Semiempirical Quantum Chemical Calculations Accelerated on a Hybrid Multicore CPU-GPU Computing Platform.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• High-Performance Computing

Background:

• Semiempirical quantum chemical methods offer a balance between accuracy and computational cost for molecular modeling.
• Traditional implementations on central processing units (CPUs) can be computationally intensive, limiting the size and complexity of treatable systems.
• Hybrid multicore CPU-GPU architectures present an opportunity for accelerating demanding computational tasks.

Purpose of the Study:

• To investigate the potential of graphics processing units (GPUs) for accelerating semiempirical quantum chemical calculations.
• To identify and optimize computationally bottlenecked routines within semiempirical methods for GPU execution.
• To quantify the performance improvements achieved on a hybrid CPU-GPU platform.

Main Methods:

• Systematic profiling of semiempirical calculations (MNDO, AM1, PM3, OM1, OM2, OM3) on diverse test systems (fullerenes, water clusters, solvated crambin).
• Porting time-consuming code sections to the GPU.
• Optimization using existing GPU libraries and custom GPU kernels for pseudodiagonalization (Jacobi transformations).
• Benchmarking single-point energy calculations and geometry optimizations on a hybrid CPU-GPU system.

Main Results:

• Identification of key computational bottlenecks in semiempirical methods.
• Successful implementation of GPU-accelerated routines for pseudodiagonalization.
• Achieved a significant reduction in computation times, up to one order of magnitude.
• Demonstrated consistent speedups across all tested semiempirical model Hamiltonians and system types.

Conclusions:

• Leveraging GPUs as coprocessors on hybrid platforms dramatically accelerates semiempirical quantum chemical calculations.
• The developed GPU acceleration provides a substantial performance enhancement for large molecule simulations.
• This approach enables more efficient and feasible computational chemistry studies using semiempirical methods.