The Year of Science and Technology in China
On February 14, 2021, AMD announced that it would acquire FPGA leader Xilinx at a valuation of $50 billion.
As AMD says, in the next decade, high-performance computing will be at the heart of almost all the major megatrends that will shape the world of the future. Whether it is in cloud computing, edge computing, or more and more intelligent terminal devices, there will be a growing demand for new experiences and services. While CPUs and GPUs will continue to be critical computing components for these devices, accelerating these emerging and changing workloads is critical in a world where algorithms are advancing and new standards are emerging, and we foresee a growing demand for adaptive computing power.
To that end, AMD believes the acquisition of Xilinx will create the industry's leader in high-performance and adaptive computing, bringing together a highly complementary set of products, customers and markets, as well as differentiated IP and world-class talent. The combination of AMD and Xilinx will benefit product, technology, marketing and finance.
In fact, since the acquisition, the two sides have done a lot of integration and innovation. Now, as we approach the second anniversary of the acquisition, AMD is bringing a new product, the AMD Embedded+ architecture, which includes Ryzen's embedded x86 processors and VerSal adaptive SoC technology.
The explosion of data at the edge brings new processing requirements
Sensor data, both visual and non-visual, is more important than ever in a wide range of applications. Chetan Khona, senior director of industry, vision, medical and science at AMD, shared his observations on the industry. He further said that the amount of data we generate today is more objective than ever, but it is only by putting it well to the best that it can be used to make the most of your decisions. "A lot of decisions also need to be made in a very short period of time, within a defined timeframe, especially for industrial and medical applications, which is often in the millisecond range. Chetan Khona emphasized.
Therefore, in his view, in order to be able to process data in a timely manner, it is necessary to do data processing close to the sensor device side of the data generation, which is what we often call the limit edge, embedded edge, or number boundary, which in turn promotes the rapid growth of the entire edge processing market. This trend can be clearly seen from the data provided by ABI Research in the chart below.
But then a new problem arises, that is, without a consumer interface such as USB, it is difficult to import the sensor into the PC.
In industrial and medical applications, custom or non-consumer ports are often used. This forces us to need an optimized process or healthcare PC that can import your sensors into it. "Of course, you can use FPGAs or embedded SoCs and SOMs to bypass PCs and get things done." But that's not very feasible, as you'll still need a PC for certain tasks. Chetan Khona continued.
The new AMD Embedded+ addresses some of the above requirements with an architecture that simplifies the path of data import to a PC. "Whether the data is coming from multiple discrete sensors or from an industrial network, AMD Embedded+ enables developers to take data, process it in real time, and even provide visualization options. Chetan Khona said.
An architecture that combines an x86 processor and a VerSal adaptive SoC
According to reports, EmbedDED+ is a scalable method developed by AMD's engineering department, which is mainly to connect Versal adaptive SoC and Ryzen x86 embedded processors through PCIe on the same motherboard. It is worth mentioning that the end product itself is provided through an ecosystem partner authorized by AMD.
In terms of architecture itself, AMD Embedded+ has three main features:
First, the use of Versal adaptive SOC programmable IO, which can adapt to various sensors and networks;
Second, it can reduce the burden of handling tasks. The x86 processor is the core of the architecture, but there are also some load-reducing components around the x86 processor, which can be used for deterministic and low-latency communication and processing programmable logic, AI engine for high-performance performance inference, and video integrated display that can be used for more exquisite visualization, and of course, there are other load-reducing modules, such as dedicated codecs;
Third, it reduces the time to market. Because this architecture is optimized for sensor fusion, AI inference, industrial interconnection, control, and visualization, it can use the general software technology architecture to connect the various parts of the solution just mentioned.
First of all, in terms of sensor friendliness, according to Chetan Khona, the combination of Ryzen's Embedded and Versal AI Edge can provide connectivity options with traditional PCs and networking devices, as well as real-time sensors and industrial networks, giving everyone the best of both worlds in a very complete system. "The Versal adaptive SoC can provide programmable IOs that interact with a variety of sensors and networks. The Embedded+ architecture provides users with 84 IOs. Chetan Khona added.
According to AMD, the EmbedDED+ architecture can be applied to a variety of applications such as machine vision, industrial connectivity, robotics, smart city, security, retail, and healthcare. Especially in the medical field, the developers are even more impressed.
Generally speaking, medical devices such as B-ultrasound or endoscopic probes have a dedicated sensor interface, which can be connected to the programmable IO of the EmbedDeDed+ system, and can be used to complete high-resolution wave shaping, image signal processing and image reconstruction and other complex algorithms with ultra-low latency using the Versala SoC, which can then be rendered and visualized by Radeon graphics cards. In terms of system specifics, AMD's architecture currently supports Ubuntu's Linux. Going forward, the company will also support Windows.
As for the "uninstall handling" feature, it provides developers with more flexibility. It is understood that AMD's Embedded+ architecture can provide an interconnect architecture for extended connectors, programmable IO, and FPGAs. If real-time control is required, the architecture can provide an interconnected architecture of ARM's subsystem and FPGA; If you need AI inference, you can add Versal AI Edge series products to the Embedded+ AI engine to meet your needs. If what is needed is decoding, rendering, and displaying, the Embedded+ architecture is able to provide H264 265 ** codecs and Radeon graphics cards.
Using x86 processors with programmable logic as a task decentralization engine can be used in many applications. In addition to the range of functions that can be achieved, we can even implement safety-critical functions. Chetan Khona said.
In terms of hardware, in addition to expansion connectors, ODM manufacturers will also provide some threaded expansion boards, so that end users can connect to some commonly used interfaces. Of course, custom boards are also very popular because they make it easier for users to use. Coupled with the two communication flows above the software architecture, the semi-custom or fully customized mode is selected, and a tool called Vitis XRT is used to communicate between Ryzen and Versal devices. This makes it easy for users to call custom cores from Versal devices from Ryzen™ Embedded processors. Achieve the goal of "fast time to market" mentioned above.
As mentioned earlier, AMD's new architecture is offered through authorized OEM partners, with its first product coming from Sapphire Technology.
According to reports, this ODM solution based on the EmbedDED+ architecture is the Sapphire Edge+ VPR-4616-MB, which is a low-power mini-ITX motherboard from Sapphire Technology. Powered by the Ryzen™ Embedded R2314 processor and the Versal AI Edge VE2302 adaptive SoC, it delivers a full suite of features at power consumption as low as 30W. In addition, the VPR-4616 is system-wide, including memory, storage, power supply, and chassis.
Adrian Thompson, Senior Vice President of Global Marketing at Sapphire Technology, said, "With a proven and reliable computing architecture, we are able to focus our resources on enhancing other aspects of our products, thereby reducing time-to-market and R&D costs. Embedded+ is a great simplified platform for building solutions with leading-edge performance and functionality. ”
Applications such as machine vision, robotics, medical imaging, industrial networking, smart cities, security, and retail all require the input, processing, and deterministic response of the lowest latency sensors, Chetan Khona concluded. The AMD Embedded+ architecture combines the best performance of Ryzen™ embedded x86 processors and Versal AI Edge adaptive SoC into an integrated computing platform. Of course, AMD is also leveraging AMD's PC ODM partner ecosystem to launch products with EmbedDED+ architecture to achieve rapid deployment.
The integration of AMD and Xilinx brings a 1+1 greater than 2 advantage in terms of technology. Chetan Khona reiterated.