This article presents the novel algorithmic developments and performance analysis of the GPU-optimized REActor Physics Monte Carlo (GREAPMC) graphical processing unit (GPU)–accelerated multigroup Monte Carlo (MC) code tailored specifically for pressurized water reactor simulations. GREAPMC tackles the thread divergence issue inherent in history-based neutron tracking on GPUs by introducing two new optimization strategies. The first novel approach dynamically replaces inactive particles with new ones during the execution of the transport loop, while the second strategy enhances efficiency by capping the history length during active cycles at a predefined maximum number of interactions. Subsequently, it sorts and invokes the kernel with only the surviving neutrons. The maximum number of interactions is automatically adjusted while considering cycle time during inactive cycles. Both methods significantly accelerate computation compared to MCS, a high-fidelity MC code developed at Ulsan National Institute of Science and Technology, with the latter approach demonstrating the most substantial acceleration. GREAPMC further enhances efficiency by adopting cell-based geometry modeling. This approach eliminates cell search overhead, ensuring consistent execution times even as the number of cells increases. Overall, these algorithmic developments in GREAPMC achieve substantial computational acceleration against MCS. A single GPU card in this study demonstrates performance equivalent to approximately 570 cores from the specific CPU model used.