RM182X M.2 AI Accelerator Module RM1820MC0 RM1828MC0 Rockchip 20TOPS LLM Inference Module

RM182X M.2 AI Accelerator Module RM1820MC0 RM1828MC0 Rockchip 20TOPS LLM Inference Module

The RM182XMC0 series are high‑performance AI co‑processor modules from Rockchip, including 

RM1820MC0 (for 3B LLMs) and RM1828MC0 (for 7B LLMs). Adopting M.2 Key‑M gold‑finger interface

with PCIe2.1 communication, the compact 22×80mm form factor enables easy integration into edge and

 embedded devices with minimal base‑board design effort, speeding up your AI product development.

Delivering up to 20 TOPS @INT8 AI computing power,the module supports INT4 / INT8 /

INT16 / FP8 / FP16 / BF16 data types for local LLM inferenceand machine‑vision tasks.

It comes with high‑bandwidth on‑board DRAM: 2.5GB for RM1820MC0 and 5GB for RM1828MC0.

Pre‑adapted mainstream models include Qwen2.5 / Qwen3 series, FastVLM, MobileNet, ResNet50V2,

YOLOv5s, YOLOv6s, YOLOv8s and more. Complete software toolchains are provided: RKNN3 Toolkit,

RKNN3 Runtime API and RKNN3 Model Zoo for model conversion, deployment and rapid validation.
Power input: 3.3V from M.2 slot plus additional 12V power supply. Different power and latency performance

are available under energy‑saving / standard operation modes for large‑model inference.
As the original manufacturer, we offer technical support and customization services for system integrators

and AI solution developers.

Core Features
1. Rockchip RM182X AI co‑processor module, 20 TOPS@INT8 computing performance
2. Two variants: RM1820MC0 supports up to 3B LLMs; RM1828MC0 supports up to 7B LLMs
3. M.2 Key‑M interface via PCIe2.1, compact 22×80mm size, simple hardware integration
4. On‑board high‑bandwidth DRAM: 2.5GB (RM1820MC0) / 5GB (RM1828MC0)
5. Multi‑precision support: INT4 / INT8 / INT16 / FP8 / FP16 / BF16
6. Compatible with Qwen series LLMs, YOLO, MobileNet, ResNet and other AI models
7. Full software stack: RKNN3 Toolkit, Runtime API, Model‑Zoo deployment examples
8. Dual‑power requirement: 3.3V (M.2 slot) + external 12V power input

Application Scenarios
Local large‑language‑model edge inference
Embedded machine vision & object detection
Edge AI computing box / AI gateway upgrade
Smart industrial equipment & on‑site AI terminals
AIoT embedded solutions, secondary development projects