Radxa CM3 Industrial Compute Module (Embedded SoM)

A system-on-module aimed at embedded AI

The Radxa CM3 is a compute module rather than a standalone board, built for industrial applications that need real processing power and AI capability in a small package. At its core is the Rockchip RK3566, a quad-core Arm Cortex-A55 processor running at up to 1.8GHz, paired with an Arm Mali-G52 2EE GPU that supports OpenGL ES and Vulkan 1.1 for graphics-heavy interfaces.

What sets it apart is the onboard 1 TOPS neural processing unit, which supports common deep learning frameworks including TensorFlow, PyTorch and ONNX. That means models can run inference directly on the module, which is useful for embedded vision, audio classification or other on-device AI tasks without shipping data to a separate server.

Video, connectivity and physical design

Video capability includes a 4K@60fps H.265/H.264/VP9 hardware decoder and HDMI output supporting up to 4K resolution, so the module can drive a display directly for kiosks, digital signage or embedded HMI panels. Networking covers a Gigabit Ethernet PHY for wired connections plus 802.11ac WiFi and Bluetooth 5.0 for wireless setups, and the module exposes HDMI, MIPI-DSI/CSI, PCIe, SATA, USB and SDIO interfaces for whatever carrier board it sits on.

The module itself measures 55mm x 40mm and is rated for operation between 0 and 50C, and supports Debian, Linux and Android, giving flexibility in the software stack chosen for a given deployment.

Who it is for

Engineers designing industrial gateways, embedded vision systems or edge-AI appliances that need a compact module with real inference capability will get the most value from the CM3. As a system-on-module, it needs a compatible carrier board to break out its interfaces for a finished product.

Questions buyers ask

Does it need a carrier board?

Yes, the CM3 is a compute module and needs a carrier board to expose its interfaces for use.

What AI frameworks does the NPU support?

TensorFlow, PyTorch and ONNX, according to Radxa's listing.

What is the NPU's performance?

1 TOPS of AI inference performance.