Overhead comparison

overhead comparison

back to main page

Stream types

For this comparison, the following streams will be used:

Video stream of about 2h 08min:
1. encoded with XVID in VFW mode:
  - AS @ L5
  - max. consecutive BVOPs: 4
  - packed bitstream
  - RAW stream size in AVI: 979,009,500 bytes
  - RAW stream size after remuxing to native MKV: 978,291,879 bytes
2. coming soon: XVID in VFW mode:
  - AS @ L5
  - max. consecutive BVOPs: 2
  - no packed bitstream
  - RAW stream size in AVI: 979,004,381
  - RAW stream size in MKV: 978,764,818
The same audio stream encoded in
- DTS (754.5 kbps, 5.1)
- AC3 (448 kbps, 5.1)
- LC-AAC (about 254 kbps, 5.1)
- HE-AAC (about 223 kbps, 5.1)
- MPEG1 Layer 3 (about 192 kbps, stereo)
- MPEG2 Layer 3 (about 64 kbps, stereo)
- Vorbis (89 kbps, stereo)

RAW stream sizes

The following sizes will be used as RAW stream sizes when calculating file overhead (file size - raw stream sizes):

video: sum of all chunks in the file (= stream size in AVI). Although technically the RAW stream size in MP4 and Matroska (native mode) differs from RAW AVI stream size, the overhead will be given relative to the AVI RAW stream size, unless something else is explicitely stated: This way, you see what you gain or lose when converting a file as used in this comparison to something else than AVI. Think of overhead as of 'difference' if the idea of negative overhead lets your brain spin around. Note that packed bitstream files in AVI contain a lot of crap that is removed when muxing to native mkv or mp4, so that remuxing packed bitstream files from AVI to MKV or MP4 will gain more space than doing the same with non-packed bitstream AVI files.
AAC: ADTS file with ADTS headers removed (= RAW stream size in AVI and Matroska)
MP3, DTS, AC3: file size
Vorbis: OGG file size without OGG page headers (= RAW stream size inside matroska)

Muxing applications and settings

AVI: AVI-Mux GUI 1.17.4
- audio interleave: 100 kB
- rec lists: enabled
- in case of Open-DML: legacy index enabled with 1 MB for the first RIFF list
- AC3 audio: 5 frames per chunk
- MP3 audio: 10 frames per chunk
- DTS audio: 10 frames per chunk
MKV: AVI-Mux GUI 1.17.4
- Cluster settings
  - maximum size: 512 kB
  - maximum duration: 30 seconds
  - write PrevClusterSize: no
  - write Position: no
  - index clusters in Seekhead: no
- Cue settings
  - minimum interval: 2000ms
  - reserved space: 20 kB/h/stream
  - enabled for video and audio
  - write CueBlockNumber: yes
  - video lacing: no
  - audio lacing:
    - vorbis: 50ms
    - mp3: 500ms
    - dts: 100ms
    - all other stream types: 200ms
  - timecode scale: 1,000,000
OGM: VirtualDubMod 1.5.10.1
MP4: MP4Box 0.4.1-DEV
- interleave: 500ms (default)

Remarks on those settings

Vorbis is only laced into blocks of 50 milliseconds because larger settings seem to be problematic. Sometimes, even laces of 1000 milliseconds work, sometimes even 100 milliseconds lead to jerky playback, depending on splitter and decoder. As I only want to use playable files in this comparison, I've used only the highest safe setting, which is 50 milliseconds. For other stream types, 200 milliseconds is a good compromise between overhead (larger laces mean less overhead) and seek accuracy.

For AC3-in-AVI, 5 frames per chunk is the largest value which really works, larger values can cause playback errors with some decoders.

When proper Cues are present, there is no need for Clusters to be indexed in the SeekHead, as it would only waste space. The Cues settings used in this comparison ensure that far more CuePoints than required for proper seeking are created.

Result

The overhead is indicated for:

only the video stream
the video stream plus one of the audio streams

The difference between (2) and (1) is the overhead caused by the audio stream, and the overhead you have to add if you want to add another stream of this type (this is not 100% accurate, but certainly 99.9%). In the case of matroska, both Matroska v1 (i.e without SimpleBlocks) and Matroska v2 (i.e. with SimpleBlocks) is checked, and in the case of DTS, AC3 and MP3, the overhead is checked with and without header stripping. The first number represents the overhead with header stripping, the second one the overhead without header stripping. Note that, in case of audio formats with long synchronisation bytes sequences and short duration, like DTS, the space gained by header stripping can exceed the overhead matroska causes, leading to an overhead below 0. In the case of DTS audio, even one single DTS stream gains enough space for the total file overhead to be lower than 0.

streams	container
	OGM	AVI			MKV		MP4
		AVI 1.0	AVI 2.0		MKV
		AVI 1.0	plain Open-DML	low overhead	v1	v2
video 1 only	10,320,645	4,926,908	3,313,528	1,844,120	VFW mode: 2,461,675 native: 1,436,437 / 2,205,189	VFW mode: 1,376,081 native: -51,098 / 717,584	1,214,770
video 2 only	10,305,184	4,961,091	3,346,273	1,877,513	VFW mode: 2,582,034 native: 1,913,303 / 2,682,110	VFW mode: 1,436,145 native: 427,840 / 1,196,557	1,693,988
video + dts	18,357,147	6,957,224	4,623,844	2,716,532	353,474 3,235,798	-924,267 1,958,075	-
video + ac3	15,580,849	6,292,268	4,197,584	2,443,768	2,392,921 2,873,313	1,211,273 1,691,665	-
video + he-aac	-	9,038,499	6,070,977	3,377,655	3,037,528	1,859,594	1,943,544
video + lc-aac	-	13,101,375	8,779,471	4,728,207	3,220,869	2,039,182	2,602,393
video + mpeg 1 layer 3	12,957,532	5,745,560	3,847,916	2,219,160	2,805,789 3,126,041	1,675,835 1,996,087	2,559,350
video + vorbis	11,352,901	-			3,951,485	2,577,660	-
video + mpeg 2 layer 3	11,607,777	5,692,520	3,813,660	2,166,168	2,650,275 2,970,545	1,520,359 1,840,629	2,539,350