When Broadcom’s acquisition of Qualcomm was blocked by US President trump, the next “target” identified by Wall Street analysts was Xilinx, a San Jose chip manufacturer. In this regard, Victor Peng, President and CEO of Xilinx, responded to the media for the first time, saying that “we do not comment on such rumors and speculation. Both myself and our senior management team and board of directors focus on our own strategy and how to make the business more successful and serve customers better.”

On January 29 this year, Victor Peng officially became the fourth CEO in Xilinx’s history. Since joining Xilinx in 2008, the company has won three consecutive titles in three generations of process products of 28nm, 20nm and 16nm, and won the first place in each industry. At the same time, through the breakthrough of integration and programming mode, FPGA has moved towards a broader application field. Recently, Victor Peng introduced the future vision and strategic blueprint of Xilinx for the first time as the CEO at the helm, and released a breakthrough new product beyond the function of FPGA, called ACAP (adaptive computing acceleration platform), which led Xilinx to go beyond the limitations of FPGA and provide support for the rapid innovation of various technologies from endpoint to edge and then to the cloud.

Xilinx’s three strategies to deal with technological change

In Victor Peng’s view, the data explosion of geometric multiple growth, artificial intelligence applications from endpoints, edges to the cloud and computing in the post Moore’s law era that cannot be satisfied by a single architecture will become three major trends affecting Xilinx and the world in the future. The explosion of data has higher and higher requirements for the speed of computing. In Greater China, where the innovation speed maintains rapid growth, the speed of Moore’s law is gradually slowing down. The original chip solutions can no longer meet the needs of the company, and there is an urgent need to develop new products, new technologies and new business models.

In order to better meet this new world of intelligent interconnection, Xilinx continues to take “flexible platform” as the product core, seize new industrial opportunities, and formulate three development strategies to support a wider range of market applications. Victor Peng pointed out that the first strategy is “data center first”. In the field of data center, it is very important to understand that Xilinx can not only support the computing acceleration and application of data center, but also support the storage and network that create value.

In the field of data center, Xilinx will change the original technical direction of supporting hardware developers, support software developers more, and let more data center users use Xilinx’s flexible platform. Victor Peng said that the reason why he attaches so much importance to the data center is that it is a very broad market and develops very rapidly. Many industry subversions occur in the data center. In the field of data center, many new architectures no longer focus on CPU, but turn to heterogeneous computing with many accelerators, and can adjust the workload according to the processing of geometric data and machine learning. Therefore, for Xilinx, the data center is a new growth opportunity.

Not only that, Xilinx has also made great efforts in building the ecosystem of the data center, from ccix, members of the open standards organization and ccix such as AMD, arm, IBM and mellanox, to super large data companies such as Amazon, Huawei and Baidu, as well as deep knowledge technology at the application level. In terms of accelerating applications, Xilinx supports a wide range of AI applications, including machine learning, machine learning inference and so on. According to the data given by Shenjian technology, Xilinx provides an efficient FPGA accelerated speech recognition engine. Compared with the CPU, the performance of the engine is 43 times that of the latter, and the performance per unit power consumption is increased by 40 times. In terms of genome sequencing and analysis, the server using edico FPGA reduces the time of genome sequencing from one day to tens of minutes, so it is of great significance and key to the survival of newborns.

In terms of the second strategy, Xilinx will accelerate the growth of the mainstream market and support the rapid growth of customers in the fields of automobile, wireless infrastructure, wired communication, aerospace, industry and medical treatment, consumer electronics and so on. Victor Peng stressed that Xilinx can participate in the process from endpoint to edge to cloud, which is our unique advantage. It can truly realize full coverage from endpoint to edge to cloud.

In order to drive flexible computing, Victor Peng said that Xilinx will launch a new series of product types – ACAP (adaptive computing acceleration platform). If FPGA is a product type with many architectures, ACAP is an individual product type that can be compared with CPU and GPU to help us realize intelligent interconnection and flexible world.

A new generation FPGA Architecture comparable to CPU and GPU

Victor Peng emphasized that AC + + AP is a highly integrated multi-core heterogeneous computing platform, which can be flexibly modified from the hardware layer according to the requirements of various applications and workloads. ACAP has the flexible adaptability of dynamic adjustment in the working process, and realizes the performance and power consumption ratio that can not be achieved by CPU and GPU. In terms of software, it can be programmed. It not only supports hardware developers as in the past, but also supports more software developers and software tools such as C / C + +, OpenCL and python.

