Project Blueprints

Reference architectures for common Rockchip products: SoC choice, system blocks, software stack and the pitfalls that cost teams time. Use them as a starting point for your own design.

Multi-Camera AI NVR on RK3588

RK3588

Ingest 8–16 RTSP streams, decode on the VPU, run YOLO on all three NPU cores and record to NVMe/SATA.

System blocks

IP cameras → GbE / 2.5 GbE → MPP hardware decode (H.264/H.265)
RGA resize/letterbox → RKNN (one model instance per NPU core)
CPU: tracking, NMS, rules engine → MQTT / REST / web UI
Recording: NVMe over PCIe 3.0 x4 or SATA through the combo PHY

Software stack

Debian or Buildroot on the vendor 6.1 kernel
MPP + RGA + RKNN runtime (C API)
GStreamer or FFmpeg with Rockchip MPP support
Web dashboard + MQTT

Watch out for

DDR bandwidth saturates before NPU TOPS do
Thermal throttling under sustained decode + NPU load
Stream reconnect and watchdog logic matter more than the model

Build something like this →

Industrial Gateway / HMI on RK3568

RK3568

Dual-GbE gateway with CAN-FD, RS-485 and an LVDS/MIPI touch display, on industrial-temperature RK3568J.

System blocks

2× GbE (field network + uplink), optional 4G/5G over mini-PCIe/M.2
CAN-FD ×2, RS-485 via UART + transceivers, isolated DI/DO
LVDS or MIPI DSI panel with capacitive touch over I2C
eMMC + SPI NOR for A/B firmware and recovery

Software stack

Yocto or Buildroot with a mainline-based kernel
Qt/QML or a web-kiosk HMI
Modbus, OPC UA and MQTT bridges
Signed A/B OTA updates (RAUC / SWUpdate)

Watch out for

RGMII delays on the second GMAC
IO-domain voltages on mixed 1.8 V / 3.3 V banks
EMC/ESD on field I/O

Build something like this →

Low-Cost Smart Camera on RV1106

RV1106

A sub-$20-BOM camera with on-device person or face detection, H.265 streaming and PoE.

System blocks

MIPI CSI-2 sensor (2–5 MP) → ISP with IQ tuning file
H.265 encoder → RTSP / ONVIF
NPU (0.5–1 TOPS): person or face detection at reduced resolution
10/100 Ethernet (integrated PHY) + PoE, SPI NAND storage

Software stack

Buildroot from the RV1106 SDK
rkaiq 3A + IQ files for the chosen sensor
RKNN runtime (C API)
Lightweight RTSP server

Watch out for

Tight RAM: plan buffers and model size up front
Sensor driver + IQ availability drives sensor choice
Night mode (IR-cut) control and IQ switching

Build something like this →

Edge AI Smart Display on RK3576

RK3576

Android or Linux smart display with camera-based presence and gesture detection, running a local assistant.

System blocks

MIPI DSI / eDP panel with touch, HDMI 2.1 out for a second screen
MIPI CSI camera → ISP → NPU for presence and gesture detection
Wi-Fi 6 / BT over SDIO or PCIe, dual GbE optional
LPDDR5 + UFS/eMMC

Software stack

Android 14 (vendor SDK) or Debian
RKNN + RKLLM for on-device small language models
Kiosk launcher and OTA

Watch out for

Panel timing and init sequence bring-up
Camera IQ tuning for indoor lighting
Keeping NPU and GPU workloads from starving the UI

Build something like this →

Built something on Rockchip you'd like featured? Share your project.

¿Desarrollas un producto basado en Rockchip?

Del diseño de hardware a la integración de Linux, Android e IA: consigue soporte de ingeniería para tu proyecto Rockchip.

Solicitar soporte de ingeniería Ver qué hacemos