Edge AI chips are modules specifically designed to handle a large number of computing tasks in artificial intelligence applications in edge computing environments. It is usually integrated into edge devices, enabling these devices to perform real-time data processing, analysis, and decision-making locally without transmitting data to the cloud for processing.
The emergence of edge AI chips has greatly improved the intelligence level of edge devices, making various application scenarios more efficient, flexible, and secure. At present, there are several well-known edge AI chip suppliers on the market, such as NVIDIA, AMD, Qualcomm, Intel, etc. These companies continue to launch new edge AI chip products to meet the needs of different application scenarios.
NVIDIA
NVIDIA has a series of important products and layouts in the field of edge AI chips. Among them, the Jetson series is the representative of NVIDIA's edge AI chips. The Jetson platform is a series of modules and development tools designed by NVIDIA for edge computing, covering the main application scenarios of edge AI such as robots, autonomous driving, industrial manufacturing, and smart cities. These products combine NVIDIA's GPU technology and deep learning technology to provide powerful computing power and AI reasoning capabilities for edge devices.
The Jetson family of products includes modules of different models and specifications, such as Jetson Nano, Jetson TX2, Jetson AGX Xavier, Jetson Orin, etc., which have different performance characteristics and applicable scenarios.
The Jetson Nano module is a compact AI computer with superior performance and power consumption to meet the needs of running modern AI workloads, running multiple neural networks in parallel, and processing data from multiple high-resolution sensors simultaneously. This makes it an ideal entry-level choice for adding advanced AI to embedded products.
The expanded Jetson TX2 embedded module series provides up to 2.5 times the performance of Jetson Nano, while consuming only 7.5 watts of power. Jetson TX2 NX is pin- and form-factor compatible with Jetson Nano, while Jetson TX2, TX2 4GB and TX2i are the same form factor as the original Jetson TX2. Jetson TX2i is considered ideal for building devices including industrial robots and medical equipment.
Jetson AGX Xavier is a computer built specifically for autonomous machines. This compact, energy-efficient module provides hardware acceleration and high-speed I/O performance for the entire AI pipeline, allowing customers to apply new AI applications to the edge. Jetson AGX Xavier is suitable for complex applications that require real-time processing and advanced AI capabilities, such as autonomous driving and smart manufacturing.
The Jetson Orin module has a computing power of up to 275 trillion floating-point operations per second (TOPS), which is 8 times the performance of the previous generation. It is suitable for multiple concurrent AI reasoning pipelines, and it can also support multiple sensors through high-speed interfaces. This makes Jetson Orin an ideal solution for a new era of robot development.
These Jetson modules are usually equipped with NVIDIA's GPU and deep learning processors, supporting a variety of deep learning frameworks and algorithms, enabling developers to build and deploy efficient AI applications on edge devices.
Intel
Intel has made significant research and progress in the field of edge AI chips. Recently, Intel and its subsidiary Altera announced the launch of new edge-optimized processors, FPGAs, and market-ready programmable solutions at the Embedded World, dedicated to extending powerful AI capabilities to edge computing. These products will power AI edge devices for retail, healthcare, industry, automotive and other industries.
Intel's edge-optimized processors include Intel® Core™ Ultra, Intel® Core™ and Intel Atom® processor series, as well as Intel Sharp™ discrete graphics cards (GPUs). These processors not only provide powerful image classification inference performance, but also provide higher performance and flexibility for edge AI devices.
From Intel's official website, we can also see that Intel's products for edge AI include: Intel® Xeon® processors, Intel® Core™ Ultra and Intel® Core™ processors, Intel Atom® processors, etc. These processors are an important part of Intel's edge AI product portfolio, mainly targeting key areas such as retail, industrial manufacturing and medical.
For example, Intel® Core™ Ultra processors provide excellent performance centered on graphics processing and AI through high-efficiency edge processors in BGA packaging; such as the fifth-generation Intel® Xeon® Scalable processors for the edge, which use built-in accelerators, hardware-level security and power optimization to improve the energy efficiency of edge workloads, and can improve the performance-to-power ratio for applications in AI, edge, healthcare, industry, retail and energy.
AMD
AMD also has important layout and development in the field of edge AI chips. In the past few months of this year, it has continuously launched new products. In February of this year, AMD announced the launch of AMD Embedded+. This new architecture solution combines AMD Ryzen™ embedded processors and AMD Versal™ adaptive SoCs on a single integrated board. Combining embedded processors with adaptive SoCs can accelerate the listing process of edge AI applications.
In early April this year, AMD officially released the Ryzen Embedded 8000 series processors, which are the first embedded products equipped with AMD XDNA architecture NPU, optimized for workloads of industrial AI applications. AMD Ryzen 8000 series embedded processors use 4nm process, AMD "Zen 4" CPU architecture, RDNA 3 architecture GPU and XDNA architecture NPU, with AI computing power of 39 TOPS, of which NPU provides 16 TOPS.
Recently, AMD announced the expansion of the AMD Versal™ adaptive system-on-chip (SoC) product portfolio, launching the new second-generation Versal AI Edge series and the second-generation Versal Prime series adaptive SoC, which integrates pre-processing, AI reasoning and post-processing in a single device, and can provide end-to-end acceleration for AI-driven embedded systems.
The second-generation Versal series devices balance performance, power consumption, board area, and advanced functional safety and information security. The new functions and features they provide support the design of high-performance edge-optimized products for the automotive, industrial, vision, medical, broadcast and professional audio and video markets.
Qualcomm
Qualcomm also has important layouts and contributions in the field of edge AI chips. Its edge AI chips combine high performance, low power consumption and optimized AI processing capabilities to provide strong support for various edge computing application scenarios.
Qualcomm has launched a number of products in the field of edge AI chips, such as QCS8550 and QCM8550. These two processors are designed for performance-intensive IoT applications and integrate powerful computing power, edge AI processing, Wi-Fi 7 connectivity, and enhanced graphics and video capabilities. They support and rapidly deploy IoT applications with high performance requirements, such as autonomous mobile robots and industrial drones.
For example, Cloud AI 100 is a chip for edge inference that can perform more than one trillion operations per second. Cloud AI 100 began production in the second half of 2020, providing powerful AI reasoning capabilities for edge computing devices.
In addition to the above products, Qualcomm continues to launch new edge AI chips to meet the development needs of the market and technology. These chips have significant advantages in power consumption, performance, security, and programmability, enabling them to play an outstanding role in various edge computing application scenarios.
Written at the end
The competition in the edge AI chip market is very fierce. In addition to AMD and NVIDIA, many other companies are also actively investing in the research and development and promotion of edge AI chips. Therefore, when choosing an edge AI chip, you need to consider multiple factors such as performance, power consumption, cost, and technical support to choose the chip that best suits your application needs.
