high performance computing on graphics processing units: hgpu.org

Protein and RNA biomolecular folding and assembly problems have important applications because misfolding events are associated with diseases like Alzheimer’s and Parkinson’s. However, simulating biologically relevant sized biomolecules on timescales that correspond to biological functions is an extraordinary challenge due to computational bottlenecks that are mainly involved in force calculations. We briefly review the molecular dynamics algorithm and highlight the main bottlenecks, which involve the calculation of the forces that interact between its substituent particles. We then present novel GPU-specific performance optimization techniques for MD simulations, including 1) a new Verlet-type neighbor list algorithm that is readily implemented using the CUDPP library and 2) data type compression scheme, as well as standard GPU-optimization techniques such as parallel random number generator and floating point operation issues. These and other GPU performance optimizations were applied to coarse-grained MD simulations of the ribosome, a protein-RNA molecular machine for protein synthesis composed of 10,219 residues and nucleotides. We observe a size-dependent speedup of the simulation code with over 30x speedup over the CPU-optimized approach for the full ribosome when all optimizations are taken into account.