RK1828 AI Dual Matrix Kit
3D RAM Stacking Architecture, Boosting Edge AI
Dual AI Matrix Kit is a motherboard designed for edge large model deployment. It supports LLM, VLM, AI Agent, and multimodal model implementation, while also remaining compatible with traditional algorithms such as CNN and RNN. The device adopts a heterogeneous modular architecture featuring a main controller plus an independent AI coprocessor, both designed as modules. This allows flexible selection of modules with different computing power and memory specifications based on application requirements. The device integrates CAN-FD and RS485 industrial interfaces, making it suitable for both general AI and industrial intelligence deployment needs.
Overview
Features
Dual-core Heterogeneous Design
Adopt a split heterogeneous architecture consisting of primary and secondary modules interconnected at high speed via PCIe. The primary module handles overall task scheduling and data preprocessing, while the secondary module focuses on core AI inference computing to enable model parallelism and multi-task coordination.
8K Codec & Multi-channel 4K Parallel Processing
The host module integrates high-performance VPU for 8K encoding/decoding, with hardware-accelerated image scaling and cropping to process multi-channel 4K streams. Combined with AI computing of slave modules, it delivers real-time video AI analysis for edge HD intelligent processing.
High-Compute Module for Diverse Scenarios
The slave module is compatible with high-compute high-bandwidth modules like RK1820, RK1828 and HOUMO LQ50, providing solid computing base for LLM, VLM, MoE and AI Agent. Its processing-in-memory structure lifts DRAM bandwidth to boost large model inference TPS and reduce initial frame latency.
Low-latency Edge Computing with Millisecond-level Response
Local AI processing eliminates cloud upload and network dependency, avoiding bandwidth fluctuation impacts to deliver millisecond response and robust data privacy. All business data is stored and computed locally to comply with data security and industry privacy regulations.
Onboard SATA and PCIe interfaces for massive storage expansion
Equipped with onboard SATA3.0 and high-speed M.2 PCIe interfaces compatible with SATA and NVMe SSDs. High-speed solid-state drives can be flexibly installed to expand storage to terabyte level, accommodating large model files and massive business data to meet local high-capacity storage requirements.
Abundant Expansion Interfaces
It features dual Gigabit Ethernet, USB3.0, USB2.0, HDMI, M.2, Mini PCIe, RS485, CAN-FD and opto-isolated DI/DO ports for flexible access to industrial peripherals and extended industry applications.
Robust Network Connectivity
Equipped with dual Gigabit Ethernet ports (RJ45) and an M.2 interface for WiFi module expansion. It also supports 5G and 4G LTE extensions, delivering significantly higher data transmission rates.
Comprehensive Development Materials
Supporting source code, practical tutorials, detailed technical documents and development tools are provided to facilitate secondary development, streamline procedures and lower project development barriers.
Application
Robot
Features multi-sensor fusion and real-time motion planning for autonomous navigation and precise tasks in complex environments, ideal for inspection and service.
Edge Computing
Decentralizes computing to the edge for real-time local AI inference, minimizing latency and bandwidth costs for IoT endpoints.
Industrial Automation
Integrates smart control and sensing for automated production, driving cost-effective digital transformation in manufacturing.
Smart Security
AI-powered visual analysis enables 24/7 monitoring and active anomaly alerts, shifting security from passive to proactive.
