GPU accelerated implementation of NCI calculations using promolecular density.

Area of Science:

• Computational Chemistry
• Molecular Modeling
• Bioinformatics

Background:

• Non-Covalent Interaction (NCI) analysis is crucial for understanding molecular interactions, especially in drug discovery and ligand-protein binding.
• Existing computational methods for NCI can be computationally intensive, limiting their application.
• The development of efficient computational tools is essential for advancing molecular modeling and chemical research.

Purpose of the Study:

• To present a custom implementation of the NCI approach utilizing promolecular density.
• To leverage NVIDIA Graphics Processing Unit (GPU) accelerators and the CUDA programming model for enhanced computational performance.
• To evaluate the performance and energy efficiency of the GPU-accelerated NCI method.

Main Methods:

• A custom NCI calculation code was developed using promolecular density.
• The implementation was optimized for NVIDIA GPUs using the CUDA programming model.
• Performance was assessed on 144 systems, comparing dual-GPU execution against an optimal 16-core CPU (OpenMP) implementation.
• Energy consumption was measured for both CPU and GPU-based calculations.

Main Results:

• The GPU-accelerated NCI approach drastically reduces computational time.
• A 39-fold performance improvement was observed for the largest systems on a dual-GPU setup compared to a 16-core CPU run.
• The GPU implementation demonstrates substantial energy savings compared to CPU-based calculations.

Conclusions:

• GPU acceleration is highly effective for NCI calculations, offering significant speedups.
• The developed CUDA implementation provides a computationally efficient and energy-saving tool for studying non-covalent interactions.
• This approach enhances the feasibility of detailed molecular interaction analysis in computational chemistry and drug design.