Rockchip RK1828 M.2 2230 AI Accelerator Module – 20 TOPS @ INT8 for local LLM & VLM

Rockchip RK1828 M.2 2230 AI Accelerator Module  – 20 TOPS @ INT8 for local LLM & VLM

The Rockchip RK1828 M.2 2230 AI Accelerator Module is a high-performance edge AI co-processor designed to accelerate on-device Large Language Models (LLMs), Vision Language Models (VLMs), and deep learning inference. Featuring an integrated 20 TOPS INT8 Neural Processing Unit (NPU), the RK1828 enables fast, power-efficient AI processing without relying on cloud services.

Designed in the compact M.2 2230 form factor, the module integrates seamlessly into embedded systems, industrial computers, AI gateways, robotics platforms, and edge servers through a PCIe interface. Built on Rockchip's advanced AI architecture, it supports multiple AI computation formats including INT4, INT8, INT16, FP8, FP16, and BF16, allowing optimized deployment of modern neural network models. It is specifically targeted at local deployment of up to 7B parameter LLMs and multimodal AI applications.

With support for popular AI frameworks such as TensorFlow, PyTorch, MXNet, and Caffe (via model conversion), the RK1828 simplifies AI development while delivering high throughput and low latency for applications including intelligent vision, speech recognition, industrial automation, smart retail, healthcare, and edge AI computing.

Datasheet

Features

• 20 TOPS AI performance at INT8 precision
• Optimized for local Large Language Models (LLM) and Vision Language Models (VLM)
• Supports deployment of up to 7B parameter AI models
• Compact M.2 2230 form factor for embedded systems
• PCIe interface for high-speed host communication
• Supports INT4, INT8, INT16, FP8, FP16, and BF16 inference
• High-bandwidth integrated memory architecture for efficient AI workloads
• Compatible with TensorFlow, PyTorch, MXNet, Caffe, and other AI frameworks via conversion tools
• Low-power edge AI acceleration without cloud dependency
• Suitable for industrial AI, robotics, machine vision, smart retail, healthcare, and AI gateways

Specifications

• AI Processor: Rockchip RK1828 AI Co-Processor
• Form Factor: M.2 2230
• AI Performance: Up to 20 TOPS (INT8)
• AI Precision: INT4, INT8, INT16, FP8, FP16, BF16
• CPU: Triple 64-bit RISC-V Cores
• Target Models: Local LLM, VLM, CNN, Transformer Models
• Maximum Model Support: Up to 7B Parameter Models
• Host Interface: PCIe
• Framework Support: TensorFlow, PyTorch, MXNet, Caffe (via conversion)
• Applications: Edge AI, Robotics, Industrial Automation, Smart Vision, AI Gateway, Intelligent Devices

Applications

• Local AI assistants
• Large Language Model (LLM) inference
• Vision Language Model (VLM) deployment
• Edge AI servers
• Industrial machine vision
• Robotics and autonomous systems
• Smart surveillance
• Intelligent retail
• Healthcare AI devices
• Embedded AI development

Package Includes

• 1x Rockchip RK1828 M.2 2230 AI Accelerator Module  – 20 TOPS @ INT8 for local LLM & VLM