Maximizing resource usage in multifold molecular dynamics with rCUDA

The full-understanding of the dynamics of molecular systems at the atomic scale is of great relevance in the fields of chemistry, physics, materials science, and drug discovery just to name a few. Molecular dynamics (MD) is a widely used computer tool for simulating the dynamical behavior of molecules. However, the computational horsepower required by MD simulations is too high to obtain conclusive results in real-world scenarios. This is mainly motivated by two factors: (1) the long execution time required by each MD simulation (usually in the nanoseconds and microseconds scale, and beyond) and (2) the large number of simulations required in drug discovery to study the interactions between a large library of compounds and a given protein target. To deal with the former, graphics processing units (GPUs) have come up into the scene. The latter has been traditionally approached by launching large amounts of simulations in computing clusters that may contain several GPUs on each node. However, GPUs are targeted as a single node that only runs one MD instance at a time, which translates into low GPU occupancy ratios and therefore low throughput. In this work, we propose a strategy to increase the overall throughput of MD simulations by increasing the GPU occupancy through virtualized GPUs. We use the remote CUDA (rCUDA) middleware as a tool to decouple GPUs from CPUs, and thus enabling multi-tenancy of the virtual GPUs. As a working test in the drug discovery field, we studied the binding process of a novel flavonol to DNA with the GROningen MAchine for Chemical Simulations (GROMACS) MD package. Our results show that the use of rCUDA provides with a 1.21× speed-up factor compared to the CUDA counterpart version while requiring a similar power budget.