FPGA Based Accelerators for Financial Applications Book
I am still looking at Matlab’s Simulink as an option to rapidly deploy this
This book covers the latest approaches and results from reconfigurable computing architectures employed in the finance domain. So-called field-programmable gate arrays (FPGAs) have already shown to outperform standard CPU- and GPU-based computing architectures by far, saving up to 99% of energy depending on the compute tasks. Renowned authors from financial mathematics, computer architecture and finance business introduce the readers into today’s challenges in finance IT, illustrate the most advanced approaches and use cases and present currently known methodologies for integrating FPGAs in finance systems together with latest results. The complete algorithm-to-hardware flow is covered holistically, so this book serves as a hands-on guide for IT managers, researchers and quants/programmers who think about integrating FPGAs into their current IT systems.
This was mentioned https://www.maxeler.com/products/software/
Use maxcomliler in java with python to throw out useless
c++ code with object orientation and templates. Slows down the processing. Duh.
It take .5-3 years to implement a model into production because no one follows coding standards.
All support both windows and gcc compilers but now support native python to. Some groups wants excel support as well. The different classes and inherited templates morphed into weeds to slow down the execution. This was the result of many years of coding models.
Code design efficiency Is key to save on power consumption and get speed on execution. Super pipelining and ultra parrelllelism is critical.
In video 1:05:48 lists all tools including maxcompiler, python API,triton lib, etc
Proteus is used for risk management http://rm-inv.enisa.europa.eu/tools/t_proteus.html
They can run their entir enterprise book in 12 seconds with 40 node fpga system as opposed to 8 hours.
Jpm was som impressed they bought 20% of maxeler that must mean they want to control who the technology goes to including the banks of competition.
They are using fpga for decision tree for fx trading and hft.
It does look like jpm does use Xilinx fpga hardware