Radxa Launches M.2 AI Accelerator with Axera AX8850 and 24 TOPS NPU
The Radxa AICore AX-M1 is an M.2 M Key AI acceleration module designed for edge computing systems that require high-throughput neural processing. Built around the Axera AX8850 system-on-chip, the module combines an octa-core Cortex-A55 processor with a 24 TOPS INT8-capable NPU and an 8K-capable video processing unit, delivering AI processing capabilities in a compact footprint.
According to Radxa, the onboard VPU supports 8K H.264/H.265 encoding and decoding, as well as 16-channel 1080p decoding, for use in computer vision and multi-stream video analytics. With 8GB of LPDDR4X RAM, the module can process memory-intensive AI models without relying on external memory.
Measuring 22 mm × 80 mm, the AX-M1 is compliant with the M.2 M Key form factor and is compatible with a wide range of host systems, including x86-based platforms from Intel and AMD as well as Rockchip-based ARM systems. It is supported on Linux distributions such as Ubuntu, Debian, and CentOS, and has a typical power draw of less than 8 W.
Radxa officially lists the ROCK 5A, 5B, 5B+, and ROCK 5 ITX boards as tested motherboards. For installation, the AX-M1 requires an available M.2 M Key slot, and setup involves both hardware mounting and driver installation. The AXCL runtime driver enables system recognition and control, allowing users to validate the module’s operational status using the axcl-smi utility.
