Multi-GPU RI-HF Energies and Analytic Gradients─Toward High-Throughput Ab Initio Molecular Dynamics.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Molecular Dynamics

Background:

• Accurate calculation of molecular energies and gradients is crucial for simulating chemical processes.
• Ab initio molecular dynamics (AIMD) simulations require computationally intensive electronic structure calculations.
• Existing methods for Hartree-Fock (HF) calculations, especially on graphics processing units (GPUs), face challenges in scalability and efficiency.

Purpose of the Study:

• To develop and implement an optimized algorithm for resolution-of-the-identity Hartree-Fock (RI-HF) calculations and analytic gradients.
• To leverage multi-GPU parallelism for high-throughput AIMD simulations of small to medium-sized molecules (10-100 atoms).
• To achieve significant performance improvements over existing GPU-accelerated and traditional HF methods.

Main Methods:

• Developed a novel algorithm exploiting multi-GPU parallelism and efficient workload balancing.
• Implemented techniques for symmetry utilization, integral screening, and leveraging sparsity for memory optimization.
• Applied the algorithm to ab initio molecular dynamics simulations.

Main Results:

• Achieved over 3x speedup in single GPU AIMD throughput compared to previous GPU-accelerated RI-HF and traditional HF methods.
• Demonstrated superlinear speedup with multiple GPUs when additional memory enabled storage of decompressed three-center integrals.
• Efficiently distributed computational tasks across multiple GPUs.

Conclusions:

• The optimized multi-GPU RI-HF algorithm significantly enhances the efficiency of AIMD simulations.
• The implementation offers substantial performance gains, making larger-scale molecular simulations more feasible.
• Multi-GPU utilization presents a promising avenue for accelerating quantum chemistry calculations.