News and Information
Insight into forward-looking trends, brand market dynamics
Behind the Implementation of Smart Cities: What Hardware Support Do Edge Computing Nodes Require?
Smart cities are no longer just a concept. They are becoming part of everyday life.
When we walk through a city, smart streetlights can automatically adjust brightness according to pedestrian flow. Traffic lights can dynamically change signal timing based on real-time vehicle volume. Electronic screens at bus stops can accurately display the arrival time of the next bus. Smart trash bins in parks can automatically notify sanitation workers when they are nearly full.
Behind all these “smart” applications, there is one essential technology: edge computing.
Today, let’s talk about the hidden hero behind smart city implementation: edge computing nodes, and what kind of hardware support they require.
To understand the value of edge computing, we first need to understand the limitations of cloud computing.
In a typical cloud computing architecture, all data must be transmitted to a central data center for processing. Video captured by cameras is sent to the cloud for analysis, sensor data is uploaded to the cloud for computing, and control commands are then sent back from the cloud to terminal devices.
This “device-to-cloud” architecture has several natural limitations.
Sending data to the cloud and receiving a response may take hundreds of milliseconds. For autonomous driving, a few hundred milliseconds can mean several meters of braking distance, which can be critical.
A single HD camera can generate tens of gigabytes of data. If tens of thousands of cameras across a city all upload data to the cloud, even very large bandwidth capacity can become overwhelmed.
Once the network is interrupted, the cloud connection is lost. Terminal devices then become “blind” and “deaf,” unable to respond intelligently.
Edge computing was created to solve these problems.
Its core idea is simple: deploy computing nodes close to the data source so that data can be processed nearby at the edge. Services that require real-time response can be completed locally, while data that requires global analysis can still be uploaded to the cloud.
In smart cities, edge computing nodes act like the city’s “nerve endings.” They sense, process, and make decisions locally, helping the city truly become intelligent.
Different smart city scenarios have different requirements for edge computing nodes. Here are several typical examples.
At an intersection, an edge computing node connects multiple cameras and radars to analyze traffic flow, pedestrian flow, vehicle speed, and traffic violations in real time.
When a pedestrian runs a red light, the node needs to identify the behavior within milliseconds and trigger a voice warning.
When traffic flow data from multiple intersections is collected, the node can coordinate and optimize traffic light timing to reduce congestion.
When a traffic accident occurs, the node must quickly identify the event and notify traffic police and emergency services.
In key areas, edge computing nodes connect HD cameras and access control systems to perform 24-hour facial recognition, behavior analysis, and abnormal event detection.
When suspicious loitering is detected, the node needs to issue a real-time alert.
When a crowd gathering is detected, the node can analyze whether there is a potential public safety risk.
When an intrusion occurs, the node can work together with the access control system to lock down the area.
In office buildings or industrial parks, edge computing nodes connect access control systems, cameras, environmental sensors, lighting systems, and air conditioning systems to enable intelligent building management.
During the morning rush hour, the node quickly processes facial recognition requests for entry.
During working hours, it automatically adjusts air conditioning temperature and lighting brightness based on occupancy density.
After work, it switches to energy-saving mode while keeping essential security monitoring active.
In urban management, edge computing nodes can be installed on streetlight poles, bus stations, trash bins, and other city facilities.
When a trash bin is nearly full, the node can notify sanitation workers.
When a manhole cover is moved, the node can trigger an alarm.
When a streetlight fails, the node can report a maintenance request.
When illegal parking is detected, the node can capture evidence automatically.
These scenarios show that edge computing nodes face very different operating conditions from servers in data centers. This creates special hardware requirements.
Servers in data centers operate in temperature-controlled rooms with stable power supply and professional maintenance teams.
Edge nodes may be deployed on roadsides, rooftops, poles, or outdoor cabinets, exposed to wind, rain, heat, cold, dust, and vibration.
Edge nodes should operate reliably within a temperature range of -20°C to 60°C. Whether in a cold northern winter or a hot southern summer, the equipment must remain stable.
Outdoor nodes need IP50 or even IP65 protection to prevent dust and moisture from entering the system.
Roadside nodes may be continuously affected by vibration from passing vehicles. Hardware must remain stable over long-term operation in such environments.
Power supply conditions for edge nodes are often limited. A node deployed on a streetlight pole may only be able to draw power from the streetlight circuit. A node in a remote area may rely on solar power.
