RK182X 3D RAM Stacking Development Kit

$959.00

Taxes included. Shipping calculated at checkout.

Spec:RK3588S (8G+64G) + RK1820 RK3588S (8G+64G) + RK1820 RK3588S (8G+64G) + RK1828 RK3588S (16G+256G) + RK1820 RK3588S (16G+256G) + RK1828 RK1828 SoM RK1820 SoM

Manual WikiDownload

Brand Firefly

Availability

5 in stock

RK182X Development Kit

RK182X Development Kit is an edge-AI mainboard designed for large-model deployment, supporting LLMs, VLMs, AI Agents, and multimodal large models, while remaining compatible with traditional algorithms such as CNN and RNN. Built on a heterogeneous split architecture with a host controller and a discrete AI co-processor — both modular — it allows flexible pairing of compute and memory modules to match your business requirements. With integrated CAN-FD and RS485 industrial interfaces, it meets the needs of both general AI and industrial intelligence applications.

Dual-Core Architecture

It features a heterogeneous dual-module architecture interconnected via high-speed PCIe. The primary module manages task scheduling and data preprocessing, while the secondary module delivers dedicated AI inference compute, supporting model parallelism and multi-task coordination. Optimized for edge scenarios such as robot navigation and multimodal perception, it ensures efficient resource allocation, low-latency response, and stable performance under complex workloads.

8K Video Codec Support

Integrating high-performance VPU, the main module delivers hardware 8K video encoding and decoding, along with hardware-accelerated image scaling and cropping — effortlessly handling parallel processing of multiple 4K video streams. Paired with the RK182X co-processor's AI compute, it performs real-time AI analysis of video streams with high efficiency, fully meeting the demands of HD video intelligence in edge scenarios.

High-Bandwidth DRAM

Integrates high-bandwidth embedded DRAM (2.5GB/5GB variants), delivering a token generation rate of over 100 tokens/s with an end-to-end latency as low as 0.1s for efficient local AI processing.

Millisecond-Level Low Latency

With local AI processing, network dependency is minimized — no need to upload data to the cloud, and operations are unaffected by network bandwidth or speed fluctuations, enabling millisecond-level response and data privacy protection. All business data is stored and processed locally, meeting data security regulations and industry privacy compliance requirements.

Powerful Network Capabilities

Features dual Gigabit Ethernet ports (RJ45), with WiFi module expansion via M.2 and support for 5G / 4G LTE connectivity for higher network speeds.

Onboard SATA/PCle Interfaces

Onboard SATA 3.0 and M.2 PCIe interfaces support both SATA and NVMe SSDs, allowing flexible expansion to TB-level capacity — ample space for large model files and massive business data to meet local high-capacity storage needs.

Rich Expansion Interfaces

Equipped with a rich set of interfaces — dual Gigabit Ethernet, USB 3.0, USB 2.0, HDMI, M.2, Mini PCIe, RS485, CAN-FD, and opto-isolated DI/DO — it flexibly supports diverse industrial peripherals and application expansions.

Development Materials

A full suite of development resources—including source code, tutorials, and technical documentation—is provided to simplify and accelerate the development process.

3D DRAM Stacking Architecture

Built on a heterogeneous split architecture combining a host SoC (standard Rockchip RK3588) with a discrete AI co-processor, dual M.2 slots accommodate two Rockchip RK182X accelerator cards. Engineered for high-performance edge AI inference, the RK182X features a 3D-stacked RAM design whose ultra-high DRAM bandwidth removes data transfer bottlenecks, enabling smooth execution of 3B/7B-parameter large models. It significantly reduces response latency for LLM and VLM workloads while markedly boosting AI processing efficiency.

Application Scenarios

Widely adapt to intelligent gateway server, robots, edge computing, industrial automation, Intelligent security, smart city and other products and fields.