Technical Report: SPIDAL Summer REU 2016: Distance Array and RMSD Speed-Up in MDAnalysis

2016-09-12T23:29:53Z (GMT) by Robert Delgado Oliver Beckstein
MDAnalysis (mdanalysis.org) is an open source library that allows for the analysis of molecular dynamics simulations. Currently, computation of the distance_array and rmsd for each frame in a molecular dynamics trajectory in MDAnalysis is a computationally intensive process.  The goal of this REU was to speed up parts of MDAnalysis. We created a GPU based program through PyCUDA that allows us to compute distance arrays. PyCUDA (https:// mathema.tician.de/software/pycuda/) is a package that allows us to access NVIDIA’s CUDA API via python. Unfortunately, even though that it was evident that GPU’s do compute distance arrays faster than CPU’s, the transfer time between the local memory and the GPU is long enough such CPU computation is still faster. We also made slight modifications to the rmsd calculations to improve speeds by up to 30%.