Anyway, my point was that even with today's processors, the clock speed isn't the problem nearly as often as memory bandwidth. So even if the 970 has a less optimal AltiVec implementation, the combination of somewhat higher clock speed and vastly higher memory bandwidth should translate into real-world performance advantages a lot better than the clock speed increase would suggest.
Assuming you aren't memory bound. You've only got 166 Mbps TOTAL between the two processors, which well tuned AltiVec codec can saturate with a single processor, let alone two.
Yeah, it seems like less than 5% of the US has GSM coverage, but in the four months I've had a GSM phone, I haven't ONCE been somewhere without access, except in a few spots on the Acela from Boston to NYC. And this is with a pretty heavy travel schedule. From my base in Portland, I've been to San Jose, San Francisco, New York, Boston, Wallingford CT, Albany OR, Las Vegas.
I could see it being a problem for a tractor salesman or anyone else who finds themselves in rural America a lot, but as a technology consultant, GSM has had perfect coverage.
You're inaccurately harsh on MPEG-1. It can provide quite good quality at adequate data rates, which a high-quality professional encoder. if you use something like Canopus's ProCoder, even a VCD should look fine. For CD-ROM stuff, a 2-pass VBR 320x240 at 2 Mbps should be free of visible artifacts for most sources.
MPEG-1 certainly requires more bits per pixel to provide high quality compared to other formats, but when that's an option, MPEG-1 can do great.
Well, Icecast is normally a http streaming system, right, not RTSP.
RTSP is important, since it provides infrastructure for much better loss recovery, and scalability. HTTP is fine for short content, but for anything much more than 10 minutes, it makes random access really icky, and real-time performance is very dicy unless the connection speed is a lot higher than the clip data rate.
These are problems that WM9 and to a lesser degree MPEG-4 have already solved and rolled out into mature, deployed implementations. Which is why I worry that Theora is too far behind to really make much of an impact on the general public web video space.
The HD DVD development process has been underway for quite a long time, and will almost certainly be using MPEG-4 AVC for the video codec. Given that the Theora bitstream isn't even locked down, I'd say there's nil chance for Theora getting in as part of the official spec.
Also, AVC, which is already standardized, almost certainly is technically superior to whatever Theora 1.0 provides. It's a really, really good codec that uses a wide variety of patented techniques that Theora doesn't have access to.
As for Apple and AAC-LC, the DRM is file based, so they could have applied it to any codec. But AAC-LC is a great codec, and it's part of MPEG-4, which Apple has been pushing in a big way. I was mildly surprised they didn't go for HE AAC, but I suppose they didn't want to delay the launch to get that technology incorporated into QuickTime.
I don't anticipate Theora becoming popular in mainstream non-computer devices. MPEG-4 and Windows Media 9 are filling the ecological niches where Theora could potentially play. And unlike Vorbis v. MP3, Theora doesn't have any significant technical advantages over the market leaders.
Still, I wish it well. Having a fully open technology like this would be useful in certain niches. And I could see it used in places like games, where compression efficiency isn't as important as having no license fees and having easy implementation.
Well, it really depends on what you want to do. The reason why MPEG-2 is "good enough" for DVD is that it is compatible with the DVD Video spec, which means it'll play back on set top boxes. It's not like we didn't have better codecs than MPEG-2 five years ago, but they aren't enough better that it'd make it worth ditching the installed base of DVD players!
Still, we'll certainly be moving away from MPEG-2 when we move to HD DVD, since backwards compatibility will be punted anyway.
There are lots of features that MPEG-2 has that are useful for video archiving and distribution, like support for interlaced video, support for non-square pixel files, and low per pixel-second CPU requirements. I don't see that Theora has any of these advantages.
Conversely, it costs $2.50 to license MPEG-2 decoder support for a product, and Theora will be free. And Theora will be able to provide better quality at low data rates.
It all depends on what you need to do.
Personally, I doubt Theora will get a lot of uptake by corporations. Its openness advantages are unlikely to overcome its disadvantages in maturity. Heck, Ogg Vorbis is quite mature, but no major media companies are using it as a distribution format. Apple picked AAC-LC instead, even though they have to pay a fee, as part of their general support of MPEG-4.
Bang a cymbal, and let it fade out into nothingness. You can definitely hit audible limits of 16-bit PCM in that case. PCM->FFT->PCM will make it worse.
Also, codec like Dolby Digital are capable of decoding in more than 16-bit, so with capable equipment, you're really able to take advantage of available dynamic range.
