Speed-Up Computational Finance Simulations with OpenCL on Intel Xeon Phi.

Computational finance is a domain where performance is in high demand. In this work, we investigate the suitability of Intel Xeon Phi for computational finance simulations. Specifically, we use a scenario based ALM (Asset Liability Management) model and propose a novel OpenCL implementation for Xeon Phi. To further improve the performance of the application, we apply several optimization techniques (data layout and data locality improvement, loop unrolling) and study their effects. Our results show that the optimized OpenCL code deployed on the Phi can run up to 135x faster than the original scalar code running on an Intel i7 GPP. Furthermore, we also show that choosing the optimal work-item/work-group distribution has a compelling effect on massively parallel and heavily-branching code. Overall, these results are significant for the computational finance specialists, as they enable a major increase in model accuracy, because 10x more simulations can be performed in less than a 10th of the original time.