GPU Acceleration of Large-Scale Full-Frequency GW Calculations.

Area of Science:

• Computational Physics
• Quantum Chemistry
• Materials Science

Background:

• Many-body perturbation theory, specifically the GW approximation, is crucial for simulating electronic excitations.
• Density Functional Theory (DFT) provides the starting point for GW calculations.
• Scaling GW calculations to high-performance computing (HPC) systems is essential for tackling larger, more complex systems.

Purpose of the Study:

• To investigate and implement GPU acceleration for the full-frequency GW method within the WEST code.
• To enhance the computational efficiency and scalability of GW calculations on modern HPC architectures.
• To demonstrate the feasibility of large-scale GW simulations for realistic material systems.

Main Methods:

• Utilized optimized GPU libraries (cuFFT, cuBLAS) for accelerated computations.
• Implemented a hierarchical parallelization strategy to minimize data transfer between CPUs and GPUs.
• Employed nonblocking MPI communications and mixed precision techniques to maximize computational throughput.
• Conducted performance benchmarks on leadership HPC systems.

Main Results:

• Achieved substantial speedups for the GPU-accelerated WEST code compared to its CPU-only version.
• Demonstrated excellent strong and weak scaling performance across a large number of GPUs (up to 25,920).
• Successfully performed large-scale GW calculations on systems with up to 10,368 valence electrons.

Conclusions:

• GPU acceleration significantly enhances the performance and scalability of the full-frequency GW method.
• The WEST code, with GPU acceleration, is capable of performing large-scale electronic excitation calculations for complex material systems.
• This work paves the way for more extensive theoretical investigations of electronic properties in molecules and materials.