high performance computing on graphics processing units: hgpu.org

The radiosity method permits the obtaining of high quality images through the evaluation of the global illumination of the scene. The computational complexity and the memory requirements of the algorithm are the main problems when a large scene has to be processed. To reduce the memory requirements, Monte Carlo radiosity method is often used. In this paper we present an efficient implementation of the Monte Carlo radiosity algorithm on the GPU using CUDA. We have developed different strategies to increase the performance of the implementation: utilization of an additional simplified version of the mesh to reduce the computational requirements, partitioning of the scene to increase the data locality and an efficient thread scheduling to exploit the characteristics of the GPU. The results are good in terms of execution time, increasing the flexibility of previous solutions.