AI development is at its dawn, and AMD uses innovative chips and tools to assist in AI implementation

Release Time:2023-6-24 12:12:50

Now, many companies are conducting AI-related application research in different fields. Many AI projects will choose to use FPGA for verification at the beginning. For complex applications, large FPGA chips will be selected, and for simple applications, small FPGA chips will be selected.If the AI application scenarios and network are determined, ASIC chips will have more advantages in terms of efficiency. "The premise is that there are certain elements. When some elements have not been determined, it is not easy to solidify an ASIC." Tang Xiaolei pointed out when sharing .
Therefore, she believes that AMD's adaptive architecture IP---AMD XDNA will be a better choice.She explained that XDAN is an IP combination of adaptive architecture, mainly for AI, and the architecture between traditional signal processing and data flow.It is a highly scalable array of engines with local memory and data movers, and is an algorithm tool that leverages AMD's deep FPGA and adaptive SoC compilation algorithm expertise.

961f01ab02e0e546aebe1f823f191f46_wKgaomSS-ryADcTKAANi-V5dqPo838.png

Among them, the AI engine can enable AI and signal processing to achieve high performance and high energy efficiency.Adopting an AI engine means constantly investing in research and development when changing, but the overall platform does not need to be changed significantly, and customers can accumulate knowledge on the platform.Tang Xiaolei also revealed that in the future, AMD will apply more chiplet technology to cope with the arrival of the AI wave.

31d5e8ff397d2e91bb192f589905c793_wKgZomSS-s-AaLkyAAR1AUcfvQA147.png

In addition to XDNA, AMD also launched the Kria system-on-module (SOM) three years ago, focusing on vision-related applications such as security systems, city cameras, and road cameras.The SOM can help developers open up the peripheral interfaces, allowing them to implement their algorithms faster and put them into practice.
According to Tang Xiaolei, future SOM products will be divided into two product lines, one will focus on cost optimization, focusing on electric drive and other size and cost-constrained applications; the other will focus on providing higher computing power.Because computing power has become an unavoidable topic, developers’ demand for computing power is increasing exponentially every day, and SOM’s computing power cannot be completely dependent on the cloud. More and more application scenarios require end-to-end, Computing support on the edge side to process data with high real-time requirements.

37d36e91405bb467c3b1227fd0b5d944_wKgaomSS-uCAEekjAAOteSXWp0I494.png

In addition to these hardware innovations, the importance of tools cannot be ignored on the FPGA and X86 platforms.Especially with the advancement of AI and the addition of more software engineers, the demand for tools is more important.Therefore, AMD also updated the Versal development environment.The development environment is oriented to all developers, including heterogeneous platforms at the bottom of the hardware, such as Versal ACAP, scalar engine, programmable logic, and AI engine; in the middle layer, the middleware provided by AMD can provide developers with an interface to use tools On this basis, AMD provides the EDA back-end tool Vivado; on top of the EDA back-end tool is the heterogeneous software platform Vitis, on this platform, software developers can easily use their familiar C, C++, or Python for development ; In addition, AMD also provides the option of no hardware compilation, using cloud servers for innovation without software compilation.
Tang Xiaolei said that AMD is using adaptive computing to accelerate customer innovation in core markets, including automotive, industrial, medical, audio and video, and TME (test, measurement and simulation) and other fields.

Automotive: From immersive cockpits to autonomous driving

Today's automotive industry is facing unprecedented changes, and intelligent transformation is happening rapidly.However, the intelligentization of automobiles has different paths, whether it is a technical route or a landing method. Among them, autonomous driving is mainly developing in the direction of Robotaxi.Traditional cars or new energy vehicles put more emphasis on immersive experience, smart cockpit, and improving the driver's sense of safety.
First of all, today's automotive industry is very complicated, and the design cycle has changed from planning products five years later this year to planning next year's products this year, which brings different challenges to customers.In Tang Xiaolei's view, the first customer facing challenges in this wave is the car factory.Although the iteration speed and application scenarios are increasing, AMD still hopes to provide a scalable platform for customers to use.For example, in the immersive smart cockpit, the demand from 4K screen (with fish screen) to six 8K screens supporting 3A games provides different market requirements for AMD.The market's demand for the ASIC design cycle has been greatly accelerated.Nowadays, iteration cycles and product upgrades can no longer meet product design cycles. AMD's adaptive platform provides customers with a different product design experience.If customers want to drive two screens, they may need a smaller FPGA, namely vision hub.An expansion through AMD's FPGA products can do more, such as supporting 13 cameras and 4 screens in the future.

