According to a report by Electronic Enthusiasts (by Li Wanwan), in recent years, various edge devices equipped with AI computing chips have emerged following NVIDIA’s launch of the Jetson series. Due to their box-like appearance and algorithm inference capabilities, the industry often refers to them as edge boxes, AI Boxes, algorithm boxes, etc.
With the richness and diversity of algorithms, edge boxes have been widely applied in various scenarios such as chain stores, gas stations, chemical plants, construction sites, factories, power systems, telecom rooms, smart security, smart communities, campuses, scenic spots, and parks, rapidly supporting visual algorithms for detection, classification, segmentation, and feature extraction.
What Computing Chips Are Used with AI Edge Computing Boxes
AI edge computing boxes can typically be paired with various AI computing chips, depending on their design goals, application scenarios, and performance requirements. For instance, GPUs, although initially designed for graphics rendering, are now widely used for deep learning and other AI tasks. GPUs have a large number of parallel processing units, which can efficiently perform matrix operations, making them very useful for training and inference of neural networks.
For example, NPUs are specifically designed for executing neural network computations, particularly deep learning tasks. They usually have high parallelism, low power consumption, and optimized floating-point operation capabilities, making them very suitable for deploying and running complex AI models on edge computing devices.
There are also ASICs, FPGAs, and CPUs. ASICs are chips customized for specific applications, providing extremely high performance and energy efficiency for certain AI tasks. FPGAs are programmable chips that users can configure according to their needs. Although CPUs are not specifically designed for AI tasks, they remain key components in edge computing devices. Modern CPUs typically contain instruction sets and features that accelerate AI tasks.
Specifically, a certain AI edge computing box may be equipped with one or more of the aforementioned chips. For example, some boxes may use a combination of high-performance NPUs and GPUs to provide powerful AI computing capabilities.
Currently, many AI edge computing boxes on the market use NVIDIA’s Jetson series. Pairing AI edge computing boxes with NVIDIA’s Jetson series is a powerful combination that brings deep learning and computer vision capabilities directly to edge devices, enabling real-time data processing and analysis.
NVIDIA’s Jetson series includes various modules and platforms, such as Jetson Xavier NX, Jetson Orin NX, and Jetson AGX Orin, all of which feature high-performance GPUs and dedicated AI acceleration cores that can quickly process image and video data, providing strong computational support for edge computing.
For instance, Yanzhi Technology has developed the edge intelligent AI-BOX product series WES-WNX00-3220 based on the NVIDIA Jetson Xavier NX system. This product is suitable for high-performance computing and AI applications in embedded and edge systems, serving as the preferred platform for running multiple modern neural networks in parallel while processing high-resolution data from multiple sensors.
Technology provides the TW-T906 smart box based on Jetson AGX Orin, with an AI computing power of up to 200 TOPS, designed for low-speed autonomous driving, and can be widely used in scenarios such as unmanned delivery vehicles, unmanned sanitation vehicles, smart driving commercial vehicles, and unmanned cleaning boats.
Hua Jie Technology has launched an edge computing AI box based on the NVIDIA® Jetson™ Orin NX/Orin Nano core module. This edge computing AI box is aimed at industrial applications, featuring excellent computing performance, efficient passive cooling, and industrial standard design, making it suitable for widespread deployment in edge environments, applicable to various scenarios such as smart power.
Advantech’s AI edge computing box based on the NVIDIA Jetson series can provide intelligent solutions for industrial applications. Shenzhen Zhishida Intelligent Technology Co., Ltd. has developed an edge computing box based on the NVIDIA Jetson series, effectively meeting the intelligent edge computing power upgrade needs in various fields such as smart transportation, AMR, smart security, and intelligent manufacturing.
What Domestic AI Edge Box Computing Chips Are Available
Currently, many manufacturers are using domestic chips. For example, the edge computer T1206 is designed based on the Rockchip RK3588 processor. The built-in NPU supports INT4/INT8/INT16/FP16 mixed operations, has a complete set of peripheral interfaces, and boasts an ultra-long MTBF for stable operation, making it an ideal carrier for deploying AI computing for deep learning in autonomous machines such as robots, unmanned delivery vehicles, low-altitude defense, intelligent inspection, and smart buildings.
Dingchang Electronics has developed an edge box based on Rockchip RK3588, which has 6 TOPS of AI computing power and can be used as a commercial display face recognition host, streaming media server, industrial control visual computing, human recognition, object recognition, gas recognition, AI monitoring, video image filtering, intelligent logistics cargo recognition, and various AI algorithm application scenarios.
The company has also developed an edge computing box based on Rockchip RK3568, which has a built-in NPU with 1 TOPS of computing power and can be used for lightweight AI projects, suitable for smart commercial display advertising machines, data collection, video playback, edge AI computing, intelligent O2O retail terminals, industrial control hosts, and other industry fields.
Rockchip has long focused on the AIoT field, forming a series of SoC chip platforms for various computing powers and scenarios, and deeply laying out AI algorithms in vision, audio, and video, while efficiently supporting mainstream model architectures, meeting the demands for deploying small models at the edge and end, empowering various AIoT smart hardware products at the edge and end, and providing strong support for intelligent upgrades and digital transformation across various industries.
