GPU Linear Algebra Libraries and GPGPU Programming for Accelerating MOPAC Semiempirical Quantum Chemistry...

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Scientific Computing

Background:

• Semiempirical quantum chemistry methods, like MOPAC2009, are valuable for molecular modeling but can be computationally intensive.
• Accelerating these calculations is crucial for tackling larger and more complex molecular systems.

Purpose of the Study:

• To modify the MOPAC2009 code for faster single-point energy calculations (1SCF) on medium-size molecular systems.
• To leverage GPU coprocessors and multithreaded CPUs to enhance computational speed.

Main Methods:

• Integrated highly optimized linear algebra libraries (Intel MKL for CPU, MAGMA/CUBLAS for GPU).
• Optimized time-consuming steps: pseudodiagonalization, full diagonalization, and density matrix assembling.
• Implemented parallel (GPU-GPU) MOPAC code.

Main Results:

• Achieved significant speedups using optimized CPU linear algebra libraries.
• Demonstrated a 14x speedup for a 2400-atom methanol system.
• Observed 2.1x speedup with multithreaded CPUs and 3.8x with GPUs.

Conclusions:

• The optimized MOPAC2009 code enables efficient modeling of larger systems in inorganic chemistry, biochemistry, and materials science.
• Feasible application of semiempirical methods in lengthy molecular simulations using hybrid QM/MM and QM/QM potentials.
• Opens new avenues for computational research in diverse scientific fields.