-worthy single-board computers of the year. I cannot wait to plug one in.

Sources: CNX Software, June 1 2026; bret.dk, 2026; Radxa Dragon product page