Intel released a programmable accelerator card equipped with Intel aria 10gx FPGA chip; Dell and Fujitsu, two OEM manufacturers, announced that they would integrate this programmable accelerator card into their own servers on a large scale to support various workloads in the data center.

At the same time, Dell and Fujitsu will also integrate a new version of Intel accelerator stack for FPGA oriented Xeon scalable processor in their respective products. This is a cross industry common software stack, which can provide the underlying tools required for FPGA integration, including data transfer, configuration accelerator, support for virtual machine, support for cloud data center migration, etc.

Liu Bin, Asia Pacific market development manager of Intel’s programmable solutions division, showed the accelerated application of Intel FPGA in finance and data center. He also said that the cooperation announced today with Dell and Fujitsu means that Intel’s programmable accelerator card is also officially mass produced. In the future, more OEM manufacturers will announce to join the cooperation.

Coincidentally, Xilinx, another player in the FPGA market, also announced the data center first strategy. At the same time, NVIDIA’s GPU and a number of SOC startups have also focused on the data center field. Liu Bin told Zhidong that the increasingly fierce market competition just proves that data center is the next huge growth point of FPGA. At present, the accelerated computing business in the data center is still in the “blue ocean” stage, and emerging businesses such as AI and data analysis are still developing rapidly. All enterprises have room for positive development.

Competition between Intel FPGA accelerator and NVIDIA

Intel’s advantage lies in its long-term accumulation of computing power. On the one hand, it has a series of powerful chips and software computing power. On the other hand, FPGA can cooperate to provide flexible hardware computing and create a powerful heterogeneous computing combination. Accelerator card is a form of FPGA entering the hardware field, but at the same time, Intel is also considering other ways and other forms of products, such as accelerators for deep learning inference.

At present, although the data center market dominated by cloud services has a certain volume worldwide, database analysis is still a rapidly growing and popular field. According to the market research media report, from 2017 to 2022, the database acceleration market will also develop at a CAGR of 23%, reaching a market size of US $3.85 billion by 2022.

In the case given by swarm64, a partner of Intel, swarm64, after adopting the Intel FPGA platform, the speed of real-time data analysis is increased by an average of 20 times, 2 times higher than the performance of traditional data warehouse and 3 times higher than the performance of old storage compression.

At the same time, Liu Bin also showed the application of Intel FPGA in financial risk analysis. Risk analysis is a huge market. According to the marketsandmarters report, the wind direction analysis market will develop rapidly at a compound annual growth rate of 15% from 2017 to 2022, reaching a market size of US $35.9 billion by 2022.

However, the accelerated calculation of financial risk analysis also faces several major challenges. First, financial risk analysis is essentially a big data problem. The amount of data can be up to several terabytes. At the same time, it is a new application, which consumes a lot of computing power. Moreover, the model of financial risk analysis is also changing, which makes the calculation more complex. The third problem is that if it is implemented with pure software, the performance of financial risk analysis and calculation will be limited, and the calculation delay will be large, which can not meet the requirements.

In the case given by Intel’s partner levyx, they introduced a financial risk analysis system to the market. Compared with the results obtained by running database query with swarm64 acceleration and software implementation, the results show that the acceleration speed of specific symbolic algorithm with Intel’s solution is increased by 850%, In the simulation acceleration of end-to-end option trading, the acceleration ratio is more than twice.

Although the data center market dominated by cloud services has a certain volume worldwide, the data center is still a rapidly growing and popular field. Over the past year, the data center chip market has been booming. In addition to old chip players such as Intel, NVIDIA and Xilinx, many chip start-ups at home and abroad have also targeted this market. While the market scope is expanding, the competition is becoming increasingly fierce.

Leave a Reply

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