RM182X · M.2 2280 AI Compute Co-Processor Module

• RM182X Hardware Specification
• RM182X Hardware Design Guide
• RM182X Downloads

Overview

RM182XMC0 (including RM1820MC0 and RM1828MC0) is a high-performance AI compute co-processor module supporting 3B / 7B large models (RM1820MC0 supports 3B, RM1828MC0 supports 7B). It is designed for AI inference and machine vision — especially artificial-intelligence-related applications.

The module carries Rockchip's RK182X co-processor and uses an M.2 Key M gold-finger connector, measuring 22mm × 80mm. Customers only need to design a simple carrier board to integrate the module into edge or embedded devices, enabling fast development.

Product Appearance

Highlights

The core value

Large-model inference delivered as a standard M.2 2280 card. As long as your host has a free PCIe M.2 slot, dropping in one card adds 20 TOPS of local compute — no host replacement, no cloud round-trip.

Feature	Specification
Compute	20 TOPS @ INT8
Data types	INT4 / INT8 / INT16 / FP8 / FP16 / BF16
On-chip DRAM	RK1820: 2.5GB ｜ RK1828: 5GB (hundreds of GB/s bandwidth)
CPU	3× 64-bit RISC-V (SRV, VRV0, VRV1)
Interface	M.2 Key M, PCIe 2.1 communication
Dimensions	22mm × 80mm
LLM support	RK1820 → 3B ｜ RK1828 → 7B

Supported Models

Category	Models
Large language models	Qwen 2.5 0.5B / 1.5B / 3B, Qwen 3 0.6B / 1.7B
Multimodal	FastVLM
Image classification	MobileNetV1, MobileNetV2, ResNet50V2
Object detection	YOLOv5s, YOLOv6s, YOLOv8s

Software Stack

Component	Description
RKNN3 Toolkit	PC-side development kit for model conversion and quantization
RKNN3 Runtime	On-device runtime API
RKNN3 Model Zoo	Model conversion and deployment examples

Measured Performance

Input tokens: 128

Model	Mode	TTFT (ms)	TPS (tokens/s)	Avg power	Peak power
Qwen2.5 3B	Power-save	144.455	54.537	5.35W	15.14W
Qwen2.5 3B	Standard	96.05	77.945	7.2W	24.86W
Qwen2.5 7B	Power-save	288.255	35.044	7.02W	16.11W
Qwen2.5 7B	Standard	186.435	51.476	10.4W	27.28W

TTFT and TPS

TTFT (Time To First Token) is first-token latency — how long before output starts. TPS (Tokens Per Second) is generation speed — how fast output flows. Together they define the interactive feel.

Power Requirements

Besides 3.3V drawn from the M.2 socket, the module requires an additional 12V supply from the carrier board.

Rail	Source	Specification
12V	Dedicated header on carrier board	12V / 3A
3.3V	M.2 socket on carrier board	3.3V / 3W

12V is mandatory

The RK1820 M.2_2280 Module requires the carrier board to supply an additional 12V/3A rail. The M.2 socket's 3.3V alone cannot drive the module — a dedicated 12V supply must be designed on the carrier board.

Application Block Diagram

        HOST                              RM182XMC0
  ┌──────────────┐                    ┌──────────────────┐
  │              │   PCIe 2.1 1lane   │                  │
  │   PCIe RC    │◄──────────────────►│   PCIe0  (EP)    │
  │              │                    │                  │
  │  PCIE0_PERSTN├───────────────────►│  RESETN_IN       │
  │  PCIE0_CLKREQ├───────────────────►│  PCIe CLKREQn    │
  │  PCIE0_WAKEN ├───────────────────►│  PCIe WAKEn      │
  │              │                    │                  │
  │   FAN_CTRL   ├───────────────────►│  Fan speed CTL   │
  │   FAN_SPEED  │◄───────────────────│  Fan speed DET   │
  │              │                    │                  │
  │   3.3V_IN    ├───────────────────►│  VCC_IN 3.3V     │
  │   12V        ├───────────────────►│  VCC_IN 12V      │
  └──────────────┘                    └──────────────────┘

Next Steps

• Full specifications → RM182X Hardware Specification
• Carrier board design → RM182X Hardware Design Guide
• Get materials → Downloads