HEVC Under the Hood
HEVC keeps the broad shape of H.264 but loosens the partition and deepens the toolset. The result is a codec that reaches the same distortion at a lower bitrate, at the expense of more compute per frame. This sheet records where the savings come from.
Variable partitions
The fixed macroblock becomes a coding tree unit that splits into a quadtree of blocks, each as large as sixty-four samples or as small as a few. A flat sky becomes one large block; a busy edge becomes many small ones. The encoder searches the split tree and keeps the shape that gives the best rate-distortion trade, which is why HEVC scales to high resolution far better than H.264.
Deeper prediction
HEVC adds more intra directions, a merge mode that borrows motion from neighbours, and weighted prediction for fades. Together these cut the residual, and a smaller residual can be quantised harder for the same visible result. Ten-bit reconstruction is common, which removes banding on gradients and is now a default for the high-definition tiers.
| Tool | Gain |
|---|---|
| split tree | Adapts partition to content |
| merge mode | Cuts motion coding |
| weighted prediction | Handles fades and blends |
| ten-bit path | Removes banding |
The compute trade
The split search and the motion tools demand time. On a general processor, HEVC is markedly slower per frame than H.264 at the same setting; on a fixed engine the gap narrows. The desk therefore records the engine with every HEVC timing, because the same setting means different work on different silicon.
Where it pays
From 1080p upward, HEVC usually wins on size at a fixed distortion, which is why it anchors the mid and high rungs of most ladders. The ladder design sheet shows how the rungs are chosen; the accelerator sheet covers the engines that run it at speed.