I just finished writing a blog post, New Ideas in Raytracing. It's about my two favorite new-ish raytracing techniques: stochastic progressive photon mapping, and divide-and-conquer raytracing. I don't do a lot of raytracing in my day job, but like lots of graphics programmers I'm an enthusiast of the subject, and I try to keep up somewhat with what's going on in that field from year to year. I thought these methods were worth sharing.
DACRT is really interesting, actually I'm a little embarrassed I didn't think of it when I wrote my tracer.
It's such a clean solution, I'm amazed no one thought of it a while ago.
There's a new "supercomputer on a chip" come out for \\$99 that we are thinking of doing something with, this might be a good starting point.
Funny, just yesterday I checked out the new generation of Nvidia gfx cards (kepler) and they were talking about how good they are for raytracing.
DACRT does indeed look interesting. I wonder how well this would work with sorting algorithms other than quicksort. A while back I was playing with an idea to use radix sort to build a hirarchical structure that can be used for raytracing. E.g. a point in space would fit into some space in an octree. Instead of using the x,y,z coordinates of the pixel, you could store the node indices and then sort them to get an optimised structure to use with OpenCL. This way you would be able to have moving objects in a scene since they can arbitrarily change their location bevore beeing sorted. But in the end... who has the time... I had to put the porject to rest