Towards production FPGA-accelerated molecular dynamics: Progress and challenges

Abstract

Recent work in the FPGA acceleration of molecular dynamics simulation has shown that including on-the-fly neighbor list calculation (particle filtering) in the device has the potential for an 80× per core speed-up over the CPU-based reference code and so to make the approach competitive with other computing technologies. In this paper we report on progress and challenges in advancing this work towards the creation of a production system, especially one capable of running on a large-scale system such as the Novo-G. The current version consists of an FPGAaccelerated NAMD-lite running on a PC with a Gidel PROCStar III. The most important implementation issues include software integration, handling exclusion, and modifying the force pipeline. In the last of these we have added support for Particle-MeshEwald and augmented the Lennard-Jones calculation with a switching function. In experiments, we find that energy stability so far appears to be acceptable, but that longer simulations are needed. Due primarily to the added complexity of the force pipelines, performance is somewhat diminished from the previous study; we find, however, that porting to a newer (existing) device will more than compensate for this loss.

Revisiting FPGA Acceleration of Molecular Dynamics Simulation with Dynamic Data Flow Behavior in High-Level Synthesis

“…As an initial effort, we mainly focus on the acceleration of the short-range, non-bonded force computation that is the critical path of the MD algorithm, similar to prior studies [1,13,14,15,16]. Therefore, we will only report the performance of one simulation iteration for the short-range, non-bonded force computation throughout this paper.…”

“…Compared to ASIC-based and GPU-based acceleration approaches, FPGA-based acceleration emerges as an attractive alternative because FPGAs provide better programmability than ASICs, and better power and energy efficiency than GPUs. In prior studies, there have been several attempts (e.g., [1,13,14,15,16]) to accelerate MD using FPGAs. Most of these prior studies mainly focus on the acceleration for the critical path of the MD algorithm, which is the computation of short-range, non-bonded forces for each atom in every simulation iteration.…”

“…In this paper we revisit the FPGA acceleration for MD simulation in high-level synthesis (HLS) [18,19,20,21] that raises the abstraction level from RTL to C/C++. Our goal is achieve a comparable performance of prior reference RTL acceleration of MD simulation [13,14], but with more affordable programming cost in HLS. Similar to prior studies [1,13,14,15,16], we focus on the acceleration of the critical path of the MD algorithm, that is, the computation of short-range, non-bonded forces for each atom, as explained in Section 2.…”

“…Our goal is achieve a comparable performance of prior reference RTL acceleration of MD simulation [13,14], but with more affordable programming cost in HLS. Similar to prior studies [1,13,14,15,16], we focus on the acceleration of the critical path of the MD algorithm, that is, the computation of short-range, non-bonded forces for each atom, as explained in Section 2. To provide more insights into the community, we start with a naive HLS design and optimize the HLS design all the way to generate the reference architecture written in RTL [13,14].…”

“…Similar to prior studies [1,13,14,15,16], we focus on the acceleration of the critical path of the MD algorithm, that is, the computation of short-range, non-bonded forces for each atom, as explained in Section 2. To provide more insights into the community, we start with a naive HLS design and optimize the HLS design all the way to generate the reference architecture written in RTL [13,14]. In summary, this paper makes the following contributions.…”