Post date: Aug 21, 2008 5:15:46 PM
Where the hardware in question is probably Larrabee, most occlusion tests nowadays are done using rasterization techniques on the GPU to exploit its parallelism. We could naively expect this to be the same on Larrabee, but with its more programmable architecture might it not be more prudent to return to more traditional approaches (that are parallelized of course)? Unfortunately I don't know, and without hardware to benchmark I won't be able to say for certain anytime soon, unless someone from Intel tells me.