Related Experiment Video
Updated: Aug 1, 2025

Computation of Atmospheric Concentrations of Molecular Clusters from ab initio Thermochemistry
Published on: April 8, 2020
Porting fragmentation methods to GPUs using an OpenMP API: Offloading the resolution-of-the-identity second-order
Buu Q Pham1, Laura Carrington2, Ananta Tiwari2
1Department of Chemistry and Ames Laboratory, Iowa State University, Ames, Iowa 50011, USA.
This study accelerates the resolution-of-the-identity second-order Møller-Plesset perturbation (RI-MP2) method using graphical processing units (GPUs). The optimized RI-MP2 method shows significant speedups for large molecular calculations in both standalone and effective fragment molecular orbital frameworks.
Area of Science:
- Computational Chemistry
- Quantum Chemistry
- High-Performance Computing
Background:
- The resolution-of-the-identity second-order Møller-Plesset perturbation (RI-MP2) method is crucial for accurate electronic structure calculations.
- Scaling limitations of RI-MP2 hinder its application to large molecular systems.
- Graphical Processing Units (GPUs) offer massive parallelism for accelerating computationally intensive tasks.
Purpose of the Study:
- To accelerate the RI-MP2 method by off-loading computations onto GPUs.
- To integrate GPU-accelerated RI-MP2 into the GAMESS electronic structure program and the effective fragment molecular orbital (EFMO) framework.
- To evaluate the performance and scalability of the GPU-accelerated RI-MP2 method.
Main Methods:
- Implementation of RI-MP2 calculations on GPUs using OpenMP Application Programming Interface.
- Development of a novel scheme to maximize GPU data digestion and linearize CPU-GPU data transfer.
- Interfacing GAMESS Fortran code with NVIDIA cuBLAS and cuSOLVER libraries for efficient matrix operations.
- Utilizing NVIDIA V100 GPUs and IBM P9 CPUs for computational acceleration.
Main Results:
- Standalone GPU RI-MP2 achieved speedups of up to 7.5× on fullerenes.
- A single Summit node (6 V100s) computed a 175-water molecule cluster's RI-MP2 energy in ~0.85 hours.
- GPU RI-MP2 in the EFMO framework demonstrated near-linear scaling with a large number of GPUs for a complex nanoparticle system.
- High parallel efficiencies (98.0% and 96.1%) were observed with 2304 and 4608 V100s, respectively.
Conclusions:
- GPU acceleration significantly enhances the performance of the RI-MP2 method for large-scale electronic structure calculations.
- The developed GPU RI-MP2 implementation is efficient both as a standalone method and within the EFMO framework.
- This work enables more accurate and efficient computational studies of larger and more complex molecular systems.
Related Concept Videos
Mechanistic Models: Compartment Models in Algorithms for Numerical Problem Solving
In individual population analyses, different algorithms are employed, such as Cauchy's method, which uses a...
Parallel Processing
Extraction: Partition and Distribution Coefficients
For extracting a solute from an aqueous phase into an...

