Codecs, containers and bitrate — how it all fits together

Container vs. codec

A container (MOV, MP4, MXF, MKV) is a file format: it wraps one or more tracks of picture, sound, subtitles and timecode, keeps them in sync and carries metadata. A codec determines how the data in an individual track is compressed. The same container can carry different codecs, and the same codec can sit in different containers.

DIAGRAM · WHAT’S INSIDE THE FILE

That is why ‘the MP4 won’t play’ can happen: the player knows the container but not the codec inside (HEVC or AV1 on an older device, for example).

Intra vs. long-GOP

Codecs compress the picture in two ways:

• Within a frame (intra). Each frame is compressed on its own, like a photograph. Apple says of all ProRes codecs that they are ‘frame-independent’ — each frame is encoded and decoded independently. DNxHR works the same way. Files are larger, but the editing software can jump to any frame without calculating the ones around it.
• Between frames (long-GOP). The codec stores a complete key frame (I) and then only the differences: P frames are predicted from previous frames, B frames from both previous and following ones. The group from one I frame to the next is the GOP. At the same quality the files are much smaller — which is why distribution works this way (H.264, H.265, AV1). When editing, though, the computer has to decode other frames to display a single one, which slows playback down with 4K and several tracks.

H.264 and H.265 also have all-intra profiles, used by some professional cameras. So the name of the codec alone doesn’t tell you whether it is long-GOP; you need to look at the profile.

Bit depth and chroma subsampling

Bit depth determines the number of levels per channel: 8 bits = 256, 10 bits = 1,024, 12 bits = 4,096. For log recording and grading, 10 bits is a sensible minimum.

Chroma subsampling exploits the fact that the eye is more sensitive to brightness than to colour. The J:a:b notation tells you how many colour samples there are relative to luminance samples in a block of 4 × 2 pixels:

In normal viewing you often won’t see a difference between 4:2:2 and 4:2:0. In keying (green screen) and heavy grading you will — edges and fine colour transitions break up sooner.

Bitrate and file size

The data rate (bitrate) is the amount of data per second. File size is simply:

size [B] = data rate [b/s] × duration [s] ÷ 8

Example: 100 Mb/s × 3,600 s ÷ 8 = 45,000,000,000 B = 45 GB per hour.

With CBR (constant bitrate) the result is exact. With VBR (variable bitrate) the encoder gives more bits to demanding scenes and fewer to simple ones — the final size is close to the average, not exact. ProRes is VBR with a target number of bits per frame; Apple says the maximum is about 10% above the target and many simple frames are smaller. DNxHR has a fixed number of bytes per frame for each level and resolution, so its data rate can be calculated exactly.

INTERACTIVE · STORAGE CALCULATOR

Storage calculator

Resolution HD 1920 × 10802K DCI 2048 × 1080UHD 3840 × 21604K DCI 4096 × 2160

Codec / preset ProRes 422ProRes 4444ProRes 422 ProxyProRes 422 LTProRes 422 HQProRes 4444 XQ DNxHR 444DNxHR LBDNxHR SQDNxHR HQDNxHR HQX H.264 — enter the bitrate H.265 / HEVC — enter the bitrate AV1 — enter the bitrate Other codec — enter the bitrate

Frame rate 23.976 fps24 fps25 fps29.97 fps30 fps50 fps59.94 fps60 fps100 fps119.88 fps120 fps

Data rate (Mb/s)

Recording length

h min s

Space required —

per minute

—

per hour

—

data rate

—

per frame

—

The DNxHR values are calculated from bytes per frame: bytes × 8 × 25 ÷ 10⁶. Avid also gives them in MB/s in its table (e.g. 57.42 MB/s for UHD SQ at 25 fps — in binary megabytes). Here GB per hour is calculated in decimal (10⁹ B).

Which codec when