In terms of the first application product, Victor Peng told reporters that he was promoting the action code named “Everest”. After four years of research and development, the cumulative R & D investment exceeded US $1 billion, and more than 1500 software and hardware engineers participated in the design of “ACAP and Everest”. At present, software tools have been delivered to major customers. The first “Everest” product adopts TSMC 7-nm process, which will be streamed today (2018) and delivered to customers in 2019. This product will focus on medium and high-end products, and there will be more than 50 billion transistors in high-end products.

With the latest 16 nm virtex ® Compared with vu9p FPGA, “Everest” is expected to improve the performance of deep neural network by 20 times! The 5g remote radio head end based on “Everest” can increase the bandwidth by 4 times compared with the latest 16 nm radio. At that time, the performance and power efficiency of various applications across multiple market areas can be significantly improved. These markets include automotive, industrial, scientific and medical, aerospace, testing, measurement and simulation, audio and video and broadcasting, and consumer electronics.

Comparing the differences between ACAP, SOC and zynq, Victor Peng pointed out to the on-site reporter that SOC and zynq are the pioneer products of the older generation. For example, MPSoC has DRAM and ultraram, which is different from traditional FPGA. For example, traditional FPGA has only one programmable logic and no SOC. ACAP can communicate directly with the whole system; In addition, we have an on-chip network to communicate directly with the subsystem, which is also the difference between ACAP and SOC; Finally, although ACAP also has multi-core, we have an engine that can be programmed in both hardware and software on the chip. This is a brand-new design, which our competitors do not have.

With regard to ACAP, Victor Peng is very confident and calls it the most outstanding engineering achievement since the invention of FGPA. Compared with the future CPU and GPU, it can cover a wider range of market applications and meet the needs of flexible workload.

Adhere to their own advantages and competition is everywhere

For a long time, the competition between FPGA, CPU and GPU has been continuous. Victor Peng analyzed that the traditional CPU architecture can no longer adapt to such a large-scale data computing and workload in the industry. Although GPU has advantages in the acceleration of some applications and workloads, the applications supported by FPGA are more extensive, and heterogeneous computing will be more and more needed to adapt to different applications in the future. ACAP can accelerate for different workloads and has highly adaptive strain ability. With the passage of time, ACAP will become a mainstream platform. Like CPU and GPU, ACAP will become a very basic construction platform for the future world.

The reason why the acceleration of data center is the new focus of Xilinx’s development is that data center is a field of rapid popularization of technology. Victor Peng added that the data center can program and adjust the workload of its server according to the computing needs, such as video transcoding during the day and image recognition at night. This is of great significance and can be applied in fields including artificial intelligence (AI) inference, video and image processing, genomics and so on.

Compared with NVIDIA and Intel, Xilinx has a longer history and profound heritage. Victor Peng stressed that in many market areas, such as the automotive industry, we have been deeply cultivated for more than ten years, which is a new market for Intel, which can enter the automotive market by acquiring Mobileye. NVIDIA has been doing infotainment system before. Now it is doing automatic driving, but it pays more attention to the safety of the car. Xilinx started from zynq products in the automotive industry, including spartan-6 products, with a total shipment of 40 million. In the field of these endpoints, we pay more attention to quality and reliability. We have great advantages in markets including medical industry and aerospace.

From this year’s revenue performance, Xilinx is still in the early stage of laying the foundation in the field of data center, and its revenue will not increase significantly. Victor Peng pointed out that we are still committed to building an ecosystem to help developers develop and apply faster and better. We cooperate with many innovative enterprises, including Shenjian technology, to cultivate the ecosystem in the form of investment. At the same time, we will better cooperate with colleges and universities to build an open-source community and promote the construction of the whole ecosystem.

Finally, Victor Peng said that Xilinx has cooperated with the three Chinese bat giants and some American technology companies in super large data centers in terms of cloud and Internet. At present, Xilinx has cooperated with all seven super large data center enterprises in the world, as well as with some OEMs and enterprises in the cloud environment.

Leave a Reply

Your email address will not be published. Required fields are marked *