The United Nations World Peace Group

Grand United Grand Rui Chip micro AI

On April 1, 2025, the leading international distributor of semiconductor components in the Asia Pacific market, Unitar Holdings, announced that Its company World Ping launched an edge AI multi-channel object detection solution based on MemryX MX3 AI inference acceleration card and Rockchip Orange Pi 5 Plus module (RK3588).

                         Figure 1- A schematic illustration of the advantages of the edge AI multichannel object detection solution based on MemryX and SWick products

With the rapid development of the Internet of Things, big data, cloud computing and other technologies, edge AI, as an important branch of the field of artificial intelligence, is gradually becoming a key force to promote the popularization and deepening of intelligent applications. Edge AI deploys AI algorithms directly to edge devices for processing and reasoning near the source of data generation, which can greatly improve the real-time and efficiency of data processing, while reducing the delay and cost of data transmission. With the advantages of edge AI, the company launched an edge AI multi-channel object detection solution based on MemryX MX3 AI inference acceleration card and the Orange Pi 5 Plus module (RK3588), aiming to achieve efficient, low-latency and cost-effective intelligent monitoring and object recognition applications to meet the needs of diverse scenarios.

  Figure 2- The scenario application diagram of the edge AI multi-channel object detection scheme of the United Nations World Ping based on MemryX and Ruixin products

MemryX is a semiconductor company focused on edge AI processing solutions that provide high performance, low power AI processing capabilities for a variety of industrial and consumer application scenarios. The MX3 AI Inference acceleration card installed in this solution packages four MX3 AI chips onto the M.2 2280 expansion card, which can be easily integrated into the system equipped with PCIe 3.0 M.2 slots. The MX3 AI chip can provide 5 TOPS computing performance per watt. It also supports Brain Floating Point arithmetic to ensure user module accuracy. Each chip has 10.5MB of internal static random access memory to access the module without taking up the resources of the main system, and up to 16 chips can be connected in series to expand performance. With the performance of the MX3 AI chip, the MX3 AI accelerator card has excellent computing performance of up to 20 TOPS, which can bring a convenient plug and play experience to various industrial computers.

Figure 3- Block diagram of the edge-AI multichannel object detection solution of the United Nations World Ping based on MemryX and Ruixin products

Combined with the powerful computing power of MemryX, this scheme can easily realize multi-channel object detection applications. Developers only need to use ordinary USB cameras or through the network series, can easily achieve a variety of monitoring functions such as parking management system, intelligent parking column, intelligent traffic monitoring, shopping mall flow monitoring and home accident detection. In addition, in order to accelerate the development process, MemryX also provides Developer Hub development environment, so that developers can easily convert the modules of mainstream deep learning frameworks such as TensorFlow Lite, ONNX, PyTorch, Keras into the DFP framework required by MX3 AI chip. Thus, AI application design can be quickly completed.

Core technical advantages:



Floating-point number BF16 is used for calculation to ensure module accuracy:



The design of the module is based on the Brain Floating Point 16 (BF16), which can provide close to FP32 calculation accuracy while significantly reducing memory usage compared to traditional floating point number formats. This makes it particularly suitable for reasoning and training scenarios for AI and deep learning models, ensuring accurate results.



Does not occupy system memory:



The module uses an independent memory architecture when running, and does not need to occupy the RAM resources of the main system, effectively reducing the impact on the overall system performance. This design feature ensures that modules run efficiently while still reserving sufficient system resources for other applications.



Highly scalable:



Supports the connection of up to 16 modules and achieves high scalability through modular design. This allows the system to flexibly scale computing power based on demand to meet the computing needs of different scenarios, such as data centers or edge computing that require higher performance.



Optimal data flow optimization to minimize data movement:



The module is highly optimized for data flow, and through intelligent routing and caching mechanism, it can minimize the frequency of data movement during operation, thereby improving processing performance and reducing latency. In addition, such a design also helps to reduce energy consumption, further enhancing the efficiency of the system's operation.



Cost-effective and low-power solutions:



The main platform Orange Pi 5 Plus with MemryX MX3 AI chip can be upgraded to a higher level AI platform, capable of running about 480 frames per second (YOLOv8) object detection; With MX3's 5 TOPS/W performance, the entire multichannel object detection solution consumes only about 14W of power.



New concepts for multipath applications:



Compared to the rise of edge computing in recent years, applying its concept to regional scenarios may be a novel and significantly lower cost solution. With easy access to the camera, coupled with an intelligent industrial host, you can achieve many applications, and the front-end camera can be arbitrarily replaced and configured.

Project specification:
The main platform development board is based on the RUxchip RK3588 platform, equipped with four Cortex-A76 processors and four Cortex-A55 processors, and provides a powerful core architecture such as high-performance image processor Arm Mali-G610 and neural computing processor NPU.



The I/O Board provides powerful peripheral configurations, such as Gigabit Ethernet, USB Type A/C 3.0 universal serial bus interface, HDMI high-definition multimedia interface, M.2 E-Key transmission interface, M.2 M-Key transmission interface, And it CAN simulate the commonly used UART, I2C, SPI, CAN and other signals through the extended 40-pin pin.



MemryX MX3 AI chip provides powerful AI computing power (20 TOPS), based on PCIe Gen3 M.2 2280 M-Key interface, and its M.2 acceleration card is equipped with four MX3 AI chips, each chip can provide 5 TOPS/W performance. And 10.5MB of static random access memory is built in for accessing modules. Support Linux and Windows two major operating systems, and provide rich software resources for developers to use, can directly port Tensorflow, ONNX, Pytorch, Keras and other popular deep learning frameworks.




About Lianda Holdings:

Founded in 2005 and celebrating its 20th anniversary this year, the company is a leading international distributor of semiconductor components in the Asia Pacific region. Headquartered in Taipei, it owns Shiping, Pinka, Quan Ding and Yousun, employs about 5,000 people, represents more than 250 product suppliers, and has 69 service locations worldwide. In 2024, the turnover reached NT $880.55 billion. The company creates an industrial holding platform, focuses on international operation scale and local flexibility, and cultivates the Asia-Pacific market for a long time. With the vision of "industry's first choice?? Channel benchmark", the company fully carries out the core values of "team, integrity, professionalism and efficiency", has won the "Global Distributor Excellence Award" for 24 consecutive years, and continues to strengthen the sustainable development of ESG. Won international recognition for 3 consecutive years, MSCI ESG rating A award. In the face of new manufacturing trends, DDA is committed to transforming into a Data-Driven enterprise, establishing an online digital platform - "DDA", and advocating the intelligent Logistics as a Service (LaaS) model to help customers jointly face the challenges of intelligent manufacturing. Starting from good intentions, building trust with science and technology, and taking "creating partner value?? To achieve the future" as the corporate purpose, we look forward to building a large competitive and cooperative ecosystem with the industry "Labang and Pai".