M5Stack LLM Module Kit (AX630C) w/ 4GB RAM & 32GB eMMC
The M5Stack LLM Module Kit is an integrated offline AI inference platform powered by the AiXin AX630C SoC, delivering 3.2 TOPS for on-device large language model (LLM), speech recognition, and text-to-speech tasks. It runs on Ubuntu and is programmable via Arduino IDE or UiFlow for edge AI applications.
Features
- 3.2 TOPS NPU for Offline AI – AX630C SoC with dedicated NPU handles INT8-precision LLM inference, speech recognition (ASR), and text-to-speech (TTS) without cloud dependency.
- High-Capacity Memory & Storage – 4GB LPDDR4 RAM and 32GB eMMC support parallel multi-model loading and chained inference for complex AI workflows.
- Integrated Audio I/O – Onboard MEMS microphone and 1W speaker enable immediate voice-based interaction and audio feedback for assistant applications.
- Versatile Development Frameworks – Compatible with Arduino IDE, UiFlow, and the built-in Ubuntu system, supporting OpenAI-standard API plugins for rapid AI integration.
- Expandable Interface Module – Includes LLM Mate module with RJ45 Ethernet (100Mbps), USB-to-serial (CH340N), Type-C, and M5BUS for system integration and power stacking.
Out of stock
The M5Stack LLM Module Kit is a dedicated edge AI computing platform designed for terminal devices requiring efficient, private, and offline intelligent interaction. At its core is the AiXin AX630C system-on-chip, featuring a dual-core Cortex-A53 CPU and a high-efficiency neural processing unit (NPU) capable of 3.2 TOPS at INT8 precision. This hardware is natively optimized for Transformer models, enabling smooth execution of pre-installed large language models like Qwen2.5-0.5B and supporting parallel multi-model processing.
The kit comprises two stackable modules: the primary LLM module housing the AX630C SoC, 4GB LPDDR4 RAM, 32GB eMMC storage, microphone, speaker, and RGB LED; and the LLM Mate interface module, which adds 100Mbps Ethernet via RJ45, USB-to-serial debugging, and expansion pads. Communication between modules and hosts is handled via stable serial links over FPC connectors. With an operating power consumption of approximately 1.5W under full load, the system is engineered for energy-efficient, long-term deployment in smart home, robotic, and industrial control scenarios.
Additional Information
| Weight | 0.051 kg |
|---|---|
| Dimensions | 5.4 × 5.4 cm |
