Employing OpenCL to Accelerate Ab Initio Calculations on Graphics Processing Units.

Area of Science:

• Computational chemistry
• High-performance computing
• Materials science

Background:

• Quantum chemistry calculations are computationally intensive.
• Graphics Processing Units (GPUs) accelerate these calculations.
• Existing GPU acceleration often relies on NVIDIA's CUDA platform.

Purpose of the Study:

• To extend a GPU-accelerated quantum chemistry package to support OpenCL.
• To evaluate the performance of AMD GPUs for quantum chemical computations.
• To compare AMD GPU performance against NVIDIA GPUs using CUDA.

Main Methods:

• Implementation of OpenCL compute kernels for a quantum chemistry package.
• Execution of hybrid density functional theory calculations.
• Performance benchmarking on AMD GPUs and comparison with NVIDIA GPUs.

Main Results:

• OpenCL kernels enable execution on diverse hardware, including CPUs and AMD GPUs.
• AMD GPUs demonstrate performance comparable to or exceeding NVIDIA GPUs for the tested calculations.
• The study validates AMD GPUs as a practical option for accelerating quantum chemistry.

Conclusions:

• The OpenCL-based extension provides broader hardware accessibility for GPU-accelerated quantum chemistry.
• AMD GPUs present a competitive and viable alternative to NVIDIA GPUs in this domain.
• This work expands the options for researchers seeking to leverage GPU computing power.