Acceleration of coarse grain molecular dynamics on GPU architectures

Area of Science:

• Computational chemistry
• Molecular dynamics simulations
• Parallel computing

Background:

• Coarse-grained (CG) molecular models offer reduced computational cost for simulating complex systems over longer timescales compared to atomistic models.
• Accelerating CG simulations on parallel architectures like Graphics Processing Units (GPUs) introduces unique challenges requiring careful evaluation.
• Understanding the interplay between CG model characteristics and parallel performance is essential for advancing simulation capabilities.

Purpose of the Study:

• To investigate and characterize how specific features of coarse-grained molecular models influence parallel simulation performance on GPUs.
• To identify key algorithmic and system-specific factors that impact the speed-up and accuracy of GPU-accelerated CG molecular dynamics.

Main Methods:

• Implementation of a GPU-accelerated coarse-grained molecular dynamics simulator.
• Application of specialized optimizations for CG models, including dedicated data structures for diverse bead interactions.
• Performance evaluation across three distinct GPU architectures, using the NVIDIA GTX480 (Fermi) as a case study.

Main Results:

• Achieved a maximum speed-up factor of 14 on an NVIDIA GTX480 GPU.
• Demonstrated significant performance gains through tailored optimizations for CG models on parallel architectures.
• Provided a comprehensive analysis of how CG model features affect simulation speed-up and result accuracy.

Conclusions:

• GPU acceleration offers substantial performance benefits for coarse-grained molecular dynamics simulations.
• Optimized data structures and algorithms are critical for maximizing speed-up and maintaining accuracy in parallel CG simulations.
• The findings provide valuable insights for developing efficient GPU-accelerated CG simulation methodologies for complex systems.