Abstract: Ray tracing is the most commonly used realistic rendering algorithm, which plays a great role in 3D animation, virtual reality, digital twin and other fields. The increasing demand for rendering quality has brought a lot of attention to large-scale scene ray tracing. In this paper, we review and analyse the related work on large-scale scene ray tracing. Firstly, we introduces the data organization of large-scale scenes from data compression and decomposition; secondly, we reviews the out-of-core ray tracing methods, and analyses the application of LOD in these methods; thirdly, we introduces the distributed ray tracing methods, and classifies them into to image-space parallel, data parallel, and hybrid parallel for analysis and comparison; Finally, we summaries the research progress of large-scale scene ray tracing, analyse the problems of existing methods in data storage, data access, rendering quality, rendering speed, etc., and point out the possible future research directions, including neural network-based scene data compression, instantiated scene data decomposition, computational hotspot prediction, and adaptive distributed strategies.