AI is everywhere.
Among them, based on the advantages of low traffic consumption, low delay and strong privacy, edge AI has been widely used in transportation, medical, intelligent retail, intelligent factory, smart city and other industries.
On April 9, AMD announced the expansion of the AMD Versal ™adaptive system-on-a-chip (SoC) portfolio with a new second-generation Versal AI Edge series and second-generation Versal Prime series adaptive SoC to provide end-to-end acceleration for AI-driven embedded systems According to reports, the first devices in the second generation Versal series portfolio are built on the basis of the first generation, with a powerful new AI engine, and the TOPS per watt is expected to increase by up to 3 times compared with the original Versal AI Edge series devices.
at the same time, the new high-performance integrated Arm ®CPU is expected to provide up to 10 times more scalar computing power than the first generation Versal AI Edge and Versal Prime series devices.
AMD, and the second generation of Versal series devices integrate preprocessing, AI reasoning and post-processing into a single device, providing end-to-end acceleration for AI-driven embedded systems.
In addition, it balances performance, power consumption, board area, and advanced functional security and information security.
the new features and features support the design of high-performance edge-optimized products for the automotive, industrial, visual, medical, broadcast and professional audio and video markets.
Subaru has become one of the first customers to announce plans to deploy AMD’s second-generation Versal AI Edge series to support the next-generation EyeSight ADAS vision system.
To bring single-chip intelligence to embedded systems, in the past decade, data has experienced “savage” growth, application operation and massive data processing based on the perception of the Internet of things, superimposed intelligent electric vehicles to a central integrated electronic and electrical architecture, traditional cloud computing is difficult to fully meet the high requirements of millisecond response time, security and so on.
As a result, edge computing arises at the historic moment and becomes more and more important.
Of course, high requirements mean larger workloads, and the current edge AI also faces more limitations, including power consumption, size, computing, security, and reliability.
Generally speaking, embedded systems driven by artificial intelligence must first preprocess the data, including sensor fusion and data intersection, then infer it through the vector processor, and finally use the high-performance embedded CPU for post-processing, so that a decision can be made through the inferred results.
AMD,AMD Adaptive and embedded Computing Division (AECG) Versal Product Marketing Director Manuel Uhm stressed that in the pre-processing stage, real-time processing must be achieved, so that the real-time processing of the whole system can be achieved.
And this requires programmable logic.
Because programmable logic can be very flexible to adapt to both sensors and various types of interfaces to achieve true real-time, while ensuring low latency, certainty, and even after field deployment.
Can also be upgraded.
” He further said that in the real-time AI-driven embedded system, the computing power of the three steps must be accelerated in order to truly realize the real-time performance of the whole system.
However, Manuel Uhm pointed out that there is no type of processor on the market that can be optimized directly for three stages, and most of them use multi-chip solutions, but this will bring a series of problems and limitations to the whole embedded system.
For example, higher power demand and power supply complexity, higher board area and end system size, higher external memory requirements and increased delay caused by inter-chip convexity, more security vulnerabilities and scrapping challenges, and inefficiency.
This is why AMD launched the second generation of Versal adaptive SoC-a function that can only be achieved by the superimposed operation of multiple devices, which can be solved by the intelligence of a single chip.
At the same time, realize the supplement to the first generation products.
Source: AMD, the single-chip intelligence of the second-generation Versal AI Edge series and the second-generation Versal Prime series adaptive SoC eliminates the need to build multi-chip processing solutions, resulting in smaller, more efficient embedded AI systems and the potential to shorten time to market.
According to the introduction, this series of devices uses FPGA programmable logic architecture for real-time processing in the preprocessing stage, flexibly connects various sensors and realizes high throughput and low delay data processing workflow.
in the AI inference stage, the next generation AI engine is constructed by vector processor array to realize efficient AI reasoning.
In the post-processing stage, the Arm CPU kernel is used to provide the post-processing capabilities needed for complex decision-making and control for security-critical applications.
At the same time, this series of devices can also bring scalar computing power up to 10 times higher than that of the original generation.
However, it can not be ignored that the significant improvement in scalar computing power also brings the corresponding increase in power consumption.
According to Manuel Uhm, the second generation will consume much more power than the first generation of VE2302.
But he added, “if you want to implement the same DMIPS and use the first-generation 2302 products, you need to use an external processor, and the resulting total power consumption is actually higher than that of the second-generation products.
” Photo: AMD, it is worth mentioning that another big highlight of this new product launch is that compared with the first generation products, the second generation is mainly aimed at the central computing engine, while the first generation is more focused on CPU acceleration.
At present, the automotive industry has entered the era of standard ADAS, more and more sensors are concentrated in a domain controller, that is, central computing.
Following this trend, the second generation of Versal products launched by AMD can be used for data fusion in the “automotive brain”.
Officially, the combination of the second-generation Versal AI Edge series and the second-generation Versal Prime series provides AI-driven systems with scalability from edge sensors to central computing, allowing customers to choose performance, power consumption, and board footprint to efficiently achieve application performance and security goals.
In addition, the product portfolio can also simplify the design cycle AMD said that AMD Vivado ™design suite tools and libraries help increase productivity and simplify design cycles for embedded hardware system developers, thereby reducing compilation time and improving the quality of results.
For embedded software developers, the AMD Vitis ™Unified Software platform supports embedded software, signal processing, and AI design and development at the user’s preferred level of abstraction, without the need for FPGA experience.
For AMFor its part, its breakthrough in FPGA and its aggressive attack in data centers are largely due to the completion of the acquisition of Xilinx, the world’s largest FPGA brand, in February 2022.
At the time, Victor Peng, former president and CEO of Cyrus and later president of AMD Adaptive and embedded Computing (AECG), said, “the rapid development of connected devices and data-intensive applications of embedded artificial intelligence has driven the growing demand for efficient and adaptive high-performance computing solutions.
The combination of AMD and Cyrus will provide a very comprehensive portfolio of adaptive computing platforms that power a wide range of intelligent applications, thus accelerating our ability to define a new era of computing.
” Photo source: AMD,AMD continues to accelerate in the AI era.
At present, the early access documents for AMD’s second-generation Versal products have been released.
It has been revealed that AMD expects to provide second-generation Versal series chip samples in the first half of 2025, evaluation kits and system module (SOM) samples in mid-2025, and mass production chips by the end of 2025.
The demand for artificial intelligence embedded applications is exploding, driving the demand for single-chip solutions that achieve the most efficient end-to-end acceleration within the power consumption and board footprint constraints of embedded systems.
Building on more than 40 years of leadership in adaptive computing, these latest generation Versal devices integrate multiple computing engines into a single architecture that will provide high computing efficiency and performance as well as scalability from low-end to high-end.
” Salil Raje, senior vice president and general manager of adaptive and embedded computing division of AMD, said.
, return to the first electric network home page >.
