A comparison of molecular dynamics simulations A comparison of molecular dynamics simulations using GROMACS with GPU and CPUusing GROMACS with GPU and CPU

Alex Camargo, Adriano Werhli, Karina Machado{alexcamargo, werhli, karina.machado}@furg.br

UNIVERSIDADE FEDERAL DO RIO GRANDE DO SULINSTITUTO DE INFORMÁTICA

ESCOLA GAÚCHA DE BIOINFORMÁTICA

Porto Alegre/2015

Why GPU? Why CPU?

Origin: FURG Destiny: UCLA

10.000 KM

–------->

LATENCY X THROUGHPUT

Airplane: 2 people, 400 km/h Lat: 25 hours, Thr: 0,08 people/hour

Ship: 40 people, 50 km/h Lat: 200 hours, Thr: 0,2 people/hour

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Overview

Mathematical modelling is central to systems and synthetic biology.

High-performance computing holds the key to making relevant biologically calculations.

Systems of up several million atoms are simulated for weeks or months prior to completion.

Graphics Processing Units (GPUs) technology brings supercomputing power to the desktop.

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Architecture: CPU x GPU

The reason behind the difference in capability between the CPU and the GPU is that the GPU is specialized for compute-intensive (highly parallel).

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Figure. Intel Xeon X5675 Figure. NVIDIA Quadro 600

Architecture: CPU x GPU

The reason behind the difference in capability between the CPU and the GPU is that the GPU is specialized for compute-intensive (highly parallel).

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Figure. CPU and GPU architecture

CUDA

CUDA is a hardware and programming model for issuing and managing computations on the GPU as a data-parallel computing device without the need of mapping them to a graphics API.

It was have implemented the proposed model using CUDA Toolkit 7.

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

CUDA

CUDA is a hardware and programming model for issuing and managing computations on the GPU as a data-parallel computing device without the need of mapping them to a graphics API.

It was have implemented the proposed model using CUDA Toolkit 7.

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Figure. CUDA particles simulation

Molecular Dynamics Simulation and GROMACS

Molecular modeling is an approach to understanding chemical and biomolecular systems. The visualization of atomic level details provides insight into the function and dynamics of structures.

In this work we use GROMACS version 5, since this software has native support for GPUs, in comparison by version 4 (OpenMP trheads native)

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Molecular Dynamics Simulation and GROMACS

Molecular modeling is an approach to understanding chemical and biomolecular systems. The visualization of atomic level details provides insight into the function and dynamics of structures.

In this work we use GROMACS version 5, since this software has native support for GPUs, in comparison by version 4 (OpenMP trheads native)

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Figure. Protein in water visualization

Performance Results

This work compares the use of the software suite GROMACS running entirely on a GPU and CPU.

Protein Lysozyme (PDB code 1AKI) in a box of water having 38.790 atoms.

The sequential MD simulations tests were executed on identical hardware with:

- Intel Xeon CPU X5675 - 3.059GHz (12 processor cores)- NVIDIA Quadro 600 GPU (96 CUDA cores)- RAM 12GB, HD 1TB - Ubuntu 14.04 64bit Desktop Edition

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Performance Results

This work compares the use of the software suite GROMACS running entirely on a GPU and CPU.

Protein Lysozyme (PDB code 1AKI) in a box of water having 38.790 atoms.

The sequential MD simulations tests were executed on identical hardware with:

- Intel Xeon CPU X5675 - 3.059GHz (12 processor cores)- NVIDIA Quadro 600 GPU (96 CUDA cores)- RAM 12GB, HD 1TB - Ubuntu 14.04 64bit Desktop Edition

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Figure. Computers used for MD simulations

Performance Results

Now, will be presented the performance comparison for main steps of Molecular Dynamics simulation.

