Making the case: The role of FPGAs for efficiency-driven quantitative financial modelling

Mark Klaisoongnoen*, Nick Brown

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract / Description of output

Aiming at low latency, Field-Programmable Gate Arrays (FPGAs) have a long history in High-Frequency Trading (HFT) and have traditionally been used for operations such as networking, routing and market feed handling.
While the low-latency benefits of FPGAs have been obvious in HFT, there have been high barriers to programming such devices and the required specialised knowledge on behalf of the developers has slowed down the adoption of FPGAs in the wider community. That is until recently when vendors such as Intel or Xilinx have invested high efforts both in creating highly-performant generations of FPGAs, such as Intel's Stratix line and Xilinx's generation of Alveo cards and more important in crafting toolchains and software environments to lower entry barriers for software developers. Operating at lower clock frequencies than traditional computing hardware such as CPUs and GPUs, and therefore requiring less power, FPGAs are now programmable directly from higher level programming languages and therefore software developers can write software code to configure such devices making performance and energy-efficiency advantages of FPGAs available to a wider user group such as the quantitative finance community. The challenge is now in the design of the algorithms to best suit the FPGA, moving away significantly from the CPU version.
We illustrate the usability, performance and energy-efficiency advantages of FPGAs with three financial use cases: Having developed our experiments on a Xilinx Alveo U280 FPGA card, our Credit Default Swap (CDS) engine achieves a 1.5 times performance increase, whilst increasing power efficiency by 7.1 times compared to the parallel version on a 24-core Intel Xeon CPU. Our Black-Scholes hedging strategy, operating over discrete time intervals on the same Alveo U280 FPGA, achieves 69.9 times the performance of the ubiquitous reference version and our FPGA option price discovery algorithm performs between 1.5 and 8 times faster than on the CPU, delivering between 8.1 and 185.1 times improvement in energy efficiency respectively.
Original languageEnglish
Title of host publicationProceedings of Economics of Financial Technology Conference 2023
Publication statusPublished - 21 Jun 2023
EventEconomics of Financial Technology Conference - Edinburgh, United Kingdom
Duration: 21 Jun 202323 Jun 2023
Conference number: 2023


ConferenceEconomics of Financial Technology Conference
Country/TerritoryUnited Kingdom
Internet address

Keywords / Materials (for Non-textual outputs)

  • quantitative finance
  • reconfigurable architectures
  • high-frequency trading


Dive into the research topics of 'Making the case: The role of FPGAs for efficiency-driven quantitative financial modelling'. Together they form a unique fingerprint.

Cite this