The RK3588’s native computing power is insufficient—does multi‑channel AI vision mean reducing the number of cameras? The XC3588E paired with a compute‑acceleration card offers a new solution.
When working on edge AI projects, have you ever encountered such a dilemma:
In the early stages, we selected the RK3588 platform and, leveraging its native 6 TOPS NPU performance, successfully completed prototype development. However, once the project entered mass production and model iterations began, a computational bottleneck suddenly emerged: with multiple IPC streams and industrial cameras simultaneously performing object detection and defect identification, frame rates plummeted and inference latency soared.
To ensure reliable recognition performance, the only option is to reduce the number of camera feeds connected. If you don’t cut back on the number of streams, you’re left with two choices: migrate to an expensive overseas computing platform, or upload video streams to the cloud for AI analysis. However, cloud‑based solutions also have significant drawbacks: they incur ongoing, high bandwidth costs; network instability can cause recognition to fail; and a complete network outage will bring the entire AI system to a standstill. So, when computing power falls short, must we really choose between reducing the number of channels or scrapping the entire hardware setup and starting over?
Xiangcheng Electronics XC3588E Smart Motherboard + DEEPX Computing Acceleration Card , bring Elastic Heterogeneous Computing Capacity Expansion Solution , All components use brand-new, original chips. There is no need to replace the existing main control board; the baseboard, firmware, and application code can be reused to the greatest extent possible.

The XC3588E is based on the RK3588 SoC and features a built-in 6 TOPS NPU, enabling it to independently handle lightweight AI‑based recognition tasks. The motherboard includes预留 M.2 expansion interface , can directly connect to compute acceleration cards ( 25 TOPS of NPU computing power) , After aggregation, the overall heterogeneous computing power of the system is significantly enhanced. Lightweight inference is handled by the motherboard’s native NPU, while computationally intensive tasks such as multi‑channel video analytics, high‑precision quality inspection, and on‑device large‑model inference are offloaded to accelerator cards. This clear division of labor and parallel processing effectively addresses the challenge of concurrent visual‑stream processing causing lag.

The greatest advantage of this solution lies in… On-demand computing power upgrades for the project. In the early stages, the product’s functionality is straightforward, allowing delivery with just the XC3588E motherboard. As requirements evolve and AI models are iterated, there’s no need to redesign the PCB or replace the entire hardware platform—simply add a compute‑acceleration card to scale up computational power. This significantly shortens the secondary development cycle and helps control R&D and inventory costs.
At the hardware level, the XC3588E serves as an industrial-grade smart motherboard, Uses a brand-new, genuine chip, It features industrial-grade wide-temperature operation and can run stably 24/7 without interruption; it is equipped with a rich array of industrial interfaces, including six serial ports, 9-way USB, dual CAN interfaces, with HDMI IN/OUT, dual Gigabit Ethernet, GPIO, and more enable direct connectivity with industrial cameras, sensors, and display terminals, supporting seamless integration with a wide range of field‑deployed devices. On the software side, it natively supports Ubuntu, Android, Galaxy Kylin, and other operating systems, facilitating easy model migration, while its domestically developed hardware platform meets the stringent requirements for independent control across numerous industries.

Typical Deployment Scenarios
Industrial Vision: Multi-camera defect detection on the production line, product appearance quality inspection
Smart Security/Industrial Park: Multi-camera human detection, fire and smoke detection, and intrusion detection
Energy and Power: Substation and Photovoltaic Plant Inspection Image Analysis
AI Robot: Multimodal visual recognition, on-site environmental perception

Many edge AI projects are held back by the twin challenges of high costs for scaling compute power and lengthy board‑revision cycles. The XC3588E+ compute‑acceleration card breaks away from the conventional mindset of “replace the board when compute falls short,” offering a pluggable, elastic compute architecture that enables multi‑channel, real‑time AI inference on‑premises. By keeping data at the edge node, it delivers low latency and offline‑capable operation, balancing performance, cost efficiency, and data security.
Recommended
From data center infrastructure monitoring to intelligent manufacturing, from energy power plants to AI‑driven security, in the era of domestically developed edge computing, Xiangcheng Electronics’ industrial control host series provides a stable and reliable hardware foundation, helping industries across the board achieve digital and intelligent transformation.
The XC3576H industrial control motherboard has now been fully adapted to Ubuntu 26.04 LTS, offering developers a high-performance, long‑support‑cycle, and feature‑rich ARM‑based open‑source solution for industrial applications.
The IPCA_3588E industrial control host features a self-developed XC3588E industrial motherboard and is built with exclusively new, original‑brand chips, delivering an integrated edge computing terminal that combines domestically produced chips with a domestic operating system.