HPC in Quant Finance: Leveraging AI/ML Revolution in Graph-Based Computation for Risk Management of Exotics on NVIDIA GPUs, Grace Hopper and Triton Inference Server

, VP, Quantitative Research, JPMorgan Chase
, VP, Quantitative Research, JP Morgan Chase
, VP, Machine Learning Center of Excellence, JPMorgan Chase
Monte Carlo methods are widely used to value and risk-manage complex financial securities. However, Monte Carlo (MC) methods can become computationally very expensive when the precision requirement is high, as is often the case in financial applications. At JP Morgan Chase & Co., we leverage thousands of NVIDIA GPUs for real-time pricing and overnight risk estimation of tens of thousands of exotics and structured derivatives.

We'll explain how we are transforming our Monte Carlo pricing engine by leveraging TensorFlow on GPU to express complex payoffs as computational graphs: (1) the use of TensorFlow and GPU make the evaluation of payoff on each MC path more efficient; and (2) TensorFlow’s AAD (Adjoint Algorithmic differentiation) allows the estimation of sensitivities in O(1) complexity on GPUs. We'll also describe how we have developed a client-server architecture powered by NVIDIA Triton Inference Server.
活动: GTC 24
日期: March 2024
级别: 高级技术
话题: AI 推理
行业: 金融服务
NVIDIA 技术: Triton
语言: 英语
所在地: