Molecular Dynamics Simulations on High-Performance Reconfigurable Computing Systems

Abstract

The acceleration of molecular dynamics (MD) simulations using high-performance reconfigurable computing (HPRC) has been much studied. Given the intense competition from multicore and GPUs, there is now a question whether MD on HPRC can be competitive. We concentrate here on the MD kernel computation: determining the short-range force between particle pairs. In one part of the study, we systematically explore the design space of the force pipeline with respect to arithmetic algorithm, arithmetic mode, precision, and various other optimizations. We examine simplifications and find that some have little effect on simulation quality. In the other part, we present the first FPGA study of the filtering of particle pairs with nearly zero mutual force, a standard optimization in MD codes. There are several innovations, including a novel partitioning of the particle space, and new methods for filtering and mapping work onto the pipelines. As a consequence, highly efficient filtering can be implemented with only a small fraction of the FPGA’s resources. Overall, we find that, for an Altera Stratix-III EP3ES260, 8 force pipelines running at nearly 200 MHz can fit on the FPGA, and that they can perform at 95% efficiency. This results in an 80-fold per core speed-up for the short-range force, which is likely to make FPGAs highly competitive for MD.

Efficient particle-pair filtering for acceleration of molecular dynamics simulation

“…We use the force pipeline specified in a previous study [1]. It uses a mix of 32-bit integer (extended for accumulation) and single precision floating point.…”

“…Since multipliers are a critical resource, we also show the two "sphere" filters implemented entirely with logic. The cost of a force pipeline (from [1]) is shown for scale.…”

“…In particular, we study the filtering of particle pairs with nearly zero mutual force, a standard optimization in MD codes. When coupled with our recent work on the optimization of MD force pipelines [1], this has the potential to improve performance of our previous full MD implementation [2] four-fold. The issue of filtering particle pairs emerges from the geometric mismatch between two shapes: (i) the cubes (or other polyhedrons) into which it is convenient to partition the simulation space and (ii) the spheres around each particle in which the short-range force is non-zero.…”

Molecular Dynamics Simulation of Iron — A Review

“…In fact, the design principle of recon¯gurable computing shifts to formulating the connection among various memory blocks and logic blocks as well as to devising the force pipeline that results in the interatomic forces used for motion integration. 159 Therefore, developers of applications in FPGA need to program the hardware connections every time a new algorithm is adopted. Despite the hardware-level approach, the idea behind the work°ow remains the same as those applied in CPU and GPU.…”

An FPGA Accelerator for Molecular Dynamics Simulation Using OpenCL

“…Therefore, ASICs are out-of-reach for most researchers, although their performances are quite excellent. A cheap way of hardware acceleration is provided by FPGAs (feild-programmablegate-arrays) 13,14,15 . Although the cost is extremely small compared to ASICs, the design time is still very large.…”