Product Overview

This product is more than an open-source 8-core Linux AI box. Built on the powerful Rockchip RK3576 16GB RAM, it is designed to make edge AI development and deployment faster and easier.

To improve the developer experience, Seeed offers reComputer AI Lab — a platform that helps you deploy AI applications with one-click tools and ready-to-use resources. It includes optimized edge AI model demos for CV, LLM, VLM, STT, and TTS, interactive tutorials, deployment tools, containerized applications, and community projects to speed up development on Rockchip.

With rich I/O, multiple OS support, wireless expansion options, and active cooling, this compact AI box is ready for a wide range of edge AI workloads.

It is ideal for smart vision systems, voice-enabled devices, AI agents, robotics, and industrial edge applications.

The reComputer RK3576 features an advanced CPU architecture with

• 4 × Cortex-A72 @2.2GHz
• 4 × Cortex-A53 @2.0GHz
• LPDDR5 (4GB/8GB/16GB)
• ARM Mali GPU-G52 MC3

With a built-in 6 TOPS NPU, it delivers efficient AI inference directly on the device, reducing latency and enhancing privacy without relying on the cloud.

AI performance highlights include:

• Vision CNN: YOLO11@77.9FPS(640*640)
• LLM: capable for deepseek-r1-distill-qwen 7b support
• VLM: capable for qwen2.5-vl 3b
• STT: whisper_base_20s (RTF 0.218) with real-time processing support
• TTS: mms_tts_eng_200 (RTF 0.069) with real-time processing support

Check the details of the benchmark: https://sensecraft.seeed.cc/ai-lab/tutorials/rk/benchmark/cv

It offers flexible expansion through

• 1 × M.2 M Key: PCIe 2.1x1, supports SSD expansion and AI accelerators such as Hailo and Rockchip, up to 26TOPS
• 1 x miniPCIe for wireless expansion, like 4G LTE, LoRaWAN, Wi-Fi HaLow

It also supports a wide range of popular AI models, from computer vision and speech to LLM-powered edge intelligence, including: YOLO, MobileNet V2, RetinaFace, CLIP, Whisper, DeepSeek-R1, Qwen2-VL, ...

Framework support includes: ONNX, PyTorch, TensorFlow, TensorFlow Lite, Caffe, Darknet, ...

Toolchain support: RKNN-Toolkit2

This broad compatibility makes it easy to migrate existing models and build integrated edge AI applications across Perception AI, Generative AI, Agent and Physical AI.

Deploying AI at the edge is often complicated — from environment setup and model conversion to performance optimization. To simplify this process, Seeed created reComputer AI Lab, a developer platform designed to lower the barrier to edge AI development.

It provides a complete collection of optimized AI model demos for CV, LLM, and VLM, along with interactive tutorials, deployment tools, containerized AI applications, project examples, and community resources. It is built to accelerate AI development and deployment on Rockchip, Raspberry Pi, and NVIDIA Jetson platforms.

For Rockchip users, reComputer AI Lab includes:

1. Ready-to-Use Models — Access pre-optimized models without starting from scratch. AI Lab includes models such as YOLO11, MobileNet, CLIP, and Whisper, optimized for Rockchip platforms. With the RKLLM toolchain, it also supports edge LLM deployment.
2. One-Command Deployment Tools — Skip complex build environments and speed up deployment with one-command tools. With Seeed-optimized RKNN-Toolkit2, models can be converted from ONNX to NPU deployment in just minutes, making AI application deployment faster and easier.
3. Comprehensive Tutorials, Projects, and Community — From basic NPU benchmarking to real-world applications such as YOLO11-based industrial detection, AI Lab provides step-by-step guides, practical project examples, and community support to help developers build faster.

The device comes with Armbian pre-installed for a fast out-of-the-box experience.

With the help of Armbian, we are able to provide long-term maintained system images, security updates, encrypted OS options, and OTA upgrade support for production-ready deployment.

It also supports multiple operating systems through the Rockchip ecosystem, giving developers more flexibility across different projects: Ubuntu, Android, Debian.

The reComputer RK3576 Series supports simultaneous 3-display output, giving developers more flexibility for digital signage, control centers, smart retail, and multimedia systems.

Supported output interfaces include:

• HDMI
• MIPI DSI
• Type-C (DP Alt Mode)

It also offers strong multimedia capabilities:

• Up to 4K@60fps video encoding with H.265 / H.264
• Up to 8K@30fps video decoding with H.265 / H.264 / AV1 / AVS2

This makes it ideal for high-resolution AI vision and multimedia applications.

Check More Demos on reComputer AI lab.

• reComputer RK35XX Series Flyer
• reComputer RK3576 Schematic
• recomputer_rk3576 Top Cover ABS
• recomputer_rk3576 Bottom Cover ABS
• reComputer RK3576 User Manual
• Rockchip RK3576 Datasheet

Array