high performance computing on graphics processing units: hgpu.org

The Intel Xeon Phi coprocessor is a PCI Express form factor card designed to work in tangent with Intel Xeon processors in order to allow faster execution of highly parallelizable code. Efficient execution of highly parallel applications is achieved through the use of many smaller, lower clock speed cores; allowing for many more simultaneous execution threads and wider vector registers. While many applications will continue to run faster on Intel Xeon processors, certain applications are well suited to take advantage of many simultaneous concurrent execution threads as well wider vectorization. In this article, the parallelization of a Computational Fluid Dynamics (CFD) code is achieved by offloading the implicit coefficients to an Intel Xeon Phi coprocessor, where the system of equations is solved using the vectorizeable BiConjugate Gradient Stabilized (BiCGStab) iterative linear solver. The same code is also run purely on the host machine featuring an overclocked Intel Xeon processor, and the performance is compared for a variety of parameters.