Less than a month after Nvidia announced the acquisition of Arm, AMD reacted quickly and was exploded with plans to acquire Xilinx for more than $30 billion. The acquisition may reach an agreement as early as next week.

Foreign media commented that AMD’s move may be a quick response to rivals Intel and Nvidia, and was driven by a series of recent mergers and acquisitions. Intel acquired Xilinx’s main competitor Altera for US$16.7 billion in 2015; not long ago, NVIDIA acquired Arm for US$40 billion, triggering an “earthquake” in the global semiconductor industry.

The successive actions of Intel, Nvidia and AMD confirm the technological changes driven by the demand for massive data of CPUs, GPUs, and FPGAs. Behind the deeper lies the upsurge of artificial intelligence and deep learning, as well as high-performance computing markets such as 5G and data centers. A strong driver for heterogeneous computing.

Since its birth, computer processing capabilities have been developing rapidly. As of the last ten years, with the continuous popularity of new technologies such as big data, blockchain, and AI, people have developed a variety of different technical ideas to increase the speed of computing processing, from single-core, multi-core, to heterogeneous computing. It is proposed to bring greater potential for computing power to the computer.

With the exponential growth of computing demand, heterogeneous computing has also experienced a variety of routes such as CPU+GPU and FPGA. In 2010, when we entered the heterogeneous computing era of CPU and GPU, we were impressed by the amazing computing power of GPU. During this period, Nvidia entered the mainstream market with its massively parallel GPU and dedicated GPU programming framework CUDA.

But while the graphics card is performing calculations, the processor is in an idle state. Therefore, processor manufacturers also want to participate in the calculation. They hope that the CPU and GPU can work together to complete those applications that have strict requirements on the amount of calculation. The processing power of the computer has pushed a new peak. Subsequently, the area covered by computing broke through the category of image recognition and entered into other tasks such as speech recognition and natural language understanding. This made FPGA, which is different from GPU, and has hardware pipeline parallel and data parallel processing capabilities, began to enter heterogeneous computing. category.

Since then, FPGAs have been trying to develop in the HPC (High Performance Computing) and data center markets. Intel, AMD, represented by CPU, NVIDIA represented by GPU, and FPGA camp represented by Xilinx, began to collide, and mergers and acquisitions followed.

A technical expert in the FPGA field told Jiwei.com: “AMD’s acquisition of Xilinx is similar to Intel’s acquisition of Altera for purposes and reasons. For Xilinx, its traditional FPGA business development has also encountered bottlenecks and needs to turn to data. In the central market, AMD needs to use FPGAs to coordinate development in the fields of heterogeneous computing, computing acceleration, AI, and data centers. The two parties are very compatible."

Intel acquired Altera in 2015 and integrated the company into its Programmable Solutions Group (PSG), integrating with its own CPU, GPU and other technologies and assets to open up new high-growth market segments. With the help of Altera's technology, Intel snatched Xilinx's cheese in order to occupy a larger market space in the field of artificial intelligence. Last year, the two sides' competition for the "world's largest FPGA" brought this contest to a higher level.

In 2017, Microsoft announced the use of Altera FPGAs in data centers; in 2018, Xilinx announced its "data center first" strategy, and the CEO announced that Xilinx "is no longer an FPGA company." As the competition intensified, in order to stabilize their strength and expand their market share, Intel and Xilinx not only upgraded their own technologies and product arrays, but also increased their firepower in mergers and acquisitions and building an ecosystem.

On July 13, 2018, Intel announced the acquisition of eASIC, a US structured ASIC supplier, and merged it into the PSG Division.

Five days later, Xilinx announced the acquisition of Shenjian Technology, one of the most watched unicorns in China's AI chips.

Subsequently, Mellanox, a network equipment supplier for Israeli startups’ server and storage connectivity solutions, has also become a "sweet cake" for giants. From November 2018 to the end of January 2019, it was rumored to be bid by Xilinx, Microsoft, and Intel. , But was eventually cut by Nvidia who threw $6.9 billion in March by "Cheng Yaojin who was killed halfway". In fiscal year 2019, the data center business became the fastest growing business for Nvidia. Nvidia GPU+Mellanox RDMA+NVIDIA CUDA formed an unshakable overall solution in the high-performance computing acceleration market.

For AMD, in the past three years it has fought a beautiful raid in the field of data centers, gaining 10% of the market share from Intel's dominant market, causing its stock price to soar this year. If AMD succeeds in acquiring Xilinx, AMD can not only enhance its core competitiveness in the data center field, but also further compete for market share from Intel, and be even more wary of NVIDIA's raids.

Once AMD successfully teamed up with Xilinx, the pattern of the high-performance computing market may eventually form a third of the world.

In addition, the above-mentioned experts also believe that as international giants switch the future track, this transaction will give domestic manufacturers more opportunities in the traditional FPGA market.