Accelerating Molecular Modeling Applications withGraphics ProcessorsJOHN E. STONE,1*JAMES C. PHILLIPS,1*PETER L. FREDDOLINO,1,2*DAVID J. HARDY,1*LEONARDO G. TRABUCO,1,2KLAUS SCHULTEN1,2,31Beckman Institute, University of Illinois at Urbana-Champaign, Urbana, Illinois, 618012Center for Biophysics and Computational Biology, University of Illinois at Urbana-Champaign,Urbana, Illinois, 618013Department of Physics, University of Illinois at Urbana-Champaign, Urbana, Illinois, 61801Received 5 April 2007; Revised 27 June 2007; Accepted 30 July 2007DOI 10.1002/jcc.20829Published online 25 September 2007 in Wiley InterScience (www.interscience.wiley.com).Abstract: Molecular mechanics simulations offer a computational approach to study the behavior of biomoleculesat atomic detail, but such simulations are limited in size and timescale by the available computing resources. State-of-the-art graphics processing units (GPUs) can perform over 500 billion arithmetic operations per second, a tremen-dous computational resource that can now be utilized for general purpose computing as a result of recent advancesin GPU hardware and software architecture. In this article, an overview of recent advances in programmable GPUsis presented, with an emphasis on their application to molecular mechanics simulations and the programming techni-ques required to obtain optimal performance in these cases. We demonstrate the use of GPUs for the calculation oflong-range electrostatics and nonbonded forces for molecular dynamics simulations, where GPU-based calculationsare typically 10–100 times faster than heavily optimized CPU-based implementations. The application of GPU accel-eration to biomolecular simulation is also demonstrated through the use of GPU-accelerated Coulomb-based ionplacement and calculation of time-averaged potentials from molecular dynamics trajectories. A novel approximationto Coulomb potential calculation, the multilevel summation method, is introduced and compared with direct Cou-lomb summation. In light of the performance obtained for this set of calculations, future applications of graphicsprocessors to molecular dynamics simulations are discussed.q 2007 Wiley Periodicals, Inc. J Comput Chem 28: 2618–2640, 2007Key words: GPU computing; CUDA; parallel computing; molecular modeling; electrostatic potential; multilevelsummation; molecular dynamics; ion placement; multithreading; graphics processing unitIntroductionMolecular mechanics simulations of biomolecules, from theirhumble beginnings simulating 500-atom systems for less than10 ps,1have grown to the point of simulating systems containingmillions of atoms2,3and up to microsecond timescales.4,5Evenso, obtaining sufficient temporal sampling to simulate significantmotions remains a major problem,6and the simulation of everlarger systems requires continuing increases in the amount ofcomputational power that can be brought to bear on a singlesimulation. The increasing size and timescale of such simula-tions also require ever-increasing computational resources forsimulation setup and for the visualization and analysis of simula-tion results.Continuing advances in the hardware architecture of graphicsprocessing units (GPUs) have yielded tremendous computationalpower, required for the interactive rendering of complex imageryfor entertainment, visual simulation, computer-aided design, andscientific visualization applications. State-of-the-art GPUsemploy IEEE floating point arithmetic, have on-board memorycapacities as large as the main memory systems of some per-sonal computers, and at their peak can perform over 500 billionfloating point operations per second. The very term ‘‘graphicsprocessing unit’’ has replaced the use of terms such as graphicsaccelerator and video board in common usage, indicating theincreased capabilities, performance, and autonomy of currentgeneration devices. Until recently, the computational power ofGPUs was very difficult to harness for any but graphics-orientedalgorithms due to limitations in hardware architecture and, to alesser degree, due to a lack of general purpose application pro-Contract grant sponsor: National Institutes of Health; contract grant num-ber: P41-RR05969*The authors contributed equallyCorrespondence to: K. Schulten; e-mail: [email protected]; web:http://www.ks.uiuc.edu/q 2007 Wiley Periodicals, Inc.gramming interfaces. Despite these difficulties, GPUs were suc-cessfully applied to some problems in molecular modeling.7–9Recent advances in GPU hardware and software have eliminatedmany of the barriers faced by these early efforts, allowing GPUsto now be used as performance accelerators for a wide varietyof scientific applications.In this article, we present a review of recent developments inGPU technology and initial implementations of three specificapplications of GPUs to molecular modeling: ion placement, cal-culation of trajectory-averaged electrostatic potentials, and non-bonded force evaluation for molecular dynamics simulations. Inaddition, both CPU- and GPU-based implementations of a newapproximation to Coulomb energy calculations are described. Inthe sample applications discussed here, GPU-accelerated imple-mentations are observed to run 10–100 times faster than equiva-lent CPU implementations, although this may be further im-proved by future tuning. The ion placement and potential aver-aging tools described here have been implemented in themolecular visualization program VMD,10and GPU accelerationof molecular dynamics has been implemented in NAMD.11Theadditional computing power provided by GPUs brings the possi-bility of a modest number of desktop computers performing cal-culations that were previously only possible with the use ofsupercomputing clusters; aside from molecular dynamics, a num-ber of other common and computationally intensive tasks in bio-molecular modeling, such as Monte Carlo sampling and multiplesequence alignment, could also be implemented on GPUs in thenear future.Graphics Processing Unit OverviewUntil recently, GPUs were used solely for their ability to processimages and geometric information at blistering speeds. The dataparallel nature of many computer graphics algorithms naturallyled the design of GPUs toward hardware and software architec-tures that could exploit this parallelism to achieve high perform-ance. The steadily increasing demand for advanced graphics withincommon software applications has allowed high-performancegraphics systems
View Full Document
Unlocking...