Accelerating Phylogenetic Inference on Heterogeneous OpenCL Platforms

MrBayes is a popular software package for Bayesian phylogenetic inference that is used to derive an evolutionary tree for a collection of species whose DNA sequences are known. At the high pace which biological data has been accumulating over the years, there has been a huge growth in the computational challenges required by this type of applications. To overcome this issue, researchers turned to parallel computing to speedup execution, for instance by using Graphics Processing Units (GPUs). At the same time, GPUs architectures of different manufacturers evolved, presenting more and more computing power. Additionally, parallel programming frameworks became more mature providing more features to programmers to exploit parallelism within GPUs. In this work, we parallelized the MrBayes 3.2 in order to accelerate and reduce the execution time using the Open Computing Language (OpenCL) programming framework. Furthermore, we studied the performance of MrBayes execution using different computing platforms and different GPUs architectures of both NVIDIA and AMD vendors to determine the best architecture for this application. Results showed that even with GPUs with similar computing power NVIDIA's obtained better performance when compared to AMD's, with the later providing an unexpected low performance. Moreover, results also showed that for this particular application, NVIDIA architectural advances over the years provide limited performance improvement.