Tile-based rendering
A GPU architecture that splits the image into small tiles and renders one tile at a time in fast on-chip memory — built to save bandwidth and power.
Rather than reading and writing the whole framebuffer in external memory, a tile-based GPU first bins all the geometry per screen tile and then renders one tile at a time, entirely in small but very fast on-chip memory. The tile is written out to main memory only once it is finished. That saves memory bandwidth, and with it power — decisive on phones.
The architecture dominates mobile graphics: PowerVR, ARM Mali, Qualcomm Adreno and Apple's own GPUs. The PowerVR and Apple variant (TBDR) also removes hidden surfaces per tile, eliminating almost all overdraw. The opposite, immediate mode, has long been the norm on desktop NVIDIA and AMD cards.