Specifications
| RK1828 AI Dual Matrix Kit | ||
|---|---|---|
| Basic Specifications | Host module | Optional modules as follows: RK3588module (Core-3588JD4): LPDDR4/LPDDR4x (4/8/16/32GB) + eMMC (32/64/128/256GB), 6TOPS(INT8) RK3576 module (Core-3576JD4): LPDDR4/LPDDR4x (2/4/8/16GB) + eMMC (16/32/64/128/256GB), 6TOPS(INT8) |
| Video encoding | RK3588: 8K@30fps H.265/H.264 RK3576: 4K@60fps H.265/H.264 | |
| Video decoding | RK3588: 8K@60fps H.265/VP9/AVS2, 8K@30fps H.264/AVC/MVC, 4K@60fps AV1, 1080P@60fps MPEG-2/-1/VC-1/VP8 RK3576: 8K@30fps H.265/VP9/AVS2/AV1, 4K@120fps H.265/VP9/AVS2/AV1, 4K@60fps H.264/AVC | |
| Accelerator expansion (Slave module) | 2 × M.2 M-KEY (Supports 2260/2280 accelerator cards & other M.2 modules, inside the device) * Users can install any brand M.2 M-KEY accelerators; standard equipped with RK1828/HOUMO LQ50 modules ○RK1828: Onboard 5GB DRAM, 20TOPS(INT8), runs LLMs up to 7B smoothly ○LQ50: 48GB LPDDR5, 160TOPS(INT8), runs LLMs up to 35B smoothly | |
| Storage expansion | 1 × M.2 M-KEY (Expandable with PCIe2.0 x1 NVMe 2280 SSD; unavailable when core-3576JD4 serves as the host module) 1 × SATA3.0 (Expandable with SATA3.0 SSD; unavailable when core-3576JD4 serves as the host module) 1 × TF Card | |
| Motherboard power | DC 12V/7A (5.5×2.1mm, support 9V~36V wide voltage input) | |
| Accelerator power | 2 × 12V (2pcs 4Pin-1.25mm wafer sockets) | |
| Power consumption | RK3588 + 1 × RK1828: Normal: 9.6W(12V/800mA), Max: 28.8W(12V/2400mA), Min(Sleep): 0.72W(12V/60mA) RK3588 + 2 × RK1828: Normal: 15.6W(12V/1300mA), Max: 42W(12V/3500mA), Min(Sleep): 0.72W(12V/60mA) RK3588 + 1 × LQ50: Normal: 8.4W(12V/700mA), Max: 45.6W(12V/3800mA), Min(Sleep): 0.84W(12V/70mA) RK3588 + 1 ×LQ50 DUO: Normal: 12W(12V/1000mA), Max: 63.6W(12V/5300mA), Min(Sleep): 0.84W(12V/70mA) RK3576 + 1 × RK1828: Normal: 8.4W(12V/700mA), Max: 28.8W(12V/2400mA), Min(Sleep): 1.2W (12V/100mA) | |
| OS | Linux | |
| Dimension | 148.0mm × 106.0mm × 36.4mm | |
| Weight | Net weight: 437g (with heatsink), Gross weight with package: 1362g | |
| Environment | Operating Temperature: -20℃～60℃, Storage Temperature: -20℃～70℃, Storage Humidity: 10%~90%RH (non-condensing) | |
| Interface Specifications | Network | 2 × Gigabit Ethernet (1000Mbps/RJ45) Wi-Fi & Bluetooth: Support dual-band 2.4GHz/5GHz Wi-Fi and Bluetooth 5.2 (Expandable via M.2 E-KEY; unavailable with core-3576JD4 host module) 4G: Expandable via Mini PCIe 5G: Expandable via M.2 B-KEY |
| Video output | 1 × HDMI2.1(Up to 4K@60Hz) | |
| Audio | 1 × Line In (3.5mm audio jack), 1 × Line Out (3.5mm audio jack) | |
| USB | 2 ×USB3.0 (Max: 1A), 2 × USB2.0 HOST (4Pin-2mm), 1 ×Type-C (USB2.0 OTG), 1 × Type-C (Debug) | |
| Button | 1 × Power key, 1 × Upgrade key, 1 × Reset key | |
| Others | 1 × SIM Card, 1 × SATA Power (4Pin-5.08mm), 2 × FAN (4Pin-1.25mm), 1 × Phoenix terminal block (8Pin-3.5mm): 1 × RS485, 1 × CAN-FD, 2 × Opto-isolated DI/DO | |
