I haven't followed progress recently, but one question that's gotten some research is how to make GPU-ized raytracing work on dynamic scenes. Static scenes can

I haven't followed progress recently, but one question that's gotten some research is how to make GPU-ized raytracing work on dynamic scenes. Static scenes can be handled fairly well by preprocessing data into efficient spatial data structures like KD-trees or octrees. For example, here's an nVidia paper from 2009 that raytraces scenes made up of huge numbers of voxels, using a representation called a sparse voxel octree: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.156.3512&rep=rep1&type=pdf http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.156... Last time I looked at this research, though, how to handle highly dynamic scenes efficiently was a bit of an open question. It seems to largely be a data-structure research problem.