Streaming DVD Video over the Internet
Sexy Commando writes "According to this article on ZDNet, the new codec, H.264, is able to stream DVD quality video using bandwidth as little as less than 1Mbps. The new codec requires 3 to 4 times as much CPU power than MPEG-2 to process the video. Now we can have two movies on 1 CD. Cheers."
Here is the VideoLocus press release for H.264/MPEG-4 AVC.
I once shot a man who posted too many, "Imagine a beowulf cluster of these"
No. Dolby AC3 would take 384 kbit/s, for all channels.
H.264 exists as MPEG-4 part 10, basically using the AVC rather than the ASP profile for encoding.
Supposedly, it offers up to 2-4x size reduction over the MPEG-4 ASP.
However...
For anyone who has extensively played with the existing ASP codecs available (basically XVID, DIVX, RV9, and WM-whatever), the quality matters a *lot* based on the implementation. And not in any consistent way, letting you pick "codec X does the best job". Nope, more like "on low-motion sequences, codec X does best. For detail, codec Y. For minimal artifacts but some bluring, codec Z", and so on.
I see no reason to expect H.264 will follow any substantially different path. In another 5 years, it might well let us get a DVD quality movie onto 1 CD. For now, don't hold your breath about this changing the scene overnight. By the time this really does make good on its potential, we'll have the bandwidth and storage to make it unnecessary.
AC3 streams use lossy compression. They can use as much as 640kbps, but typical DVDs use either 384 or 448 kbps.
Sorry, here's your reference.
The economist has a great article reviewing the latest codec offerings from different players. Specifically DivX 5.0 "is said to be particularly good at preventing tearing, a playback error that occurs when the software cannot render the video for display at the same pace that it is being decompressed and fed into the media player. And a new codec from supersecretive Pulsent claims to be object rather than block based. Whereas block-based compression and object-oriented codecs slice up backgrounds and foregrounds into grids, the Pulsent approach actually pinpoints real-world items in the frame--such as a person, tree or building--and processes each element separately. story here