A hybrid CPU/GPU method for Hartree-Fock self-consistent-field calculation.

Area of Science:

• Computational chemistry
• Quantum chemistry
• High-performance computing

Background:

• Two-electron repulsion integrals (ERIs) are fundamental to Hartree-Fock calculations.
• Both CPUs and GPUs have distinct advantages for ERI computation based on angular momentum.

Purpose of the Study:

• To develop and evaluate a hybrid CPU/GPU method for accelerating ERI calculation and Fock matrix generation.
• To leverage the strengths of both CPUs and GPUs for enhanced computational efficiency.

Main Methods:

• A hybrid CPU/GPU approach utilizing a task queue with OpenMP dynamic scheduling.
• Tasks are prioritized based on angular momentum: low for GPUs, high for CPUs.
• Overlapping CPU and GPU computations to maximize resource utilization.

Main Results:

• The hybrid method demonstrated superior efficiency compared to GPU-only methods with a single GPU.
• The advantage of the hybrid method decreased with an increasing number of GPUs.
• While scaling exponents were slightly higher, the hybrid method showed a significantly lower pre-exponent factor, indicating overall effectiveness.

Conclusions:

• The hybrid CPU/GPU method offers an effective strategy for accelerating ERI calculations, particularly in single-GPU environments.
• This approach enables computations with arbitrary angular momentum, enhancing flexibility.
• Further research may be needed to optimize performance in multi-GPU systems.