RM1828MC0

RM1828MC0

$259.00
$259.00
Skip to product information
RM1828MC0
  • RM1828MC0
  • RM1828MC0
  • RM1828MC0
  • RM1828MC0
  • RM1828MC0
  • RM1828MC0
RM1828MC0

RM1828MC0

$259.00
$259.00
Shipping calculated at checkout.

RK1828 M.2 2280 accelerator with 20 TOPS INT8 AI and 5GB 3D-stacked memory for local language and vision models.
  • American Express
  • Apple Pay
  • Google Pay
  • Mastercard
  • PayPal
  • Shop Pay
  • Union Pay
  • Visa
RM1828SA0 / RM1828MC0 AI Accelerator Cards

RM1828SA0 / RM1828MC0 AI Accelerator Cards

RM1828SA0 and RM1828MC0 are RK1828-based AI inference modules for on-device large models. A 3D-stacked RAM architecture with 5GB of high-bandwidth DRAM reduces memory bandwidth bottlenecks, while the integrated NPU delivers up to 20 TOPS. Single-card configurations support efficient 8B model inference, two cards support 12B models, and four cards support models up to 27B. SODIMM and M.2 interfaces enable integration into a range of edge devices

RM1828MC0
20 TOPS AI Performance for Local Model Deployment

20 TOPS AI Performance for Local Model Deployment

The NPU delivers up to 20 TOPS of INT8 performance and supports INT4, INT8, INT16, FP8, FP16 and BF16 mixed-precision computing. Support for 8B model inference enables local AI processing on edge devices.

Integrated 5GB High-Bandwidth DRAM

Integrated 5GB High-Bandwidth DRAM

Integrated 5GB high-bandwidth DRAM reduces external memory access overhead and improves inference responsiveness for LLMs and VLMs. Reported generation speeds reach up to 100 tokens/s, with end-to-end latency as low as 0.1 seconds.

Standard SODIMM / M.2 Interfaces with Multi-Card Support

Standard SODIMM / M.2 Interfaces with Multi-Card Support

The modules use standard SODIMM or M.2 B+M-key interfaces. The SODIMM version features a pinout compatible with NVIDIA Jetson Orin Nano. Multi-card configurations support efficient 8B model inference with one card, 12B models with two cards, and up to 27B models with four cards.

Application Scenarios

Robotics, edge AI computing, industrial automation, intelligent security, intelligent driving, smart cities, traffic management, home storage and retail automation.

Robots
Robots
Edge Computing
Edge Computing
Industrial Automation
Industrial Automation
Intelligent Security
Intelligent Security
Smart City
Smart City
Traffic Management
Traffic Management

Interfaces

Interfaces

Specifications

RM1828MC0RM1828SA0
Basic Specifications SOC

Rockchip RK1828

RGA
Built-in 2D graphics engine; supports BT.601/BT.709 pixel format conversion and dithering operation. Max resolution: source image 8192×8192, target image 4096×4096
NPU

NPU computing power: 20 TOPS (INT8); supports mixed-precision computation of INT4/INT8/INT16/FP8/FP16/BF16

Codec

JPEG Encoding: Supports baseline (sequential DCT), JFIF 1.02. Image size range: 16×16 ~ 65520×65520; supports YUV400/YUV420/YUV422/

YUV444

JPEG Decoding: Supports baseline (sequential DCT), JFIF 1.02. Image size range: 48×48 ~ 65520×65520; supports YUV400/YUV420/YUV422/

YUV440/YUV411/YUV444/RG888/RGB565

RAM

On-chip 5GB DRAM

OS

Linux

Software support

Supports YOLO object-detection algorithms.

Supports private deployment of large-scale Transformer-based models, such as Gemma, ChatGLM, Qwen, Phi and other LLM series.

Supports classic network architectures including CNN, RNN and LSTM; supports RKNN model import & export.

Power

DC 12V

DC 5V

Interface type

M.2 B-KEY / M.2 M-KEY

Gold-finger (SODIMM 260-Pin standard interface, 0.5 mm pitch)

Dimension

22mm × 80mm × 6mm (without heat sink)

25mm × 81mm × 23.05mm (with heat sink)

69.6mm × 45.0mm × 4.85mm (without heat sink)

69.6mm × 45.0mm × 20.67mm (with heat sink)

Weight

11g (without heat sink), 53g (with heat sink)

21g (without heat sink), 75g (with heat sink)

Environment

Operating temperature: -20℃ ~ 60℃, Storage temperature: -20℃ ~ 70℃, Storage humidity: 10% ~ 90%RH (non-condensing)

Interface Specifications Fan interface
1 × FAN (4Pin-1.25mm)
N/A