84a5741eced58dd17012fa14170560d2_wKgZomSS-vmAdwRdAAgrupBblTc660.png

Secondly, when your immersive experience requirements are not so complete, you may not need to play 3A games in the back row, or do some more advanced game experience, you may not need a GPU.However, if it needs to be installed in the future, customers can completely make the GPU into a box to increase the customer experience, or turn the GPU into an option for configuration.From the CPU platform itself, the migration of X86 is not difficult.If you are using the ZEN 3 architecture now, after we launch the ZEN 4 architecture in the future, it will be very simple for the migration of the entire platform and ecosystem.Moreover, the cockpit is a very independent space, and the X86 can bring a very complete and more powerful experience in terms of the supported application scenarios and the existing ecology.
Relatively speaking, Android is more suitable for mobile scenarios such as mobile phones.When pursuing some extreme experience, Android has certain usage restrictions.As the space expands, platforms with top-level processing capabilities are needed to satisfy related experiences.
Autonomous driving is completely different.Autonomous driving has very high requirements on computing power, so server-level chips are usually used in this scenario.In this field, AMD's advantage is that it has a very strong X86 to do real control.In terms of computing, how can X86 and GPU combine with FPGA to make the vehicle network more efficient and more complex.In the field of autonomous driving, AMD's unique advantages can help customers continuously upgrade iterations and adaptive computing.
Regarding the energy consumption of FPGA, Tang Xiaolei believes that this is no longer a problem, and the car is undergoing a huge transformation process from machinery to electronics to intelligence.Whether it is a new energy vehicle or a traditional vehicle, in order to provide a better experience, the automotive industry and semiconductor companies are generating unprecedented tight coupling to meet the stronger computing power required for development.This fusion process is data-driven.In this world, AMD's unique position lies in the integration of CPU, GPU, FPGA, and Semi ASIC to help customers in the automotive industry realize the transformation of big data computing.
In addition, with the development of electric driving force, energy consumption and power consumption are no longer the most critical or unbreakable issues.Now that electric vehicles have adopted a water cooling system, adding a design with higher power consumption will not cause additional problems in terms of cost.In the past, the electricity was 12 degrees, but now the car battery capacity has reached 120 degrees. In this case, it is relatively insensitive to the power consumption of the chip. "In addition, our heterogeneous computing better balances power consumption and computing power. It has subverted the thinking habits of traditional users. Therefore, through communication with customers, we found that the introduction of FPGA, GPU and CPU will not increase the energy consumption or price of the car, but a necessary solution.” Tang Xiaolei emphasized.

industrial innovation

In the field of intelligent manufacturing, AMD uses digital twin technology to make projects quickly and accurately landed.Another application scenario is smart warehousing, making logistics solutions more efficient.Also in unmanned factories, unmanned moving vehicles (AGV) and automatic meter reading (AMR) are needed to allow people to understand what is happening in the warehouse in real time and reduce manual intervention in the factory.
Smart manufacturing is inseparable from industrial robots. In the field of industrial robots, AMD has been committed to making the automation process smoother and enabling more precise control of automated robots.

a1e42c68121858517fa515d69f70923f_wKgaomSS-xWAVHC5AAevZ4WUOdg375.png

There are many other innovations about product verification. For example, product verification is increasingly transforming from human eye verification to machine verification and intelligent visual inspection.In the entire industrial innovation, from the establishment of the plant, the digital twin of the project, and the logistics in the middle, the final production, and the control of product quality, it can cover the entire intelligent manufacturing process.
This process covers a lot of underlying technologies, including sensing, control, computing, interconnection, how to make networking more efficient, and how to enable sensing to provide more data.Tang Xiaolei said that AMD also combines the underlying technologies such as control sensing, control, computing and networking, so that data can more effectively complete the mission, and promote China's manufacturing to China's intelligent manufacturing.

Help medical innovation

Medical innovations include speculum systems, medical ultrasound, surgical robots, large scanners, and patient care.Tang Xiaolei mainly introduced some relatively new and creative solutions in the field of medical innovation, such as endoscopic systems, surgical robot systems, medical ultrasound, and CT, MR and large scanners.

17ffd76d7a854728d36d7021534395e4_wKgaomSS-yqAZgsiAAQesnD0Spk744.png

According to her introduction, in terms of endoscopes, as the resolution of endoscopes increases, there are more and more requirements for bandwidth and low latency.AMD's real-time adjustment of broadband and image resolution support and flexible computing processing give AMD a great competitive advantage in endoscope resolution; in medical ultrasound, AMD implements visualization and intelligence to improve the overall The efficiency of the medical system has been greatly promoted in the real-time processing of signals; in the surgical robot system, AMD's technology has also been applied, and the surgical guidance is realized through 5G network and AI machine learning reasoning to better use medical resources; in CT For , MR and large scanners, the offloading and acceleration of image reconstruction is processed on AMD's platform, making the imaging density higher and higher.
Tang Xiaolei pointed out that medical care, as a humanistic care project, has always been an important research and development direction of AMD.In order to cooperate with the rapid implementation of medical innovation, AMD provides customers with imaging libraries and releases medical e-books, enabling them to accept, use and implement adaptive computing faster, helping customers complete rapid changes and iterations in medical care.AMD is also planning to apply AI to medical imaging libraries as soon as possible.
She also emphasized that the medical field has always been an area that AMD attaches great importance to and maintains close cooperation with global customers.Because of the long design cycle of medical equipment, various certifications and hospital procurement processes are required, the acceptance of new products and new technologies in the medical field is relatively slow.But as innovation evolves, so does the need for innovation in healthcare.Many people are thinking about how to introduce AI, such as accelerating the analysis of X-ray films and lesions.These are the driving forces to promote the improvement of medical level and improve the efficiency of resource utilization. These driving forces require a large room for improvement of medical equipment.
At present, whether it is endoscopy or medical ultrasound, AI technology is incorporated, which drives innovation and acceptance of new devices."In the past two years, the medical industry has begun to apply our advanced devices on a large scale." Tang Xiaolei said.
With the implantation of AI, the increase of processing power, the improvement of network delay and the addition of telemedicine, the medical industry has begun to pay attention to communication and networks, and smart hospitals have also begun to consider the interconnection of equipment.She further noted, "These are the drivers of innovation and the reason why AMD's high-end products are widely used in the medical field. So we see new devices, big devices, good devices being used more and more in the medical field .”
In addition, Tang Xiaolei also introduced the innovative applications of its products in embedded industry, vision, medical system synergy, professional audio and video, broadcasting, consumer applications, testing, measurement and simulation.


Disclaimer: The content and pictures of this article are reproduced athttps://www.elecfans.com/d/2146724.html.The article and its pictures are only for engineers to learn. If there is any infringement of content pictures or other problems, please contact this site for infringement and deletion!



Technical Support: WangGuang Network Copyright © Shenzhen OCIC R&D Co., Ltd 粤ICP备19007470号-1