To develop a more realistic virtual reality firefighting training platform, this study utilizes fire spread data, including smoke and heat, obtained from numerical simulations using the Fire Dynamics Simulator (FDS). FDS employs MPI and OpenMP for large-scale fire simulations by dividing computational domains into sub-domains, with OpenACC applied to support heterogeneous hardware architectures. Performance tests using CSIRO Grassland Fires demonstrated a 1.89 × speedup with combined CPU-GPU computation and a 21 × speedup with 1 GPU and 16 CPUs, validating enhanced fire analysis capabilities. Additionally, existing VR engines were improved by integrating WFDS data into Unreal Engine for realistic smoke and heat visualization using FGA files. The program dynamically visualizes flame and smoke movements based on wind speed and direction, achieving a 100% match between WFDS data and Unreal Engine output, with combustion stages rendered in real-time through mass-based material updates.