A Dataflow Overlay for Monte Carlo Multi-Asset Option Pricing on AMD Versal AI Engines

Monte Carlo simulations are widely used for financial applications, particularly in option pricing. Multi-asset option pricing has not been as widely investigated as single-asset problems and presents specific challenges. This paper presents an efficient way to design a dataflow overlay for multi-asset option pricing on AMD Versal AI Engines by leveraging their SIMD capabilities and the high-performance interconnect network linking them. The proposed dataflow overlay achieves a highly scalable parallelized implementation that can utilize nearly the totality of the available AI Engines on the AMD Versal VCK5000 accelerator. We also show that it achieves up to 25.7$\times$ speedup over a traditional FPGA programmable logic implementation using the Vitis Quantitive Finance Library, a specialized AMD library for quantitative finance on FPGAs, and up to 13.41$\times$ speedup over a highly parallelized CPU implementation over 128 threads. We show that although an Nvidia RTX A6000 GPU implementation has 0.73$\times$ the execution time of our design, our design achieves 1.82$\times$ the energy efficiency.

---

Contributors:

• Michael Samet