Total system power consumption should be controlled within tens of watts, or even lower. Edge nodes cannot operate like data center servers that consume hundreds of watts.
Edge nodes may need to integrate CPUs, GPUs, NPUs, and other computing units. Different tasks can be scheduled to different units to maximize computing power within limited power budgets.
During low-load periods, non-core components can be automatically powered down to further reduce energy consumption.
Many smart city applications require millisecond-level response, which places strict demands on hardware.
From image capture to recognition and output, end-to-end latency should be controlled within 100ms.
For control tasks such as traffic light switching, response time must be stable and predictable, without random fluctuations.
Tasks such as AI inference, video encoding and decoding, encryption, and decryption require dedicated hardware acceleration units instead of relying entirely on the CPU.
Edge nodes are widely distributed and cannot be maintained like data center servers by dedicated on-site teams. Once a failure occurs, maintenance costs can be high.
Key components should support redundancy. Power, storage, and network systems need fault protection mechanisms.
Out-of-band management should be supported. Even if the operating system crashes, maintenance engineers should still be able to remotely log in, diagnose, and recover the system.
When abnormalities are detected, the system should automatically restart or recover whenever possible to reduce manual intervention.
Edge nodes are deployed in open environments and face both physical and cybersecurity threats.
The chassis should include anti-tamper switches. Once the enclosure is opened, an alarm should be triggered immediately, and sensitive data can even be automatically erased when required.
TPM chips should be integrated to ensure a complete chain of trust from hardware and firmware to the operating system and applications.
Stored data and transmitted data should both be encrypted. Even if the device is stolen, data should not be leaked.
To meet different application requirements, edge computing nodes are available in various hardware forms.
Edge servers are suitable for edge data centers or large-scale edge nodes, such as regional aggregation nodes.
They usually adopt a standard rackmount design, but with a shorter depth than general-purpose servers, such as 450mm, to fit edge cabinets.
Typical configuration:
1–2 CPUs
64–128GB memory
2–4 drives
Expandable with 1–2 GPU cards
Power consumption controlled at 200–300W
Edge gateways are suitable for small edge nodes such as intersections and buildings.
They usually use a fanless design, full-metal enclosure, DIN-rail mounting, and compact structure.
Typical configuration:
Low-power CPU
8–16GB memory
Built-in storage
Multiple network and I/O interfaces
Power consumption controlled at 20–50W
AI boxes are suitable for scenarios that require local AI inference, such as deployment near cameras.
They usually integrate dedicated AI chips such as NPUs to provide powerful AI computing capability under very low power consumption.
Typical configuration:
1–10 TOPS AI computing power
Multi-channel real-time video stream analysis
Power consumption of 10–30W
Direct installation on camera poles
Outdoor integrated units are suitable for harsh outdoor environments such as roadsides and remote locations.
They use industrial-grade protection design, built-in temperature control systems, and can be directly exposed to natural environments.
Typical configuration:
Customized based on project requirements
Wide-temperature and wide-voltage support
IP65 or higher protection
Wall-mounted or pole-mounted installation
As a professional server OEM/ODM manufacturer, we have accumulated rich experience in edge computing hardware.
For a smart transportation project in one city, we customized an intersection edge node.
Project requirements:
Support real-time analysis of 8 HD video streams
End-to-end latency below 100ms
Operation from -20°C to 50°C
Support data retransmission after power failure
Our solution:
Industrial motherboard with wide-temperature design
Integrated high-performance GPU module
Total system power controlled within 150W
Dual power redundancy
4G / 5G backhaul support
The solution has been deployed at more than 300 intersections across the city and has operated stably for over one year.
For an environmental sanitation group, we customized a trash bin overflow detection terminal.
Project requirements:
Ultra-low power consumption below 5W
Solar power supply
4G backhaul support
Ability to detect trash overflow and fire risks
Our solution:
ARM-based low-power processor
Lightweight AI chip
Infrared camera
Multiple sensors
Total system power consumption of only 3.5W
Solar panel + battery power supply
The device can operate continuously for more than 7 days and has been deployed in more than 5,000 units.
For a technology park, we customized an AI access control terminal.
Project requirements:
Facial recognition
Mask detection
Temperature measurement
Response time below 300ms
Offline local operation
Our solution:
RK3588 chip
Integrated 6 TOPS NPU computing power
HD camera
Infrared temperature measurement module
Local storage for 100,000 face records
Real-time cloud synchronization
After deployment, park entry efficiency increased by 50%.
Related Recommendations
Learn more news and information
