Tile-based processing

A hardware-level graphics architecture that divides the screen into a grid of small tiles to process geometry and shading on-chip, slashing memory bandwidth and power consumption.

Tile-based processing (often implemented as tile-based deferred rendering) is the architectural backbone of modern mobile and Apple silicon GPUs. Unlike traditional immediate-mode desktop GPUs that constantly write and read massive framebuffers to and from external VRAM, a tile-based GPU divides the screen into a grid of small regions (typically 16x16 or 32x32 pixels). The hardware bins the scene's geometry, deferring the pixel shading phase until it determines exactly which polygons are visible within each tile. By keeping all depth testing, blending, and shading inside fast, on-chip tile memory (which is physically located on the GPU die), the system eliminates expensive round-trips to system memory. This design delivers a massive reduction in power consumption and thermal output, making it highly efficient for battery-constrained devices like smartphones, VR headsets, and ultra-thin laptops.

https://developer.apple.com/documentation/metal/gpu_devices_and_work_submission/tailor_your_apps_for_apple_gpus_and_tile-based_deferred_rendering