Which computing chips are matched with AI edge computing box

Which computing chips are matched with AI edge computing box

AI edge computing boxes can usually be matched with a variety of AI computing chips, depending on their design goals, application scenarios and performance requirements. GPUs, although originally designed for graphic rendering, are now widely used in deep learning and other AI tasks. GPU has a large number of parallel processing units, which can efficiently perform matrix operations, which is very useful for neural network training and inference.
Like NPU, NPU is specifically designed to perform neural network calculations, especially deep learning tasks. They usually have high parallelism, low power consumption and optimized floating point computing capabilities, and are very suitable for deploying and running complex AI models on edge computing devices.
There are also ASICs, FPGAs, CPUs, etc. ASICs are chips customized for specific applications. For certain specific AI tasks, ASICs can provide extremely high performance and energy efficiency. FPGA is a programmable chip that allows users to configure its logic functions as needed. Although CPUs are not specifically designed for AI tasks, they are still key components in edge computing devices. Modern CPUs typically contain instruction sets and features for accelerating AI tasks.
Specifically, an AI edge computing box may carry one or more of the above chips. For example, some boxes may use a combination of high-performance NPUs and GPUs to provide powerful AI computing power.
At present, many AI edge computing boxes on the market use NVIDIA Jetson series. The AI edge computing box and Nvidia's Jetson series are a powerful combination, which can directly bring the ability of deep learning and computer vision to edge devices to achieve real-time data processing and analysis.
Nvidia's Jetson series includes a variety of modules and platforms, such as Jetson Xavier NX, Jetson Orin NX, Jetson AGX Orin, etc. They all have high-performance GPUs and dedicated AI acceleration cores that can quickly process image and video data, providing strong computing support for edge computing.
For example, Yanzhi Technology has developed the WES-WNX00-3220 series of edge intelligent AI-BOX products based on the Nvidia Jetson Xavier NX system. This product is suitable for high-performance computing and AI applications in embedded and edge systems, and is the preferred platform for running multiple modern neural networks in parallel and processing high-resolution data from multiple sensors simultaneously.
The picture shows a TW-T906 smart box based on Jetson AGX Orin technology, with an AI computing power of up to 200TOPS, specifically designed for low-speed autonomous driving. It can be widely used in scenarios such as unmanned delivery vehicles, unmanned sanitation vehicles, intelligent driving business vehicles, and unmanned cleaning ships.
Nvidia Jetson Orin edge computing AI box launched by Huajie Technology is based on NVIDIA ® Jetson ™ Orin NX/Orin Nano core module products. This edge computing AI box is oriented to industrial applications. It has excellent computing performance, efficient passive cooling, industrial standard design and other characteristics. It can be widely deployed in edge environments, and is suitable for multiple scenarios such as smart power.
Advantech, based on the AI edge computing box of Nvidia Jetson series, can provide intelligent solutions for industrial applications. Shenzhen Zhishida Intelligent Technology Co., Ltd., based on Nvidia Jetson series of edge computing boxes, can effectively meet the needs of intelligent edge computing upgrading in multiple fields such as intelligent transportation, AMR, intelligent security, and intelligent manufacturing.

