Abstract: The Monte Carlo method uses random sampling to simulate and solve physical and mathematical problems. By virtually tracking the complete transport process of radiation particles through the human body, it provides statistically reliable dose estimates and is widely regarded as the “gold standard” for radiation dosimetry. However, the high computational cost of a single Monte Carlo run makes it difficult to achieve fast and timely dose assessment in practice. To address this limitation, GAPMC, a GPU-accelerated Monte Carlo program, was developed for photon external dose calculation based on human voxel phantoms. An event-history-based parallel simulation strategy was implemented by exploiting the high concurrency of modern GPU architecture, where each GPU thread independently tracks every single photon history. For geometric particle transport within voxel geometries, a three-dimensional digital differential analyzer (3D-DDA) algorithm was employed to compute voxel boundary crossings efficiently. Cross-section data were taken from the EPICS2017 database. Photon interactions were modeled using the Livermore physics model, in which reaction cross-sections are sampled via the rejection sampling method. Secondary electrons were handled through a local energy deposition approximation, which is justified by the fact that the mean free path of electrons is smaller than the voxel size. To exploit the spatial locality inherent in voxel traversal, the voxel material index array was stored in GPU texture memory rather than as a one-dimensional array in global memory. Both single- and double-precision floating-point versions of GAPMC were developed and the computation speed was compared. Validation against the CPU-based Geant4 toolkit shows good agreement in dose calculations over a photon energy range of 0.5 MeV to 10 MeV, with whole-body effective dose deviations within 5%. In double-precision mode, GAPMC achieves acceleration factor of 887.5 to 1028 relative to Geant4. The use of texture memory for voxel indexing yields an additional performance improvement of approximately 20% compared to a global memory implementation. The single-precision version maintains result consistency with the double-precision version while delivering a further speedup of 60% to 114%, reaching a total acceleration factor of 1454.4 to 2366.2 relative to CPU-based Geant4. GAPMC is an effective GPU-based solution for photon dose calculation in voxel geometries. The combination of GPU parallelism, texture memory optimization, and floating-point precision flexibility provides substantial computational acceleration with acceptable accuracy sacrifice, demonstrating practical potential for rapid dose assessment in radiation protection applications.