Do you have any more detail on how you think VP3 is streatched to its limits?
Also, making a ground-up, competitive codec that doesn't infringe on any patents is a risky and potentially overwhelming task. There is a LOT of patented stuff in this area, and codec development is intrinsically risky and difficult. Given it's taken them this long to productive VP3, I can't imagine they had anything like the resources neccessary to actually build a real codec that was that much better than VP3. And in fact, that's pretty much why they switched from Tarkin to Theora.
I worry because there are many different ways a file can be degraded. A perceptual codec +a noise floor due to rounding can sound a lot worse with very quiet sections than the same perceptual codec without that extra noise.
Is there any explicit support for RTSP in the Theora project so far? Having a codec is less than half of the way to having a robust real-time streaming format. VP3 was a progressive download codec in all implementations I've ever seen of it. Even though QuickTime supports RTSP, the VP3 implementation of it didn't support native hinting for reliable RTSP.
Well, Theora isn't done yet, nor is its bitstream locked down, so it's impossible to say how small files will get.
By the time it is GM, I imagine MPEG-4 will be well along in its migration to the new AVC codec, which offers much better compression efficiency than the current Simple and Advanced Simple profiles used by Divx, Xvid, etcetera. So even though the final Theora might be somewhat better than MPEG-4 today, it almost certainly will be behind MPEG-4 by the time it is released.
Bear in mind that MP3 is ten years old now. Modern audio codecs like HE AAC are definitely better than Vorbis, technically.
If Theora gets market share, it'll be because of its openness, not because of any price or quality advantages. Windows Media 9 is free-as-in-beer for most uses, and is today a lot better than Theora could possibly be in a year.
Actually, the corporate world, especially European companies, are looking to MPEG-4 as their future format of choice. The cable industry has already agreed to switch to MPEG-4 for digital cable (in the vague future). MPEG-4 is becoming the standard format for cell phones via 3GPP.
A lot of this has to do with maturity. You can actually buy interoperable, commerical MPEG-4 solutions from a variety of vendors today. Also, MPEG-4 supports real-time streaming over lossy networks. And it has profiles for everything from cell phones to HD. There's at least 100x more work into MPEG-4 than Theora.
Of course, Microsoft's Windows Media 9 is even farther along in maturity in many ways, and certainly has strong technical advantages over MPEG-4 if Windows 98 or higher is the exclusive playback platform. It has better compression efficinecy, and much better scalability over real-world consumer internet access.
It really depends on what you're trying to accomplish.
What cost savings will content creators have by using Theora? It's not like you have to pay a license fee to distribute digital video files in the propritary formats. There are some cases where you have to pay content fees with MPEG-4, but they don't apply to most users (like you get the first 50,000 users a year free, and don't have to pay for marketing content, just content that gives you revenue).
When you're doing a lossy compression like MP3 or AAC, you're going to convert from PCM to a frequency domain, and then back again. Going from 16-bit PCM to 16-bit frequency and back to 16-bit PCM introduces rounding errors. Starting with 24-bit in the first step will effectively eliminate those rounding errors, reducing noise slightly. So even if playback is via 16-bit in the end, keeping it 24-bit as late into the process as possible will pay off.
Oh, it certainly won't be a night and day difference. But it is a factor. When doing a FFT or any transfer into frequency domain, rounding errors are always a significant factor when delivering with the same precision as the source. I imagine this would matter more for content with a high dynamic range like classical, and less for typical pop/rock content.
This is of course an emperical question. A good test would be to take a 24-bit source file, and encode it to AAC-LC in QuickTime in Better and then Best modes. Better will treat the source as 16-bit, and Best mode will use the extra precision of the 24-bit source. In fact, the ONLY difference between Better and Best is the use of more than 16-bits of information in the signal path.
Over 160? Maybe you've been to too many Megadeth concerts or something, but 160 Kbps is quite audibly lossy in my experience. Now, I'm fussy about encoding artifacts, but 160 is the lowest I'll use for listening to on headphones on an airplane. It has to be at least 192 for me to not find artifacts distracting while listening on a good stereo (Paradigm Active/20 reference monitors attached to my video editing rig, in my case, self-powered with all XLR signal routing from the jukebox machine. I grant this is overkill for the casual listener).
Personally, I encode my library at 320 Kbps Normal Stereo without any filtering. This is overkill for listening, but that's enough data that I can recompress to another, more portable format like AAC on an iPod without windup up with a audible multigeneration artifacts.