What are the computing power chips in Chinese AI edge boxes

At present, many manufacturers also use domestic chips. For example, the edge computer T1206 shown in the figure is designed based on the Ruixin micro RK3588 processor chip. The NPU embedded in this product supports INT4/INT8/INT16/FP16 hybrid computing, with a complete peripheral interface and ultra long MTBF stable operation ability. It can be applied to autonomous machines such as robots, unmanned delivery vehicles, low altitude defense, intelligent inspection, and smart buildings, making it an ideal carrier for deploying AI computing power at the edge for deep learning.
Dingchang Electronics has developed an edge box based on Ruixin Micro RK3588, which has 6TOPS AI computing power and can be used as a commercial facial recognition host, streaming media server, industrial visual computing, human shape recognition, object recognition, vehicle recognition, AI monitoring, video image filtering, intelligent logistics product recognition and other AI algorithm application scenarios.
The company also developed a edge computing box based on Ruixin Micro RK3568. The embedded NPU has 1TOPS computing power and can be used as a lightweight AI project, which is suitable for smart business display advertising machines, data acquisition, video playback, edge AI computing, smart O2O retail terminals, industrial control hosts and other industries.
Ruixin Micro has been deeply involved in the AIoT field for a long time, forming AIoT SoC series chip platforms with various computing power and scenarios, and deeply laying out AI algorithms such as vision, audio, and video. At the same time, it efficiently supports mainstream model architectures, meets the needs of small models deployed on the edge and end sides, empowers various AIoT intelligent hardware products on the edge and end sides, and provides strong support for the intelligent upgrading and digital transformation of Baixing Baiye.
Its RK3568/RK3588 chips can be used to develop edge computing boxes. The NPU computing power of these two chips is different. RK3568 has 1TOPS computing power, and RK3588 has 6TOPS computing power. Therefore, small-scale AI computing or visual analysis projects generally use the RK3568 scheme, while medium to large-scale AI computing and visual analysis projects can use the RK3588 scheme.
In addition, in 2023, Ruixin Microelectronics completed the research and design work of the new generation mid to high end AIoT processor RK3576 and produced chips. RK3576 adopts advanced process design and is equipped with the latest generation NPU of the company's self-developed 6TOPs computing power. It supports Transformer model architecture related operators, significantly improving artificial intelligence computing efficiency and supporting efficient operation of various AI algorithms.
It is said that Baidu's latest Panyu AI edge computing box product, Panyu AIBOX-L01, is based on the new RK3576 chip of Ruixin Micro. Panyu AI edge computing box has more than 80 kinds of algorithm adaptation for cross industry real scenarios, which can effectively meet customers' local model computing needs.
Some manufacturers also use Cambrian AI chips. On the official website of Fengchao Interactive, there is a C16 edge computing box of its cooperative brand, which is equipped with Cambrian pure domestic AI reasoning accelerator chip MLU220. The equivalent computing power of INT8 is up to 16 Tops. This product has the characteristics of strong computing performance, large capacity storage, flexible configuration, small size, wide temperature range support, strong environmental adaptability, and easy maintenance and management. The N-PIPE+UNN visual inference platform, independently developed by integrating trend interaction, supports common visual AI models on the market. This product supports flexible deployment in various fields and scenarios, such as edges, smart parks, security, commerce, and transportation.
Huajie Technology has a edge computing box HE10C, which also uses the Cambrian MLU220 chip, but it also uses the Ruixin micro RK3568 chip. The entire machine has 32 TOPS INT8 computing power and supports mainstream deep learning frameworks such as TensorFlow, Caffe, and PyTorch. The device is designed for industrial applications and has excellent computing performance, efficient passive cooling, and industrial standard design. It can be widely deployed in edge environments and is suitable for scenarios such as smart power, smart factories, smart parks, smart security, and smart construction sites.
MLU220 is an edge AI series product launched by Cambrian in 2019, and it is also the first AI chip of Cambrian aimed at edge intelligent computing. The Siyuan 220 chip is a SOC edge acceleration chip specifically designed for deep learning, using TSMC 16nm technology. It has high computing power, low power consumption, and rich I/O interfaces. It adopts a series of innovative technologies from the Cambrian era in the field of processor architecture. Its architecture is the latest generation of Cambrian intelligent processor MLUv02, achieving a maximum computing power of 32 TOPS (INT4) with a power consumption of only 10 watts.
In addition, some edge computing boxes use energy computing chips, such as Huajie Technology, which previously developed edge computing AI BOX based on energy computing BM1684 high-performance chip. The whole machine has 17.6 TOPS INT8 computing power, supports TensorFlow, Caffe, PyTorch and other mainstream in-depth learning frameworks, and is oriented to industrial applications. It has excellent computing performance, efficient passive cooling, industrial standard design and other characteristics. It can be widely deployed in edge environments, and is suitable for smart power, smart factories, smart parks, smart security, smart construction sites and other scenarios.
In addition, there is British Code Technology. The edge computing box IVP03X intelligent workstation developed by the company based on the fourth generation AI processor BM1684X is characterized by high computing power, strong encoding and decoding capability, and ultra-low power consumption. Compared with the edge computing box of the previous generation BM1684 series, the overall performance has been greatly improved. The English code IVP03X intelligent workstation can be widely used for AI intelligent video analysis and lightweight model training in fields such as smart cities, smart security, smart communities, and smart transportation.

Write at the end

Just as AI cannot be realized without computing power and algorithms, AI edge computing box is also a collection of computing power and algorithms. In terms of computing power, AI edge computing box can not be separated from the support of AI computing power chip, such as Nvidia's Jetson series, Ruixin Micro's RK3568/RK3588 chip, Cambrian MLU220, and Computable BM1684/BM1684X. Of course, in addition to these enterprises, there are many others who can provide such chip support, including Yuntian Lifei, Kunyun Technology, and so on.