Raising the Performance of the Tinker-HP Molecular Modeling Package [Article v1.0]

Abstract

This living paper reviews the present High Performance Computing (HPC) capabilities of the Tinker-HP molecular modeling package. We focus here on the reference, double precision, massively parallel molecular dynamics engine present in Tinker-HP and dedicated to perform large scale simulations. We show how it can be adapted to recent Intel ® Central Processing Unit (CPU) petascale architectures. First, we discuss the new set of Intel ® Advanced Vector Extensions 512 (Intel AVX-512) instructions present in recent Intel processors (e.g., the Intel ® Xeon ® Scalable and Intel ® Xeon Phi ™ 2nd generation processors) allowing for larger vectorization enhancements. These instructions constitute the central source of potential computational gains when using the latest processors, justifying important vectorization efforts for developers. We then briefly review the organization of the Tinker-HP code and identify the computational hotspots which require Intel AVX-512 optimization and we propose a general and optimal strategy to vectorize those particular parts of the code. We present our optimization strategy in a pedagogical way so it can benefit other researchers interested in improving performances of their own software. Finally we compare the performance enhancements obtained to unoptimized code, both sequentially and at the scaling limit in parallel for classical non-polarizable (CHARMM) and polarizable force fields (AMOEBA).

High-resolution mining of the SARS-CoV-2 main protease conformational space: supercomputer-driven unsupervised adaptive sampling

“…The results presented here benefit from a GPU acceleration in the newly developed Tinker-HP GPUs code 22 that was first used here for COVID-19 simulations. However the procedure is completely general and can be applied to any homogeneous or heterogeneous computational platforms compatible with Tinker-HP 21,27 or any MD software. Therefore, in view of the particular distribution of available numerical resources, the simulations are organized by iterations as follows.…”

Tinker-HP: Accelerating Molecular Dynamics Simulations of Large Complex Systems with Advanced Point Dipole Polarizable Force Fields Using GPUs and Multi-GPU Systems

“…At first, our approach was designed to directly offload the CPU vectorized real space compute kernels that use small arrays to compute pairwise interactions in hopes of aligning the memory access pattern at the vector level and therefore accelerate the code. 22 This requires each gang to privatize every temporary array and results in a significant overhead with memory reservation associated with a superior bound on the gang’s number. Making interactions computation scalar helps us remove those constraints and double the kernel performance.…”

“…The code was shown to be really efficient, being able to scale on up to tens of thousand cores on modern petascale supercomputers. 20 , 22 Recently, it has been optimized on various platforms taking advantage of vectorization and of the evolution of the recent CPUs (Central Processing Units). 22 However, in the last 15 years, the field has been increasingly using GPUs (Graphic Processor Units) 23 − 25 taking advantage of low precision arithmetic.…”

“… 20 , 22 Recently, it has been optimized on various platforms taking advantage of vectorization and of the evolution of the recent CPUs (Central Processing Units). 22 However, in the last 15 years, the field has been increasingly using GPUs (Graphic Processor Units) 23 − 25 taking advantage of low precision arithmetic. Indeed, such platforms offer important computing capabilities at both low cost and high energy efficiency allowing for reaching routine microsecond simulations on standard GPU cards with pair potentials.…”

High-Resolution Mining of SARS-CoV-2 Main Protease Conformational Space: Supercomputer-Driven Unsupervised Adaptive Sampling

“…The results that will be presented here benefit from a GPU acceleration thanks to the newly developed Tinker-HP GPUs code 23 that was first used here for production to handle COVID-19 simulations. However the procedure is completely general and can be applied to any homogeneous or heterogenous computational platforms compatible with Tinker-HP 22,28 or any MD software. Therefore, in view of the particular distribution of available numerical resources, the simulations are organized by iterations as follows.…”