RK3588 Edge Computing Solution: Real-World Testing

Edge computing has been a hot topic for years, but when it comes to actual deployment, many people ask a very practical question: which chip should you use?

In the edge computing space, NVIDIA's Jetson series is the go-to for high-end solutions. But for the vast number of scenarios requiring mid-range AI performance — industrial inspection, video analytics, smart terminals — a single RK3588 may be more than enough.

In this article, based on Qiyun Zhixun's real-world project experience, we dive deep into the edge computing capabilities of the Rockchip RK3588.

1. What Exactly Is the RK3588?

The RK3588 is Rockchip's flagship SoC, manufactured on an 8nm process node.

In short: CPU performance approaching desktop-class i5, 6 TOPS NPU, exceptional video processing capability, and rich I/O — one of the most powerful domestic SoCs for edge computing.

2. Why Choose RK3588 for Edge Computing?

Key Advantages of RK3588:

1. Exceptional price/performance: 6 TOPS at ¥300-600, one of the cheapest options at this performance level
2. Outstanding video processing: 8K decode + 8K encode, real-time multi-stream 4K processing
3. Extremely rich interfaces: PCIe 3.0, USB 3.1, SATA, multiple UART/SPI/I2C
4. High degree of domestic control: Fully controllable, ideal for domestic substitution projects
5. Manageable power consumption: Typical 5-15W, passive cooling sufficient for most scenarios

3. RK3588 NPU Benchmark Data

Data below is based on Qiyun Zhixun's standard core board, running Ubuntu 22.04 + RKNN Toolkit 2.0.

3.1 Classic Model Inference Performance

3.2 Multi-Stream Video Processing

4. Real-World Project Applications

Project 1: Industrial Visual Inspection System

Scenario: PCB surface defect detection

• 2 industrial cameras (5MP, GigE interface)
• RK3588 core board + custom carrier board
• AI model: Custom YOLOv8-based defect detection
• Detection speed: <500ms per PCB
• Detection accuracy: >99.5% defect detection rate

Project 2: Multi-Stream Intelligent Video Analytics Gateway

Scenario: Campus security video analytics

• 16× 1080P IP cameras connected
• AI functions: Person detection, license plate recognition, intrusion alerts
• 4 simultaneous AI streams, round-robin processing for all 16

Project 3: Smart Traffic Edge Terminal

Scenario: Intersection traffic flow statistics and violation detection

• 4× HD cameras
• RK3588 core board + 4G/5G communication module
• AI functions: Vehicle detection, LPR, traffic flow statistics
• Wide-temperature design (-40°C to +70°C)

5. Development Experience

NPU Model Deployment Flow

Train model (PyTorch/TensorFlow)
    ↓
Export ONNX model
    ↓
RKNN Toolkit conversion (quantization: INT8/FP16)
    ↓
Deploy to RK3588 (RKNN Runtime C API)
    ↓
Integrate into application (C/C++/Python)

Thermal Design

• Passive cooling: Large aluminum heatsink + thermal pad, controllable within 70°C
• Active cooling: Small fan + heatsink, controllable within 55°C
• Industrial scenario: Metal enclosure for whole-unit heat dissipation, fanless design

Qiyun Zhixun RK3588 Development Capabilities

Based on the RK3588, Qiyun Zhixun provides full-spectrum services from core board design to complete product development:

• RK3588 Core Board Design: High-speed signal, power design, thermal solution
• Custom Carrier Board: Interface and function design based on application scenarios
• AI Model Deployment: Model conversion, quantization optimization, NPU adaptation
• System Integration: Linux customization, driver development, application development
• Complete Product Delivery: From solution design to mass production