The RK3568/RK3588 chips can be used to develop edge computing boxes. These two chips have different NPU computing powers, with the RK3568 having 1 TOPS and the RK3588 having 6 TOPS. Therefore, for small AI computing or visual analysis projects, the RK3568 solution is generally used, while the RK3588 solution is suitable for medium to large AI computing visual analysis projects.
In addition, in 2023, Rockchip completed the research and development design work for the new generation of mid-to-high-end AIoT processor RK3576 and began tape-out. The RK3576 adopts advanced process design and is equipped with the company’s self-developed latest generation NPU with 6 TOPS computing power, supporting operators related to the Transformer model architecture, significantly improving AI computing efficiency and enabling efficient operation of various AI algorithms.
It is reported that Baidu’s latest released Pan Yu AI edge computing box product – Pan Yu – AIBOX-L01, is based on this new RK3576 chip. The Pan Yu AI edge computing box is equipped with over 80 algorithms adapted to real scenarios across industries, effectively meeting customers’ local model computing needs.
Some manufacturers are also using Cambrian’s AI chips. Fengchao Interactive’s official website has a C16 edge computing box, which is equipped with Cambrian’s purely domestic AI inference acceleration chip MLU220, with an INT8 equivalent computing power of up to 16 TOPS. This product features strong computing performance, large storage capacity, flexible configuration, small size, wide temperature support, strong environmental adaptability, and ease of maintenance and management. It integrates Fengchao Interactive’s self-developed N-PIPE+UNN visual inference platform, supporting common visual AI models on the market. This product supports flexible deployment in various fields and scenarios such as edge, smart parks, security, commerce, and transportation.
Hua Jie Technology has an edge computing box HE10C, which also uses the Cambrian MLU220 chip, but it also uses the Rockchip RK3568 chip. The whole machine has 32 TOPS INT8 computing power and supports mainstream deep learning frameworks such as TensorFlow, Caffe, and PyTorch. This device is aimed at industrial applications, featuring excellent computing performance, efficient passive cooling, and industrial standard design, making it suitable for widespread deployment in edge environments, applicable to scenarios such as smart power, smart factories, smart parks, smart security, and smart construction sites.
MLU220 is Cambrian’s first edge AI series product launched in 2019, and it is also Cambrian’s first AI chip aimed at the edge intelligent computing field. The Siyuan 220 chip is a SoC edge acceleration chip specifically designed for deep learning, using TSMC’s 16nm process, featuring high computing power, low power consumption, and rich I/O interfaces. It employs a series of innovative technologies in processor architecture from Cambrian, achieving a maximum of 32 TOPS (INT4) computing power while consuming only 10 watts.
Additionally, some edge computing boxes use chips from Sanneng, such as Hua Jie Technology, which previously developed the edge computing AI BOX based on the Sanneng BM1684 high-performance chip, featuring 17.6 TOPS INT8 computing power and supporting mainstream deep learning frameworks such as TensorFlow, Caffe, and PyTorch, aimed at industrial applications, with excellent computing performance, efficient passive cooling, and industrial standard design, suitable for widespread deployment in edge environments, applicable to scenarios such as smart power, smart factories, smart parks, smart security, and smart construction sites.
Moreover, there is Yingma Technology, which developed the edge computing box – IVP03X smart workstation based on Sanneng’s fourth-generation AI processor BM1684X, featuring high computing power, strong encoding and decoding capabilities, and ultra-low power consumption. Compared to the previous generation BM1684 series edge computing boxes, the overall performance has improved significantly. The Yingma IVP03X smart workstation can be widely used in AI intelligent video analysis and lightweight model training in fields such as smart cities, smart security, smart communities, and smart transportation.
Just as the realization of artificial intelligence relies on computing power and algorithms, AI edge computing boxes are also a combination of computing power and algorithms. At the computing power level, AI edge computing boxes depend on the support of AI computing chips, such as the NVIDIA Jetson series, Rockchip RK3568/RK3588 chips, Cambrian MLU220, and Sanneng BM1684/BM1684X mentioned above. Of course, besides these companies, many others can also provide such chip support, including Cloud Sky Li Fei and Kun Yun Technology.
Disclaimer: This article is original from Electronic Enthusiasts, please indicate the source above when reprinting. If you want to join the group for communication, please add WeChat elecfans999, for submission of interviews, please send an email to [email protected].
More Hot Articles to Read
-
Performance Divergence Among 6 Chip Manufacturers! GPU, CPU Shift to High-End, AI Chips Accelerate Penetration into Large Model Industry
-
36 Semiconductor Companies Terminate IPOs! Inquiry Passes Difficulties, Simulation Becomes “Heavy Disaster Area”
-
From the Beijing Auto Show, Innovations in Automotive Architecture, Supercharging, and Smart Driving Solutions by Major Manufacturers
-
iPhone Sales Plummet, Vision Pro Cuts Orders, Can Apple Rely on AI for Rescue?
-
Price Cuts to Grab Market! Many Semiconductor Companies Face “Negative Profit Margin,” Affecting Packaging, RF, and Discrete Devices
Click to Follow and Mark Us
Mark us as a star to not miss any updates!
If you like it, give a “look” reward!
