After communication, the second life of FPGA

After communication, the Second Life of FPGA

# After communication, the second life of FPGA If we compare quantitative trading to a race, what happens in the first 400 nanoseconds after the gun is fired essentially determines the outcome of the day. In the server room of a leading quantitative trading institution in Shanghai, a research and development manager from a domestic financial acceleration provider spent an entire afternoon at the client’s site. A number on the screen kept appearing repeatedly: 400. This represented the total latency in the process of their domestic FPGA acceleration card receiving market data from the Shanghai Stock Exchange via optical ports, undergoing complete decoding, multi-channel selection, and then being sent to the user’s strategy engine. The unit was nanoseconds. This number may be abstract for most readers, but for quantitative trading institutions, it means whether they can secure that order. Software for parsing market data often starts with hundreds of microseconds, and the operating system itself has uncontrollable fluctuations. In this competition, a single microsecond difference can change whether an order can be “secured” or not. More than ten years ago, such scenarios were almost entirely handled by Xilinx (which was acquired by AMD in 2022). Domestic FPGAs were secondary options, backups. But today, at this server room...