high performance computing on graphics processing units: hgpu.org

Nowadays, risk analysis and management is a core part of the daily operations in the financial industry, and strictly enforced by regulatory agencies. At the same time, large financial corporations have started migrating their operations into cloud services. Since the latter use a pay-per-use business model, there is a real need for implementations with high performance and energy efficiency, in order to reduce the overall cost of operation. And this is exactly where field programmable gate arrays (FPGAs) have a great advantage as accelerators, which enables the designer to efficiently offload the compute-intensive parts of the workload from the host central processing unit (CPU). The advantages associated to FPGAs have recently conferred them a place in cloud services, such as the Amazon Elastic Compute Cloud (EC2). In this work, we concentrate on widely-employed portfolio risk measures, namely value-at-risk and expected shortfall, by means of compute-intensive nested Monte Carlo-based simulations. As a proofof-concept, we have implemented and assessed the performance of these simulations on a representative complex portfolio, where the workload is efficiently distributed between CPU and FPGA on an Amazon EC2 F1 instance. Due to a lack of direct power measurement capabilities in the cloud, energy efficiency cannot be properly assessed at the moment. Nevertheless, our FPGA implementation achieves a 10x speedup on the compute intensive part of the code, compared to an optimized parallel implementation on multicore CPU, and it delivers a 3.5x speedup at application level for the given setup.