OpenCL on Xeon Phi paper is now available
Speed-Up Computational Finance Simulations with OpenCL on Intel Xeon Phi
Our work on accelerating scenario-based ALM models with OpenCL on Intel Xeon Phi processors is recently published in the Europar ’16 workshop proceedings.
We managed to submit and present a representative paper of our work on accelerating scenario-based ALM models with OpenCL on Intel Xeon Phi processors in the International Workshop on Multicore Software Engineering in August of 2016, in Grenoble. This month the conference proceedings were published thus so is our paper.
About the paper
The paper describes research we did in our mission to provide state-of-the-art risk management modeling in a split of second to private investors. Therefore, this work investigates the performance impact of two families of accelerators on computational finance simulations. Specifically, we use a scenario-based ALM (Asset Liability Management) model, we design and optimize its accelerated version using OpenCL, and test its performance gain when using Intel’s Xeon Phi and an NVIDIA GPU. Our comprehensive performance analysis includes performance results of the original scalar code, the OpenMP parallel ALM kernel, and the OpenCL version. Our results show that the optimized OpenCL code deployed on the Phi can run up to 135x faster than the original scalar code. Also, OpenCL can be up to 10x faster than the OpenMP implementation on the same Xeon Phi, but Xeon Phi is only 2–3x times faster than the regular CPU when using the same OpenCL code. Finally, the GPU outperforms Xeon Phi by almost an order of magnitude. We conclude that ALM is an excellent target for acceleration. Our results are significant for the computational finance domain as they enable a major increase in model accuracy.
This Europar ’16 contribution is available in the Conference proceedings or can be downloaded as a chapter at Springer. The earlier working paper version can be requested by email via: techlabs@-ortec-finance.com