Cloud-based AI models are quite convenient for typical users. However, some use cases may require to run these models locally in order to ensure data privacy or to work around a slow or non-existing internet connection. That said, a certain amount of performance is needed to run large language models or even AI-assisted video analysis. The latter might be useful for counting people who pass through the entrance of a building, for example.
With the M5Stack LLM-8850, another expansion card that accelerates these exact AI applications has now hit the market. It’s powered by an Axera AX8550 SoC with four Cortex-A55 cores and an NPU with a performance of 24 TOPs. By comparison, NPUs integrated into modern AMD and Intel CPUs offer roughly twice as much performance. A VPU is onboard as well and can simultaneously decode up to 16 video streams at Full HD resolution. The board further sports 8GB of RAM and comes with a heatsink and a fan.
The card is connected via an M.2 Key M slot and uses two PCIe 2.0 lanes. The board measures 1.68 x 0.94 x 0.38 inches and is compatible with the Raspberry Pi 5 and other single-board computers or even mini PCs as well as various AI frameworks. Last but not least, the M5Stack LLM-8850 is officially priced at $99, but additional shipping and import fees may apply.
Source(s)
M5Stack, via CNX-Software











