Ray tracing is a technique for image synthesis: creating a photorealistic picture of a 3-D world. But the technique tends to be a time-consuming process, and how to decrease the expensive computational cost has become a hotspot. Considering the factors of task partition and load balance, with the two levels of task partition method, a new dynamically self-adaptive distributed ray tracing algorithm was developed to enhance the efficiency of ray tracing based on Physically Based Ray Tracing (PBRT), which is a famous ray tracing framework from Stanford University. When using CPU cores to 80 in experiment, the corresponding speedup of the proposed algorithm versus the original PRBT algorithm reached to a perfect value, which is close to theoretic linear speedup. It shows that the proposed algorithm is effective and highly scalable with respect to PBRT. The proposed method can be applied to ray tracing process effectively to accelerate the efficiency of rendering.