A Performance Study of Quantum ESPRESSO’s PWscf Code on Multi-core and GPU Systems

Abstract

We describe the porting of PWscf (Plane-Wave Self Consistent Field), a key component of the Quantum ESPRESSO open-source suite of codes for materials modeling, to GPU systems using CUDA Fortran. Kernel loop directives (CUF kernels) have been extensively used in order to have a single source code for both CPU and GPU implementations. The results of the GPU version have been carefully validated and the performance of the code on several GPU systems (both x86 and POWER8 based) has been compared with traditional Intel multi-core (CPU only) systems. This current GPU version can reduce the timeto-solution by an average factor of 2 − 3 running two different input cases widely used as benchmarks on small and large high performance computing systems.

Quantum ESPRESSO toward the exascale

“…In light of this limitation, a new port, dubbed QE-GPU, has been recently rewritten from scratch, starting from the case study presented by Romero et al 65 who ported v. 6.1 of pw.x to NVIDIA GPUs. 66 In Ref.…”

“…66 In Ref. 65, the authors detail a new strategy based on CUDA Fortran-the Fortran analogue of CUDA C-and demonstrate 2× to 3× speedup consistently achieved on a variety of platforms and using different benchmarks. The new GPU-enabled releases of QUANTUM ESPRESSO extend this work, but adopting a few design solutions to streamline future development and porting to other heterogeneous architectures, as detailed below.…”

Efficient Ab-Initio Molecular Dynamic Simulations by Offloading Fast Fourier Transformations to FPGAs

“…Accelerating these simulations has led to massively parallel computations using multicore processors and hardware accelerators. Popular AIMD simulation software packages such as Quantum Espresso [1] and CP2K [2] offer CUDA implementations to offload specific routines to GPUs. This scale of processing has led to maximizing performance through efficient computation.…”

VASP hits the memory wall: Processors efficiency comparison

“…Performance prediction for HPC systems is an important new field of science and engineering. The applications of idealized models are limited by the complexity of real‐life applications and the benchmarks of the real‐life cases serve as a complementary tool for the co‐design and optimization of software‐hardware combinations …”