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Performance Results:energy minimization

The structure is relaxed to ensure that the system has no steric clashes or inappropriate geometry.

nsteps: 5.000

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Performance Results:energy minimization

The structure is relaxed to ensure that the system has no steric clashes or inappropriate geometry.

nsteps: 5.000

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Figure. GPU utilization: Energy minimization

Figure. CPU utilization: Energy minimization

Performance Results:energy minimization

The structure is relaxed to ensure that the system has no steric clashes or inappropriate geometry.

nsteps: 5.000

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Figure. Comparison of runtimes: Energy minimization

Performance Results:equilibration (phase 1)

The structure is to brought to the temperature we wish to simulate and establish the proper orientation about the solute.

nsteps: 5.000

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Performance Results:equilibration (phase 1)

The structure is to brought to the temperature we wish to simulate and establish the proper orientation about the solute.

nsteps: 5.000

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Figure. GPU utilization: Equilibration (phase 1)

Figure. CPU utilization: Equilibration (phase 1)

Performance Results:equilibration (phase 1)

The structure is to brought to the temperature we wish to simulate and establish the proper orientation about the solute.

nsteps: 5.000

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Figure. Comparison of runtimes: Equilibration (phase 1)

Performance Results:equilibration (phase 2)

Prior to data collection, it is must also stabilize the pressure (and thus also the density) of the system.

nsteps: 5.000

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Performance Results:equilibration (phase 2)

Prior to data collection, it is must also stabilize the pressure (and thus also the density) of the system.

nsteps: 5.000

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Figure. GPU utilization: Equilibration (phase 2)

Figure. CPU utilization: Equilibration (phase 2)

Performance Results:equilibration (phase 2)

Prior to data collection, it is must also stabilize the pressure (and thus also the density) of the system.

nsteps: 5.000

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Figure. Comparison of runtimes: Equilibration (phase 2)

Performance Results:production MD

Upon completion of the two equilibration phases, we are ready to run production MD for data collection.

nsteps: 500.000

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Performance Results:production MD

Upon completion of the two equilibration phases, we are ready to run production MD for data collection.

nsteps: 500.000

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Figure. GPU utilization: Production MD

Figure. CPU utilization: Production MD

Performance Results:production MD

Upon completion of the two equilibration phases, we are ready to run production MD for data collection.

nsteps: 500.000

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Figure. Comparison of runtimes: Production MD

Discussion

Here we discuss how molecular dynamics simulations can be benefit with the computing power of GPUs.

Simulations of proteins on graphics hardware using CUDA is attractive.

As future work intend to run molecular dynamics simulations with different proteins and other hardware settings.

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

Acknowledgment

This research was supported by CAPES (Coordenação de Aperfeiçoamento de Pessoal de Nível Superior).

Bioinformática Estrutural de Proteínas: modelos, algoritmos e aplicações biotecnológicas.

UFMG/UFPB/FURG

www.biologiacomputacional.c3.furg.br

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

References[1] Hess, Berk, et al. "GROMACS 4: algorithms for highly efficient, load-balanced, and scalable molecular simulation." Journal of chemical theory and computation 4.3 (2008): 435-447.

[2] Liu, Weiguo, et al. "Molecular dynamics simulations on commodity GPUs with CUDA." High Performance Computing–HiPC 2007. Springer Berlin Heidelberg, 2007. 185-196.

[3] NVIDIA, CUDA. "CUDA Toolkit Documentation - v7.0 ". Available: https://docs.nvidia.com/cuda/. (2015).

[4] Pronk, Sander, et al. "GROMACS 4.5: a high-throughput and highly parallel open source molecular simulation toolkit." Bioinformatics (2013): btt055.

[5] Rodrigues, Christopher I., et al. "GPU acceleration of cutoff pair potentials for molecular modeling applications." Proceedings of the 5th conference on Computing frontiers. ACM, 2008.

[6] Vouzis, Panagiotis D., and Nikolaos V. Sahinidis. "GPU-BLAST: using graphics processors to accelerate protein sequence alignment." Bioinformatics 27.2 (2011): 182-188.

[7] Walters, John Paul, et al. "Accelerating Molecular Dynamics Simulations with GPUs." ISCA PDCCS. 2008.

A comparison of molecular dynamics simulations using GROMACS with GPU and CPU EGB/2015

A comparison of molecular dynamics simulations A comparison of molecular dynamics simulations using GROMACS with GPU and CPUusing GROMACS with GPU and CPU

Alex Camargo, Adriano Werhli, Karina Machado{alexcamargo, werhli, karina.machado}@furg.br

THANK YOU VERY MUCH!

Porto Alegre/2015