23 ms·

Standard MPEG-1 MPEG-2 MPEG-4 Visual MPEG-4 AVC MPEG-H HEVC Bitrate vs previous 25% less 25% less 30% less 60% less The table at the end is hiding

Standard MPEG-1 MPEG-2 MPEG-4 Visual MPEG-4 AVC MPEG-H HEVC Bitrate vs previous 25% less 25% less 30% less 60% less The table at the end is hiding one factor : the computation cost of each new codecs. Have we really found new better ways to compress videos, or is it just Moore's law in action ?

wmf 9y ago
Yes, the computation per pixel has increased ~2x per codec generation so codecs are taking advantage of (a fraction of) the gains provided by Moore's Law.
- krylon 9y ago
  I remember that time when I tried to play a 4k video (x264 or x265, I am not certain) on my Mac Mini. I was pleasantly surprised that it actually ran smoothly, but the poor thing nearly melted.
matthewmacleod 9y ago
It depends what you mean by "better". More advanced video codecs have higher visual quality at lower bitrates, but are more computationally expensive. All of these video codecs use the same underlying ideas of 2d transforms, intra- and inter-frame compression, motion compensation, entropy coding etc., but using more complex techniques to preserve video quality. I wouldn't say it's a simple as "Moore's law in action".
- ttoinou 9y ago
  Better is increasing filesize/( (video quality)(computation cost) )
ecesena 9y ago
I think it's because the major cost is bandwidth, so everyone is incentivized to optimized on size primarily, maybe at the cost of being more cpu/memory intensive on enc/dec.
- ttoinou 9y ago
  Of course ! I was just wondering about the real technical improvement... Talking about bitrate/(video quality) is forgetting about (computation complexity) that's all
brigade 9y ago
Depends on what you mean by "just Moore's law." Certainly each standard was developed with a higher target decode complexity; H.264 and H.265 both targeted 2x greater than the previous standard, JVET is currently targeting 16x, and I think AV1 is aiming for less than 2x HW area. So yeah, there were obvious gains left on the table for H.264 for complexity reasons that H.265 picked up; likewise for H.265/VP9 that AV1 picked up. That said, there has been a lot of refinement that didn't necessarily increase complexity, and there have been various tools that saw reinvigorated research once complexity targets were raised.

wmf 9y ago

Yes, the computation per pixel has increased ~2x per codec generation so codecs are taking advantage of (a fraction of) the gains provided by Moore's Law.

krylon 9y ago
I remember that time when I tried to play a 4k video (x264 or x265, I am not certain) on my Mac Mini. I was pleasantly surprised that it actually ran smoothly, but the poor thing nearly melted.

matthewmacleod 9y ago

It depends what you mean by "better". More advanced video codecs have higher visual quality at lower bitrates, but are more computationally expensive. All of these video codecs use the same underlying ideas of 2d transforms, intra- and inter-frame compression, motion compensation, entropy coding etc., but using more complex techniques to preserve video quality. I wouldn't say it's a simple as "Moore's law in action".

ttoinou 9y ago
Better is increasing filesize/( (video quality)(computation cost) )

ecesena 9y ago

I think it's because the major cost is bandwidth, so everyone is incentivized to optimized on size primarily, maybe at the cost of being more cpu/memory intensive on enc/dec.

ttoinou 9y ago
Of course ! I was just wondering about the real technical improvement... Talking about bitrate/(video quality) is forgetting about (computation complexity) that's all

brigade 9y ago

Depends on what you mean by "just Moore's law." Certainly each standard was developed with a higher target decode complexity; H.264 and H.265 both targeted 2x greater than the previous standard, JVET is currently targeting 16x, and I think AV1 is aiming for less than 2x HW area. So yeah, there were obvious gains left on the table for H.264 for complexity reasons that H.265 picked up; likewise for H.265/VP9 that AV1 picked up. That said, there has been a lot of refinement that didn't necessarily increase complexity, and there have been various tools that saw reinvigorated research once complexity targets were raised.