With the rapid development of artificial intelligence technology, AI is profoundly changing all aspects of human life and work at an unprecedented speed. From smartphones to self-driving cars on the streets, from smart brains in bustling cities to intelligent robots on the factory floors, the tentacles of AI have extended to every corner of society. However, in the context of the rise of the Internet of Things and edge computing, AI chip designs also face many unprecedented challenges: they need to process massive amounts of data in real time within extremely demanding power budgets and chip size constraints, they need to adapt to harsh working environments that are much lower than those of data centers, they must have extremely high reliability and security, and they must flexibly adapt to rapidly changing application needs.
It is in this era that the industry urgently needs a highly integrated, end-to-end chip acceleration solution that can efficiently realize the whole process of data preprocessing, AI inference, and post-processing at the edge. As a pioneer in heterogeneous computing, AMD is launching the second generation of the Versal Adaptive System-on-Chip (SoC) portfolio, which aims to accelerate the entire data processing process on a single chip, achieving industry-leading performance and power efficiency under demanding power and size constraints. According to Salil Raje, senior vice president and general manager of AMD's Adaptive and Embedded Computing Group, the second-generation Versal family of products builds on AMD's more than 40 years of experience in adaptive computing and integrates multiple key compute engines on a single architecture to deliver superior computing efficiency, performance and flexible scalability in a wide range of application scenarios, from entry-level to flagship.
In terms of specific product planning, the second-generation Versal series mainly includes the Versal AI Edge series for AI edge computing and the Versal Prime series for traditional embedded fields. It is worth mentioning that the Versal AI Edge series adopts an innovative heterogeneous architecture that integrates a variety of complementary key processing units to fully accelerate the end-to-end data processing process of AI systems: a high-performance FPGA array for real-time processing of multi-source sensor data, providing flexible data pre-processing and feature extraction capabilities, a newly upgraded AI Edge array specializing in efficient inference computing of neural networks, with three times the performance per watt of the previous generation, and high-performance Arm CPU clusters provide powerful post-processing capabilities for safety-critical tasks such as decision-making and control. As Steph Gauthier, Senior Product Line Manager, Versal AI Edge, Adaptive & Embedded Computing Group, AMD, said, "This single-chip smart solution eliminates the drawbacks of traditional multi-chip processing architectures, enabling smaller, more energy-efficient system sizes and significantly reducing end-to-end latency through seamless data flow within the chip, providing tremendous value for real-time critical use cases." "
Steph Gauthier, Senior Product Line Manager, Versal AI Edge, Adaptive & Embedded Computing Group, AMD
At the same time, the second-generation Versal Prime series for the general embedded field focuses on combining high-performance scalar computing capabilities with flexible programmable logic, focusing on video processing, industrial control, software defined radio and other occasions that require extremely high computing power and adaptability. Compared to the previous generation, the series delivers up to 10x faster CPU performance, while the programmable logic architecture inherited from Xilinx provides sufficient support for different interfaces and acceleration needs. With a new hardware IP module optimized for 8K multi-channel video processing, the second generation of the Versal Prime series will be ideal for ultra-high-definition video encoding, decoding, and streaming.
Of course, the improvement of the software ecosystem and development tool chain is crucial to the performance of the chip. Manuel Uhm, Director of Product Marketing for Versal, Adaptive & Embedded Computing, AMD, said, "To take full advantage of the superior performance of the second-generation Versal family, the entire technology stack must be deeply optimized end-to-end. In response, we are introducing a new Versal software development environment, including specially optimized compilers, library functions, advanced synthesis tools, and more, which enable developers to deploy a wide range of industry applications on our silicon platform with unprecedented efficiency. At the same time, we have launched a series of reference designs and development boards for mainstream use cases to help customers significantly shorten the transition time from proof of concept to mass production. "
Manuel Uhm, Director of Product Marketing, Versal, Adaptive & Embedded Computing Group (AECG), AMD
In fact, the reason why AMD has been able to launch such a rich and diverse and highly targeted embedded chip product line is because of its deep technical accumulation in the field of heterogeneous computing. Throughout AMD's more than half-century history, innovation and openness have always been at the heart of AMD. From general-purpose computing, graphics computing, to adaptive computing, AMD has always explored the convergence of multiple computing paradigms with an open mind. The Zynq adaptive SoC, the predecessor of the Versal series, is the essence of this heterogeneous concept, and the newly released second-generation Versal series is the best of all worlds, combining AMD's advanced technologies in CPU, GPU, and interconnect architectures with Xilinx's programmable logic advantages accumulated over many years, to help the leapfrog development of embedded AI and accelerate the realization of the vision of ubiquitous intelligence.
Looking ahead, Manuel has high hopes for the second-generation Versal series: "With the accelerated evolution of emerging technologies such as 5G and AIoT, AMD has built a comprehensive heterogeneous computing product matrix, from cloud big data centers, to small cells at the edge of the network, to smart nodes on the device side, to meet the individual needs of various smart applications at different deployment stages. In the future, we will continue to uphold the development concept of openness, cooperation and innovation, work with industry partners to deepen software and hardware collaboration, accelerate the deep integration of artificial intelligence and the real economy, and jointly create a better future full of intelligence and fully connected." "
This may be a rare opportunity for developers and industry users. AMD has launched an early access program for the second-generation Versal series for key customers, and the relevant technical documentation and development tools are now available, allowing developers to apply for the first-generation Versal evaluation kit to start developing innovative applications in advance. According to AMD's official news, engineering samples of the second-generation Versal series chips are expected to be available in the first half of 2025 at the earliest, when supporting evaluation boards and system-on-modules will also be launched, and large-scale mass production is also expected to be achieved by the end of the year. Based on a new starting point, start a new journey. Looking to the future, the second-generation Versal series is expected to shine on a broader application stage, helping developers spread the "wisdom" of chips to thousands of industries and draw a new blueprint for the development of artificial intelligence.
(8661288)