Introducing GAINS, a GPU-accelerated continuous energy CSG based neutron Monte Carlo simulation code

GAINS is a GPU-accelerated Constructive Solid Geometry (CSG)-based continuous energy Monte Carlo neutron simulation code, designed for full core optimization, with regard to criticality and neutronics reactor performance. GAINS combines a hybrid event-history workflow to curb warp divergence, replaces Probability Density Functions (PDFs) with precomputed Cumulative Distribution Functions (CDFs), direct-indexed data layouts, and unrolls recursive CSG traversal into a stack-based scheme, which is evaluated using a spherical ray-march algorithm. Additionally, GAINS incorporates a localized and customizable distance cache. Accuracy and performance are assessed on three problems: the PU-239 Jezebel critical benchmark, a cylindrical fuel-element case (Base) with a movable hafnium absorber, and a full-core lattice industry benchmark. GAINS reproduces reference trends with small, traceable biases k eff of 21 pcm (Jezebel), 40 pcm average across control-rod positions (Base), and 28 pcm (Benchmark), primarily attributable to resampling inaccuracies in cross-section data leading to a faster spectrum. On an RTX 4090, GAINS completes 136 million histories in 23.26 s (Base) and 271.48 s (Benchmark), achieving 47.2 × and 6.1 × speed-ups over Serpent 2 on a dual-Xeon (160-thread) node, with significantly better cost-to-performance. Enabling the cache increases these speed-ups to an overall speed-up of 57.6x - 8.2x, respectively. The resulting capability enables fast, accurate neutronics calculations suitable for optimization loops and rapid design iteration on commodity hardware, while maintaining agreement with CSG-based CPU codes.