On the Feasibility of FPGA Acceleration of Molecular Dynamics Simulations

Abstract

Classical molecular dynamics (MD) simulations are important tools in life and material sciences since they allow studying chemical and biological processes in detail. However, the inherent scalability problem of particle-particle interactions and the sequential dependency of subsequent time steps render MD computationally intensive and difficult to scale. To this end, specialized FPGA-based accelerators have been repeatedly proposed to ameliorate this problem. However, to date none of the leading MD simulation packages fully support FPGA acceleration and a direct comparison of GPU versus FPGA accelerated codes has remained elusive so far.With this report, we aim at clarifying this issue by comparing measured application performance on GPU-dense compute nodes with performance and cost estimates of a FPGA-based singlenode system. Our results show that an FPGA-based system can indeed outperform a similarly configured GPU-based system, but the overall application-level speedup remains in the order of 2× due to software overheads on the host. Considering the price for GPU and FPGA solutions, we observe that GPU-based solutions provide the better cost/performance tradeoff, and hence pure FPGA-based solutions are likely not going to be commercially viable. However, we also note that scaled multi-node systems could potentially benefit from a hybrid composition, where GPUs are used for compute intensive parts and FPGAs for latency and communication sensitive tasks.

Heterogeneous Parallelization and Acceleration of Molecular Dynamics Simulations in GROMACS

“…With recent extensions to the fast multipole method 51 , we expect it to become the algorithm of choice for the largest parallel runs. Future technological improvements, including faster interconnects and closer onchip integration as well as advances in both traditional 3,52 and coarse-grained reconfigurable architectures 53 could allow getting closer to this performance target.…”

“…We believe this approach makes great use of limited compute resources to improve research productivity 2,3 , and it is increasingly enabling higher absolute performance on any given resource. Exploiting low-level parallelism can be tedious and has often been avoided in favor of using more hardware to achieve the desired time-to-solution.…”

Characterizing Molecular Dynamics Simulation on Commodity Platforms

“…While the performance gap with specialized hardware can hardly be closed [15,30,35,41,42], detailed workload analysis and characterization can still provide valuable insights to drive performance improvement for MD on commodity systems, thus expanding the experiments that do not require a DSA for a reasonable turnaround time [34]. While previous work [3,25,28] focused on proving good weak scaling properties and performance portability, the goal of this paper is to provide a detailed characterization of several representative and diverse MD experiments on modern high-performance commodity platforms.…”