Shinjiro Toyoda

DEVELOPMENT OF MD ENGINE : HIGH-SPEED ACCELERATOR WITH PARALLEL PROCESSOR DESIGN FOR MOLECULAR DYNAMICS SIMULATIONS

Application of molecular dynamics (MD) simulations to large systems, such as biological macromolecules, is severely limited by the availability of computer resources. As the size of the system increases, the number of nonbonded forces (Coulombic and van der Waals interactions) to be evaluated increases as 𝒪(N2), where N is the number of particles in the system. The force evaluation consumes more than 99% of the CPU time in an MD simulation involving over 10,000 particles. Hence, the major target for reduction of the CPU time should be acceleration of the calculation of nonbonded forces. For this purpose, we developed a custom processor for calculating nonbonded interactions and a scalable plug‐in machine (to a workstation), the MD Engine, in which numbers of the custom processors work in parallel. The processor has a pipeline architecture to calculate the total nonbonded force using the coordinates, electric charge, and species of each particle broadcast by the host computer. The force is calculated with sufficient accuracy for practical MD simulations. The processor also calculates virials simultaneously with forces for use in the calculation of pressure, accommodates periodic boundary conditions, and can be used in Ewald summations. An MD Engine system consisting of 76 processors calculates nonbonded interactions about 50 times faster than an UltraSPARC‐I processor (Sun Ultra‐2, 200 MHz) or an R10000 processor (SGI Origin 200, 180 MHz). On a Sun Ultra‐2 workstation with a single UltraSPARC‐I processor an MD simulation of a Ras p21 protein molecule immersed in a water sphere (13,258 particles) was accelerated by a factor of 48 using the MD Engine system. ©1999 John Wiley & Sons, Inc. J Comput Chem 20: 185–199, 1999

Development of hardware accelerator for molecular dynamics simulations: a computation board that calculates nonbonded interactions in cooperation with fast multipole method.

Evaluation of long‐range Coulombic interactions still represents a bottleneck in the molecular dynamics (MD) simulations of biological macromolecules. Despite the advent of sophisticated fast algorithms, such as the fast multipole method (FMM), accurate simulations still demand a great amount of computation time due to the accuracy/speed trade‐off inherently involved in these algorithms. Unless higher order multipole expansions, which are extremely expensive to evaluate, are employed, a large amount of the execution time is still spent in directly calculating particle–particle interactions within the nearby region of each particle. To reduce this execution time for pair interactions, we developed a computation unit (board), called MD‐Engine II, that calculates nonbonded pairwise interactions using a specially designed hardware. Four custom arithmetic‐processors and a processor for memory manipulation (“particle processor”) are mounted on the computation board. The arithmetic processors are responsible for calculation of the pair interactions. The particle processor plays a central role in realizing efficient cooperation with the FMM. The results of a series of 50‐ps MD simulations of a protein–water system (50,764 atoms) indicated that a more stringent setting of accuracy in FMM computation, compared with those previously reported, was required for accurate simulations over long time periods. Such a level of accuracy was efficiently achieved using the cooperative calculations of the FMM and MD‐Engine II. On an Alpha 21264 PC, the FMM computation at a moderate but tolerable level of accuracy was accelerated by a factor of 16.0 using three boards. At a high level of accuracy, the cooperative calculation achieved a 22.7‐fold acceleration over the corresponding conventional FMM calculation. In the cooperative calculations of the FMM and MD‐Engine II, it was possible to achieve more accurate computation at a comparable execution time by incorporating larger nearby regions.

Large Scale Molecular Dynamics Simulation Using MD-Engine

Recently, demands for computing and designing large complex systems, such as proteins, biomembrain and so on, grow tremendously. For those systems, accurate calculation of non-bonding interactions becomes crucial to predict correct stable structures and thermodynamic properties in molecular dynamics(MD) simulations. However, for examples, truncation of Coulombic interaction must be taken very large and consequently calculation cost scales as O(N2) Special purpose molecular dynamics(MD) hardware accelerator, which is called the MD-Engine, has been used to overcome this difficulty and applied to several complex systems. The MD-Engine consists of custum prosessors that calculate pairwise potentials and forces in parallel. A user just has to send coordinate of atoms and force field parameters, then receive forces and potential energy. Our test calculations show good linear scaling with the system size.