Advancing RT Core-Accelerated Fixed-Radius Nearest Neighbor Search

TL;DR

This work tackles efficient FRNN computation in dynamic particle simulations by leveraging RT Cores. It introduces a real-time BVH update/rebuild ratio optimizer (gradient), two neighbor-list-free RT-core FRNN variants (ORCS-persé and ORCS-forces), and a ray-traced periodic boundary condition method, all evaluated on Lennard-Jones interactions. The results show up to about $\sim 3.4\times$ speedups for BVH management, substantial performance and energy-efficiency gains for ORCS variants across radius distributions, and negligible penalties for periodic BC, with clear scaling across GPU generations. These contributions delineate when RT-core FRNN is advantageous and provide a practical toolkit for accelerating large-scale particle simulations while guiding future energy-aware and architectural optimizations.

Abstract

In this work we introduce three ideas that can further improve particle FRNN physics simulations running on RT Cores; i) a real-time update/rebuild ratio optimizer for the bounding volume hierarchy (BVH) structure, ii) a new RT core use, with two variants, that eliminates the need of a neighbor list and iii) a technique that enables RT cores for FRNN with periodic boundary conditions (BC). Experimental evaluation using the Lennard-Jones FRNN interaction model as a case study shows that the proposed update/rebuild ratio optimizer is capable of adapting to the different dynamics that emerge during a simulation, leading to a RT core pipeline up to $\sim 3.4\times$ faster than with other known approaches to manage the BVH. In terms of simulation step performance, the proposed variants can significantly improve the speedup and energy efficiency (EE) of the base RT core idea; from $\sim1.3\times$ at small radius to $\sim2.0\times$ for log normal radius distributions. Furthermore, the proposed variants manage to simulate cases that would otherwise not fit in memory because of the use of neighbor lists, such as clusters of particles with log normal radius distribution. The proposed RT Core technique to support periodic BC is indeed effective as it does not introduce any significant penalty in performance. In terms of scaling, the proposed methods scale both their performance and EE across GPU generations. Throughout the experimental evaluation, we also identify the simulation cases were regular GPU computation should still be preferred, contributing to the understanding of the strengths and limitations of RT cores.

Figures (13)

• Figure 1: Fixed‑radius search with RT cores: a ray launched from particle $p_3$ intersects the bounding spheres of $p_2$ and $p_4$, which are its neighbors.
• Figure 2: An example simulation of five particles. The left side shows the current state of its BVH and the simulation of one time step which moves particles $p_1$ and $p_5$. On the right, the resulting BVH by applying update or rebuild.
• Figure 3: Cost curve using a fixed number of updates $k_u$ on the BVH.
• Figure 4: The pipeline for one simulation step using RT cores. The purple boxes represent the traditional way of leveraging RT Cores, while the red and teal ones are the proposed ORCS variants with no need for neighbor lists.
• Figure 5: Variable radius case: only $p_2$'s ray triggers an intersection, therefore such thread is in charge of adding the force contributions on both $p_1$ and $p_2$.