快速且高能效的衍生品风险分析：在 Xilinx 与 Intel FPGA 上流式计算期权 Greeks
摘要
尽管FPGA在加速高频金融工作负载方面已取得成功一段时间，但其在量化金融（即使用数学模型分析金融市场和证券）中的使用迄今仍相当有限。当前，CPU是此类工作负载最常见的架构，一个重要的问题是FPGA能否缓解这些架构上遇到的一些瓶颈。在本文中，我们扩展了先前加速行业标准证券技术分析中心（STAC®）衍生品风险分析基准STAC-A2™的工作，首先将此从我们先前的Xilinx实现移植到Intel Stratix-10 FPGA，探索从一种FPGA架构迁移到另一种时所遇到的挑战及技术的适用性。然后我们提出了一种主机数据流流式方法，该方法最终在Xilinx Alveo U280 FPGA上以最大问题规模比我们先前版本快达4.6倍且能耗降低9倍，同时分别比CPU和GPU版本快达8.2倍和5.2倍。本工作的成果是针对该行业标准基准在Xilinx和Intel FPGA上运行的FPGA性能相较先前版本显著提升，此外还探索了可应用于其他HPC工作负载的优化与移植技术。
引用
@article{arxiv.2212.13977,
title = {Fast and energy-efficient derivatives risk analysis: Streaming option Greeks on Xilinx and Intel FPGAs},
author = {Mark Klaisoongnoen and Nick Brown and Oliver Brown},
journal= {arXiv preprint arXiv:2212.13977},
year = {2024}
}
备注
This work uses a benchmark of STAC, whilst this was approved at the time they have asked we remove the paper as it needs to be made more explicit that these are unofficial ports and are entirely independent from any vendor and don't follow STAC rules. As we are comparing vendor hardware in the paper, it was felt that this could easily be mistaken to be representing something that the paper is not