M5Stack LLM‑8850 card is an M.2 M-Key 2242 AI acceleration module powered by an Axera AX8850 SoC delivering 24 TOPS ( INT8) of performance, and suitable for host devices such as Raspberry Pi 5, Rockchip RK3588 SBCs, and even x86 PCs like mini PCs with a spare M.2 Key-M socket.
The card ships with 8GB RAM, a 32Mbit SPI NOR flash, and also supports H.265/H.264 8Kp30 video encoding and 8Kp60 video decoding, with up to 16 channels for 1080p videos. It is also equipped with an active cooling system to maintain stable temperatures and prevent thermal degradation inside enclosures.
M5Stack LLM‑8850 card specifications:
- SoC – Axera AX8850
- CPU – Octa-core Cortex‑A55 processor at 1.7 GHz
- NPU – 24 TOPS @ INT8
- VPU
- Video Encoder – 8K @ 30 fps H.264/H.265 encoding, supports scaling / cropping
- Video Decoder – 8K @ 60 fps H.264/H.265 decoding, supports 16 channels 1080p parallel decoding, supports scaling / cropping
- Memory – 8GB 64‑bit LPDDR4x @ 4266 Mbps
- Storage – 32Mbit QSPI NOR Flash (for Bootloader only)
- Host Interface – PCIe 2.0 x2 via M.2 Key-M edge connector
- Cooling – Micro turbo fan + integrated aluminum alloy CNC heatsink
- Power Supply – 3.3V via edge connector
- Power Consumption – Up to 7 Watts
- Dimensions – 42.6 x 24.0 x 9.7mm
- Weight – 14.7 grams
- Temperature Range
- Operating – 0 to 60°C
- Full load temperature at room temperature – 70°C
The card works with Ubuntu 20.02, 22.04, and 24.04 as well as Debian 12, but software support is not available for other operating systems, such as Windows, macOS, or even Windows WSL. That’s because it relies on the axcl-smi driver only available in Linux.
Once the driver is installed on your Raspberry Pi 5 or another Linux SBC or mini PC, you can download various demos listed in the wiki to get started. A fairly long list of models is provided:
- Vision – YOLO11, Yolo-World-V2, Yolov7-face, Depth-Anything-V2, MixFormer-V2, Real-ESRGAN, SuperResolution, RIFE
- Large Language – Qwen3-0.6B, Qwen3-1.7B, Qwen2.5-0.5B-Instruct, Qwen2.5-1.5B-Instruct, DeepSeek-R1-Distill-Qwen-1.5B, MiniCPM4-0.5B
- Multimodal – InternVL3-1B, Qwen2.5-VL-3B-Instruct, SmolVLM2-500M-Video-Instruct, LibCLIP
- Audio – Whisper, MeloTTS, SenseVoice, CosyVoice2, 3D-Speaker-MT
- Generative – lcm-lora-sdv1-5, SD1.5-LLM8850, LivePortrait
M5Stack did not provide benchmark results or try to compare the Axera AX8850 SoC against other AI accelerators, although the wiki reveals some numbers, such as 12.88 tokens/s for Qwen3-0.6B with w8a16 quantization. Based on the arguably flawed TOPS number, it looks like a potential competitor to the 26 TOPS Hailo-8 AI accelerator. However, the latter is optimized for computer vision applications, and for tasks like LLMs, the Axera 8850 will be vastly superior.
The price for the LLM-8850 card is within the same range, and even a bit cheaper than a product such as the Raspberry Pi AI HAT+ 26 TOPS ($110), and much more affordable than the ~$200 Hailo-8 M.2 card. M5Stack sells the Axera AX8850 M.2 module for $99 on AliExpress and its own online store.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress. We also use affiliate links in articles to earn commissions if you make a purchase after clicking on those links.
