Multi-dimensional characterization of electrostatic surface potential computation on graphics processors.

Area of Science:

• Computational chemistry
• Biophysics
• Molecular modeling

Background:

• Electrostatic surface potential (ESP) calculations are vital for understanding biomolecular function.
• The quadratic computational complexity of ESP calculations limits their application.
• Existing methods focus on algorithmic or hardware improvements to reduce complexity.

Purpose of the Study:

• To accelerate the calculation of electrostatic surface potential (ESP).
• To leverage a multi-scale approximation algorithm, hierarchical charge partitioning (HCP), on graphics processing units (GPUs).
• To investigate the combined effect of HCP and GPU acceleration for biomolecular modeling.

Main Methods:

• Applied hierarchical charge partitioning (HCP), a multi-scale approximation algorithm, to ESP calculations.
• Mapped the HCP algorithm onto graphics processing units (GPUs), utilizing natural biomolecular partitioning.
• Incorporated GPU optimization techniques such as shared memory and amortized divergent branching.

Main Results:

• Achieved a 25-fold acceleration of ESP calculations through GPU parallelization alone.
• The combination of GPU and HCP yielded speedups of up to 1,860-fold for large molecular structures.
• These speedups were benchmarked against a highly optimized CPU implementation.

Conclusions:

• The developed method significantly accelerates ESP calculations.
• GPU acceleration combined with HCP provides substantial performance gains.
• The accuracy of ESP calculations is maintained with the GPU-accelerated HCP approach.