Hi. Following this post about Larrabee, I read the paper and became interested in their software rendering implementation. According to the paper, their algorithm is based on this paper. The way I understand it is the following: the frame starts with a single-node covering the whole screen. Then, for each triangle rendered, all the nodes intersecting with an edge of the triangle are recursively subdivided until they fits a single-pixel, leaving unsubdivided nodes inside the triangles. I am correct? It's not clear in the Larrabee paper, but it looks like they keep zmin and zmax for each node, so new triangles can be rejected in parts or totally based on the depth coverage. Isn't it what they call hierachical z-buffer? Thanks for any comments about this! JF

bump