Harvesting graphics power for MD simulations

Meel, Jacobus A. van; Arnold, Axel; Frenkel, Daan; Zwart, Simon F. Portegies; Belleman, R.G.

doi:10.1080/08927020701744295

Cited by 144 publications

(145 citation statements)

References 12 publications

Supporting

Mentioning

142

Contrasting

Unclassified

Order By: Relevance

“…This requirement is relevant even for systems as small as a few thousand particles because many simulation platforms nowadays use limited precision to accelerate molecular dynamics simulations. [16][17][18][19][20][21][22][23] For these, the multigrator provides a way of addressing certain artifacts that arise when barostatting and thermostatting is done for large-dimensional systems or with large relaxation times.…”

Section: Multigrator Decomposition For the Martyna-tobias-klein mentioning

confidence: 99%

Accurate and efficient integration for molecular dynamics simulations at constant temperature and pressure

Lippert

Predescu

Ierardi

et al. 2013

The Journal of Chemical Physics

174

148

View full text Add to dashboard Cite

In molecular dynamics simulations, control over temperature and pressure is typically achieved by augmenting the original system with additional dynamical variables to create a thermostat and a barostat, respectively. These variables generally evolve on timescales much longer than those of particle motion, but typical integrator implementations update the additional variables along with the particle positions and momenta at each time step. We present a framework that replaces the traditional integration procedure with separate barostat, thermostat, and Newtonian particle motion updates, allowing thermostat and barostat updates to be applied infrequently. Such infrequent updates provide a particularly substantial performance advantage for simulations parallelized across many computer processors, because thermostat and barostat updates typically require communication among all processors. Infrequent updates can also improve accuracy by alleviating certain sources of error associated with limited-precision arithmetic. In addition, separating the barostat, thermostat, and particle motion update steps reduces certain truncation errors, bringing the time-average pressure closer to its target value. Finally, this framework, which we have implemented on both general-purpose and special-purpose hardware, reduces software complexity and improves software modularity.

show abstract

Section: Multigrator Decomposition For the Martyna-tobias-klein mentioning

confidence: 99%

Accurate and efficient integration for molecular dynamics simulations at constant temperature and pressure

Lippert

Predescu

Ierardi

et al. 2013

The Journal of Chemical Physics

174

148

View full text Add to dashboard Cite

show abstract

“…Several groups have started to implemented MD routines on APs. Meel et al [27] describe a CUDA implementation of Lennard-Jones MD which achieves a net speedup of up to 40 times over a conventional CPU. Although suitable for coarse-grained simulations, the lack of support for an atomistic, biomolecular force field limits the applicability of the code.…”

Section: Accelerated Modelingmentioning

confidence: 99%

The impact of accelerator processors for high-throughput molecular modeling and simulation