All things being equal, I'd use FLAC, but I really really like the iTunes interface, and 320 MP3 is the best format it has historically supported. It now does 320 AAC, and I'm toying with switching to that (although I haven't yet, since the files won't be quite as widely interoperable).
Not a good test of iTunes service
on
AAC Put To The Test
·
· Score: 4, Interesting
It looks like they're working with 16-bit source, not the 24-bit source that most of the iTunes AAC files are ripped from. So this test, while certainly very interesting, won't be useful to determine the iTunes music store quality.
Also, I didn't mean that to be a criticism of the original test. 6.0 was the current version of QuickTime when they did the test, so it looks like a fair test for the state of the technologies at the time.
Don't take 64 Kbps AAC results seriously
on
AAC Put To The Test
·
· Score: 5, Informative
I note that in the 64 Kbps test, they used the AAC-LC encoder from QuickTime 6.0. This was a pretty darn lousy one, lacking any ability to specify a sample rate at a given data rate, and had poor quality. The current version of QuickTime 6.3 (for Windows and MacOS X), has a much improved, more flexible AAC-LC encoder, so if they did that test today AAC would likely rank higher.
If using the Apple encoder, encode in "Better" mode with 16-bit source, and in "Best" mode with source that's more than 16-bits per sample (and hence isn't a CD rip). Support for mastering from 24-bit when running in "Best" is one of the reasons why the AAC-LC files as part of iTunes sound so good.
No, it'd be like playing X-rated segments from Girls Gone Wild during the Power Puff Girls.
There's spam out there with graphic pictures of women having sex with donkeys. Someone randomly sending those out without the faintest hint or intention to filter kids out of the list doesn't merit a "cavet emptor" defense!
Said spammer does, however, certainly deserve to be the model in their next donkyphilic photmontage.
Of course, if it's an indie band, than you can't use the "the Big Five is keeping me down" excuse. It's up to each band to decide to release their stuff for file sharing. Screwing a major label is one thing, but screwing a band who owns their own copyright, and hasn't released stuff for sharing is another. If anything, the moral imperative to pay is greater for indies, since they don't have those big advances to live off of.
Well, pitch, speed, etcetera could all be trivially obscured if a ripper knew what to look for, but a perceptual codec itself would likely keep those paramaters alone. Of course, not a lot of syntax could be put in there.
And yes, this isn't at all specific to MP3. The reason why formats like WMA and Apple's secure MPEG-4 use encryption+metadata instead of watermarking is exactly this. One could certainly apply a propritary extension to MP3 to do the same thing.
Personally, I've never been convinced by Digimarc. I've yet to see a convincing demo with content I've chosen that kept the watermark sufficiently robust.
Sure, you can stuff data into the audio file in any number of ways. But not in a way that'll survive any kind of recompression, or analog playback. A watermark is something that has to handle lossy reproduction by definition. Persitant metadata doesn't.
Slashdot Mirror
You're correct - my typo.
Anyway, my point was that even with today's processors, the clock speed isn't the problem nearly as often as memory bandwidth. So even if the 970 has a less optimal AltiVec implementation, the combination of somewhat higher clock speed and vastly higher memory bandwidth should translate into real-world performance advantages a lot better than the clock speed increase would suggest.
Assuming you aren't memory bound. You've only got 166 Mbps TOTAL between the two processors, which well tuned AltiVec codec can saturate with a single processor, let alone two.
Yeah, it seems like less than 5% of the US has GSM coverage, but in the four months I've had a GSM phone, I haven't ONCE been somewhere without access, except in a few spots on the Acela from Boston to NYC. And this is with a pretty heavy travel schedule. From my base in Portland, I've been to San Jose, San Francisco, New York, Boston, Wallingford CT, Albany OR, Las Vegas.
I could see it being a problem for a tractor salesman or anyone else who finds themselves in rural America a lot, but as a technology consultant, GSM has had perfect coverage.
You're inaccurately harsh on MPEG-1. It can provide quite good quality at adequate data rates, which a high-quality professional encoder. if you use something like Canopus's ProCoder, even a VCD should look fine. For CD-ROM stuff, a 2-pass VBR 320x240 at 2 Mbps should be free of visible artifacts for most sources.
MPEG-1 certainly requires more bits per pixel to provide high quality compared to other formats, but when that's an option, MPEG-1 can do great.
