MX3-2280-M-4 AI Accelerator Module

The MX3-2280-M-4 is a state-of-the-art M.2 AI Accelerator Module crafted by MemryX, designed to redefine edge computing with unparalleled artificial intelligence (AI) inference performance. Encased in the compact M.2 2280 M-Key form factor (22mm x 80mm), this module leverages a PCIe Gen 3 interface to deliver seamless integration and exceptional computational efficiency. Powered by four MemryX MX3 Edge AI Accelerator chips, it achieves up to 24 TFLOPS of AI compute power while maintaining an ultra-low power envelope of 6–8 watts.

Engineered for real-time, low-latency processing of deep neural network (DNN) models, the MX3-2280-M-4 excels in computer vision (CV) applications, making it ideal for robotics, smart surveillance, industrial automation, and IoT deployments. Its innovative dataflow architecture offloads complex AI workloads from host CPUs, ensuring high-throughput inferencing with minimal dependency on external memory or processing. Supporting 4/8/16-bit weights and BFloat16 formats, the module handles up to 80 million 4-bit parameters. Compatibility with industry-standard frameworks such as ONNX, TensorFlow, PyTorch, Keras, and TensorFlow Lite, combined with a robust cross-platform SDK, ensures effortless deployment across x86, ARM, and RISC-V platforms running Ubuntu, Windows, or Android.

The MX3-2280-M-4 features integrated on-module memory and a scalable architecture supporting up to 16 interconnected chips, enabling up to 96 TOPS of performance. Its automated model compilation tools eliminate the need for retraining or quantization, offering a plug-and-play solution that sets a new standard for efficiency, adaptability, and performance in edge AI systems.

Specification Table