A comparison between parallelization approaches in molecular dynamics simulations on GPUs

Abstract

We test the relative performances of two different approaches to the computation of forces for molecular dynamics simulations on graphics processing units. A "vertex-based" approach, where a computing thread is started per particle, is compared to an "edge-based" approach, where a thread is started per each potentially non-zero interaction. We find that the former is more efficient for systems with many simple interactions per particle while the latter is more efficient if the system has more complicated interactions or fewer of them. By comparing computation times on more and less recent graphics processing unit technology, we predict that, if the current trend of increasing the number of processing cores--as opposed to their computing power--remains, the "edge-based" approach will gradually become the most efficient choice in an increasing number of cases.

Introducing improved structural properties and salt dependence into a coarse-grained model of DNA

“…The introduction of different strengths for the AA and TT stacking in the model allows us to better capture these effects and in addition gives us more accurate sequence-dependent hairpin thermodynamics and kinetics, which are currently being exploited to further study hairpins. We note that a simulation code implementing both the original oxDNA model and the new oxDNA2 model, including an implementation using graphical processing units (GPUs), 85 is available for download from dna.physics.ox.ac.uk.…”

A nucleotide-level coarse-grained model of RNA

“…This example shows that structural investigations of systems of the order of hundreds of bases are well within the capabilities of the oxRNA model using a single CPU, and if multiple CPUs are used, or a GPU chip is used, 94 then structural properties and fluctuations around equilibrium can be studied for systems on the order of thousands of base pairs, as can be done for oxDNA. 68 In Ref.…”

RUMD: A general purpose molecular dynamics package optimized to utilize GPU hardware down to a few thousand particles

“…Rovigatti et al have recently discussed the possible advantages of "vertex-based" (atom-decomposition [54], one thread per particle) versus "edge-based" (force-decomposition [54], one thread per interaction) parallelism [55]. Our approach includes the former and a range of intermediate cases, while not taking it to the extreme of one thread per interaction.…”