Computational fluid dynamics (CFD) is a cornerstone in the understanding of many scientific and technological areas such as in marine offshore renewable energy (ORE) technologies, aero-space/defence, automotive/motorsport, and others. CFD software applications incur huge computational workloads, and therefore without the use of expensive and energy-intensive High-Performance Computing (HPC) platforms, running these applications becomes impractical. Current state-of-the-art general purpose CFD software routines are often inefficient, and many are unable to utilize more than 5% of the available computing power of modern computer systems. 

The aim of this project is to improve the execution time of the sparse matrix solver of the leading open-source CFD library (OpenFOAM). This is the performance bottleneck of many modelling applications currently employed in ORE sector, automotive/motorsport and others. The optimization will target contemporary x64 CPUs and/or Nvidia GPUs. The student will investigate/evaluate various sparse matrix memory formats and optimize the code by using parallel programming frameworks such as CUDA, MPI, OpenMP, AVX intrinsics and others.

This PhD project is also supported by an industrial partner (Engys). Engys Ltd is a leading provider of open-source CFD software for the enterprise sector and developer of the OpenFOAM-based HELYX software suite. This ambitious and pioneering PhD project necessitates close cooperation between HPC experts, CFD domain scientists and engineers. At least two real-world applications (test cases) will be considered, including industrial applications and benchmarks from the CCP-WSI EPSRC project.

Non-UK applicants are also welcome to apply as the international fees may be waived for exceptional candidates.

Apply here
https://www.jobs.ac.uk/job/DFL177/phd-studentship-accelerating-sparse-matrix-solvers-for-high-performance-computing