Giupponi

Harvey

Fabritiis

2008

Drug Discovery Today

View full text Add to dashboard Cite

The recent introduction of cost-effective accelerator processors (APs), such as the IBM Cell processor and Nvidia's graphics processing units (GPUs), represents an important technological innovation which promises to unleash the full potential of atomistic molecular modeling and simulation for the biotechnology industry. Present accelerator processors can deliver over an order of magnitude more floatingpoint operations per second (flops) than standard processors, broadly equivalent to a decade of Moore's law growth, and significantly reduce the cost of current atombased molecular simulations. In conjunction with distributed and grid computing solutions, accelerated molecular simulations may finally be used to extend current in silico protocols by use of accurate thermodynamic calculations instead of approximate methods and simulate hundreds of protein-ligand complexes with full molecular specificity, a crucial requirement of in silico drug discovery workflows.Teaser phrase: New accelerated computing devices will unleash the predictive power of molecular modeling and simulation for biotechnology.Key words: Cell processor, graphics processing units (GPUs), accelerated computing, molecular modeling and simulation, drug discovery, distributed and grid computing Preprint submitted to Elsevier Science 5 August 2008Bringing a new drug to market is a long and expensive process[1] encompassing theoretical modeling, chemical synthesis and experimental and clinical trials. Despite the considerable growth of biotechnologies in the last ten years, the practical consequences of these techniques on the number of approved drugs has failed to meet expectation [2]. Nonetheless, the discovery process greatly benefits from the use of computational modeling [3], at least in the initial stages of compound discovery, screening and optimization [4].Among the techniques available, force-based molecular modeling methods (briefly, molecular modeling) such as molecular dynamics are particularly useful in studying molecular processes at the atomistic level, providing accurate information on macromolecular dynamics and thermodynamic properties. Bridging from the molecular-atomistic (femtosecond) to biological (micromillisecond) timescales is still an unaccomplished feat in computational biology: the complexity of the modeling impedes sufficient sampling of the evolution of the system, even on expensive high performance computing (HPC) resources. For instance, cost and sampling constraints have so far limited a routine molecular dynamics run over a PC cluster to a single protein system for tens of nanoseconds, barely sufficient to compute the binding free energy using an exact thermodynamic method [5]. This information is of great importance in the discovery process for virtual screening, lead optimization and in silico drug discovery [4], but needs to be computed for hundreds of protein-ligand systems efficiently and economically in order to open the way for molecular simulations to become a routine tool in the discovery workflows us...

show abstract

“…In particular, in molecular dynamics (MD) simulation codes [3,4,5,6,7] these algorithms are most commonly employed. Although these algorithms suffer from limitations on modern SIMD architectures [8,9], there have been only a few attempts to overcome them, most of them specific to GPUs [10,11] without achieving generality.…”

Section: Introductionmentioning

confidence: 99%

“…Moreover, the throughput-oriented GPU architecture requires high level of parallelism and is sensitive to memory access patterns. In order to target GPUs, some codes combine the traditional algorithms with data regularization techniques [15,10], but such approaches can still lead to inefficient execution. Recasting the algorithms to a more regular data access has been shown to result in higher IPC on GPUs, but not without additional trade-offs [16].…”

Section: Introductionmentioning

confidence: 99%

A flexible algorithm for calculating pair interactions on SIMD architectures

Páll

Hess

2013

Computer Physics Communications

593

400

View full text Add to dashboard Cite

Calculating interactions or correlations between pairs of particles is typically the most time-consuming task in particle simulation or correlation analysis. Straightforward implementations using a double loop over particle pairs have traditionally worked well, especially since compilers usually do a good job of unrolling the inner loop. In order to reach high performance on modern CPU and accelerator architectures, singleinstruction multiple-data (SIMD) parallelization has become essential. Avoiding memory bottlenecks is also increasingly important and requires reducing the ratio of memory to arithmetic operations. Moreover, when pairs only interact within a certain cut-off distance, good SIMD utilization can only be achieved by reordering input and output data, which quickly becomes a limiting factor. Here we present an algorithm for SIMD parallelization based on grouping a fixed number of particles, e.g. 2, 4, or 8, into spatial clusters. Calculating all interactions between particles in a pair of such clusters improves data reuse compared to the traditional scheme and results in a more efficient SIMD parallelization. Adjusting the cluster size allows the algorithm to map to SIMD units of various widths. This flexibility not only enables fast and efficient implementation on current CPUs and accelerator architectures like GPUs or Intel MIC, but it also makes the algorithm future-proof. We present the algorithm with an application to molecular dynamics simulations, where we can also make use of the effective buffering the method introduces.

show abstract

Harvesting graphics power for MD simulations

Cited by 144 publications

References 12 publications

Accurate and efficient integration for molecular dynamics simulations at constant temperature and pressure

Accurate and efficient integration for molecular dynamics simulations at constant temperature and pressure

The impact of accelerator processors for high-throughput molecular modeling and simulation

A flexible algorithm for calculating pair interactions on SIMD architectures

Contact Info

Product

Resources

About