A tile is a rectangular region of a video frame that can be encoded and decoded independently of the other tiles in the same frame. Modern codecs (HEVC, AV1, VVC) let the encoder split each frame into a grid of tiles — typically 2×2 or 4×4 — and process each tile as a separate, parallel work unit. The motivation is parallelism: a single 8K frame is too big for one CPU core to encode or decode in real time, but split into 16 tiles, the work fits across 16 cores.
The technical constraint that makes tiles work: predictions cannot cross tile boundaries within the same frame. Each tile is self-contained for intra-prediction and motion vectors within the frame, so worker N decoding tile N never has to wait for worker M decoding tile M. The small cost: predictions that would have benefitted from looking across a tile boundary can't, slightly reducing compression efficiency — typically 1–3 % bitrate penalty depending on tile count. For real-time work and high-resolution streaming, the parallelism win dramatically outweighs the bitrate cost.
For a product team, tiles are the enabling technology behind real-time 4K and 8K encoding and decoding. Without tiles, a single CPU core would have to serialise every macroblock in the frame, which can't keep up with 4K60 on consumer hardware. With tiles, decoding scales across cores and silicon implementations naturally — every modern hardware decoder (NVENC, Quick Sync, VideoToolbox, NETINT VPU, smart-TV chipsets) processes tiles in parallel. The visible result: 4K HDR plays smoothly on a phone or TV that would have struggled in the H.264 era, because tiles let the silicon parallelise the work.
