Open-Source Multi-GPU-Accelerated QM/MM Simulations with AMBER and QUICK.

Area of Science:

• Computational Chemistry
• Biomolecular Simulations
• Quantum Mechanics/Molecular Mechanics (QM/MM)

Background:

• The QM/MM approach is a cornerstone in computational chemistry for studying biomolecular systems.
• Efficient electronic structure calculations are crucial for accurate QM/MM simulations.
• Leveraging GPU acceleration can significantly enhance computational performance.

Purpose of the Study:

• To integrate the QUantum Interaction Computational Kernel (QUICK) program as an electronic structure engine within the AMBER framework for QM/MM simulations.
• To provide users with a free, GPU-accelerated option for QM/MM calculations.
• To evaluate the performance and energy conservation of the integrated QM/MM approach.

Main Methods:

• Integration of the open-source, GPU-accelerated QUICK program with the AMBER simulation package via file-based interface (FBI) and application programming interface (API).
• Implementation of multi-GPU parallelization for QUICK.
• Investigation of energy conservation in microcanonical ensemble simulations.
• Benchmarking performance using N-methylacetamide (NMA) and the photoactive yellow protein (PYP) in bulk water.

Main Results:

• Demonstrated successful integration of QUICK with AMBER for QM/MM simulations.
• Achieved significant speedups using GPU acceleration, with up to 53x faster performance compared to a single CPU core.
• Observed speedups of up to 2.6x when using four GPUs versus a single GPU.
• Reported up to 3.5x speedup when utilizing the API compared to the FBI.

Conclusions:

• The integration of QUICK with AMBER provides a powerful and efficient tool for QM/MM simulations.
• GPU acceleration offers substantial performance gains, making complex biomolecular simulations more accessible.
• The API interface further enhances computational efficiency over the file-based interface.