M5Stack LLM630 Compute Kit (AX630C)

The LLM630 Compute Kit is an AI inference platform featuring an AX630C SoC (3.2 TOPs NPU), 4GB LPDDR4, 32GB eMMC, Wi-Fi 6, Gigabit Ethernet, dual USB-C ports, and an SMA antenna interface. It supports CV, speech, and LLM tasks with the StackFlow framework.

M5Stack LLM630 Compute Kit is an AI large language model inference development platform designed for edge computing and smart interactive applications. This kit’s mainboard is equipped with an Aixin AX630C SoC processor, integrating a high-efficiency NPU with 3.2 TOPs at INT8 computing power, providing powerful AI inference capabilities to efficiently execute complex vision (CV) and large language model (LLM) tasks, meeting the needs of various intelligent application scenarios.

The mainboard is also equipped with the JL2101-N040C Gigabit Ethernet chip and the ESP32-C6 wireless communication chip, supporting Wi-Fi 6 at 2.4 G as the device’s network card, providing high-speed data transmission for bridging functionality between Wi-Fi and Ethernet. Whether through a wired connection for large-scale data exchange or via wireless communication for real-time interaction with a remote server or other smart devices, this platform ensures efficient data exchange. The mainboard also integrates an SMA antenna interface to further enhance the stability and transmission range of wireless signals, ensuring stable communication in complex network environments. It comes with 4GB LPDDR4 memory (of which 2GB is for user usage, 2GB for hardware acceleration) and 32GB eMMC storage, supporting parallel loading and inference of multiple models for efficient and smooth task handling.

Working in perfect conjunction with the baseboard, the LLM630 Compute Kit’s functionalities and applicability are significantly expanded. It integrates a BMI270 6-axis sensor for precise attitude sensing and motion detection, suitable for various dynamic applications. With the built-in NS4150B Class D amplifier and microphone and speaker interfaces, it supports high-quality voice input and audio output, enabling full-duplex communication for an enhanced user interaction experience. The baseboard also features dual Grove interfaces as well as LCD/DSI and CAM/CSI MIPI interfaces, making it convenient to expand peripherals such as displays and camera modules. Additionally, the baseboard integrates an external antenna interface and a Gigabit Ethernet port, providing flexible network connections and enhanced wireless performance. The device’s user buttons enable functions such as power on/off and mode switching, enhancing usability and interactivity.

The baseboard’s charging chip and reserved battery socket support custom battery configurations, ensuring the platform can run stably for long periods even without external power. The integrated battery detection chip monitors battery status in real time. The MicroSD card slot allows for storage expansion and will later support AI model updates. The dual Type-C ports not only support high-speed data transmission but also provide OTG functionality, making device connections more flexible and ensuring high efficiency in data exchange and device connectivity.

LLM630 Compute Kit supports the StackFlow framework, allowing developers to realize edge AI applications with just a few lines of code and quickly start various AI tasks. The platform supports a wide range of AI applications, including computer vision recognition, speech recognition, text-to-speech, and wake word recognition, and supports separate invocation or automated pipeline flow for convenience. The platform also supports models such as Yolo11 DepthAnything for vision tasks, InternVL2.5-1B multi-modal large model, Qwen2.5-0.5/1.5B Llama3.2-1B large language models, and Whisper Melotts voice models, with hot update features. It will continue to support cutting-edge large models, empowering intelligent recognition and analysis, ensuring the platform keeps pace with technological developments and community trends.

LLM630 Compute Kit is suitable for security monitoring, smart retail, smart agriculture, smart home control, interactive robotics, and education, providing powerful computing capabilities and flexible expandability for edge AI applications.