GPU-accelerated molecular dynamics: State-of-art software performance and porting from Nvidia CUDA to AMD HIP

Abstract

Classical molecular dynamics (MD) calculations represent a significant part of the utilization time of high-performance computing systems. As usual, the efficiency of such calculations is based on an interplay of software and hardware that are nowadays moving to hybrid GPU-based technologies. Several well-developed open-source MD codes focused on GPUs differ both in their data management capabilities and in performance. In this work, we analyze the performance of LAMMPS, GROMACS and OpenMM MD packages with different GPU backends on Nvidia Volta and AMD Vega20 GPUs. We consider the efficiency of solving two identical MD models (generic for material science and biomolecular studies) using different software and hardware combinations. We describe our experience in porting the CUDA backend of LAMMPS to ROCm HIP that shows considerable benefits for AMD GPUs comparatively to the OpenCL backend.

Molecular dynamics analysis of elastic properties and new phase formation during amorphous ices transformations

“…Positions of particles are determined from the solution of the classical Newtonian equations of motion. The simulations are carried out for a system of 2880, 23040 and 77760 water molecules, in a cubic box with periodic boundary conditions using the LAMMPS software package 39 with the GPU acceleration for TIP4P models 40 , 41 . The cutoff radii for the Coulomb and Lennard-Jones interactions are and , respectively.…”

FPGA Accelerator for Machine Learning Interatomic Potential-Based Molecular Dynamics of Gold Nanoparticles

“…One such snippet of code with the applied directives is provided below. Table (1) shows the latency and resource utilization of the above code snippet with and without array partitioning directive. It is clear from this Table that using directives would improve latency and this comes with the increase in hardware.…”

“…With several advancements in the electronic hardware industry, High Performance Computing (HPC) applications utilize HPC servers. These servers use several conventional processors and graphical processing units (GPUs) [1]- [3]. These advancements play a significant role in increasing the computational power which is very essential for many critical HPC applications.…”

Predicting Single-Substance Phase Diagrams: A Kernel Approach on Graph Representations of Molecules

“…Despite a wider availability of GPU-accelerated software package and directive-based programming models, 38 state-ofthe-art GPU solvers still entail the usage of assembly like languages and warp-level intrinsic to obtain the best possible performance. 39 Recently, Tang et al developed the GraphDot software package for the marginalized graph kernel to overcome the aforementioned challenges. 40,41 The software builds on top of an advanced just-in-time code generation framework to create an optimized GPU code for node and edge features and feature-level microkernels that domain scientists as users can easily create using a high-level Python API.…”

“…Previous works were also constrained to use graphs with simple node and edge features due to the difficulty in implementing sophisticated feature-level microkernels. , These problems have been further exacerbated in this era of heterogeneous computing as the general-purpose GPUs are still far from friendly in terms of programmability for domain scientists. Despite a wider availability of GPU-accelerated software package and directive-based programming models, state-of-the-art GPU solvers still entail the usage of assembly like languages and warp-level intrinsic to obtain the best possible performance …”