A Deep Dive into Nanite Virtualized Geometry

Caption notes

Brian Karis explains an aim of Nanite’s early design: to have the cost of processing geometry track screen resolution more closely. The system divides triangles into clusters and uses estimated screen-space error from the current view to select detail for different regions. Neighboring clusters that switch independently can leave cracks, so he describes switching groups together and changing group boundaries across levels to avoid keeping the same edges locked. He also notes that geometry made of many small pieces, such as grass and leaves, does not fully follow that cost model.