Well, Icecast is normally a http streaming system, right, not RTSP.
RTSP is important, since it provides infrastructure for much better loss recovery, and scalability. HTTP is fine for short content, but for anything much more than 10 minutes, it makes random access really icky, and real-time performance is very dicy unless the connection speed is a lot higher than the clip data rate.
These are problems that WM9 and to a lesser degree MPEG-4 have already solved and rolled out into mature, deployed implementations. Which is why I worry that Theora is too far behind to really make much of an impact on the general public web video space.
The HD DVD development process has been underway for quite a long time, and will almost certainly be using MPEG-4 AVC for the video codec. Given that the Theora bitstream isn't even locked down, I'd say there's nil chance for Theora getting in as part of the official spec.
Also, AVC, which is already standardized, almost certainly is technically superior to whatever Theora 1.0 provides. It's a really, really good codec that uses a wide variety of patented techniques that Theora doesn't have access to.
As for Apple and AAC-LC, the DRM is file based, so they could have applied it to any codec. But AAC-LC is a great codec, and it's part of MPEG-4, which Apple has been pushing in a big way. I was mildly surprised they didn't go for HE AAC, but I suppose they didn't want to delay the launch to get that technology incorporated into QuickTime.
I don't anticipate Theora becoming popular in mainstream non-computer devices. MPEG-4 and Windows Media 9 are filling the ecological niches where Theora could potentially play. And unlike Vorbis v. MP3, Theora doesn't have any significant technical advantages over the market leaders.
Still, I wish it well. Having a fully open technology like this would be useful in certain niches. And I could see it used in places like games, where compression efficiency isn't as important as having no license fees and having easy implementation.
Well, it really depends on what you want to do. The reason why MPEG-2 is "good enough" for DVD is that it is compatible with the DVD Video spec, which means it'll play back on set top boxes. It's not like we didn't have better codecs than MPEG-2 five years ago, but they aren't enough better that it'd make it worth ditching the installed base of DVD players!
Still, we'll certainly be moving away from MPEG-2 when we move to HD DVD, since backwards compatibility will be punted anyway.
There are lots of features that MPEG-2 has that are useful for video archiving and distribution, like support for interlaced video, support for non-square pixel files, and low per pixel-second CPU requirements. I don't see that Theora has any of these advantages.
Conversely, it costs $2.50 to license MPEG-2 decoder support for a product, and Theora will be free. And Theora will be able to provide better quality at low data rates.
It all depends on what you need to do.
Personally, I doubt Theora will get a lot of uptake by corporations. Its openness advantages are unlikely to overcome its disadvantages in maturity. Heck, Ogg Vorbis is quite mature, but no major media companies are using it as a distribution format. Apple picked AAC-LC instead, even though they have to pay a fee, as part of their general support of MPEG-4.
You don't listen to a lot of classical, do you?
Bang a cymbal, and let it fade out into nothingness. You can definitely hit audible limits of 16-bit PCM in that case. PCM->FFT->PCM will make it worse.
Also, codec like Dolby Digital are capable of decoding in more than 16-bit, so with capable equipment, you're really able to take advantage of available dynamic range.
Do you have any more detail on how you think VP3 is streatched to its limits?
Also, making a ground-up, competitive codec that doesn't infringe on any patents is a risky and potentially overwhelming task. There is a LOT of patented stuff in this area, and codec development is intrinsically risky and difficult. Given it's taken them this long to productive VP3, I can't imagine they had anything like the resources neccessary to actually build a real codec that was that much better than VP3. And in fact, that's pretty much why they switched from Tarkin to Theora.
I worry because there are many different ways a file can be degraded. A perceptual codec +a noise floor due to rounding can sound a lot worse with very quiet sections than the same perceptual codec without that extra noise.
Is there any explicit support for RTSP in the Theora project so far? Having a codec is less than half of the way to having a robust real-time streaming format. VP3 was a progressive download codec in all implementations I've ever seen of it.
Even though QuickTime supports RTSP, the VP3 implementation of it didn't support native hinting for reliable RTSP.
Well, Theora isn't done yet, nor is its bitstream locked down, so it's impossible to say how small files will get.
By the time it is GM, I imagine MPEG-4 will be well along in its migration to the new AVC codec, which offers much better compression efficiency than the current Simple and Advanced Simple profiles used by Divx, Xvid, etcetera. So even though the final Theora might be somewhat better than MPEG-4 today, it almost certainly will be behind MPEG-4 by the time it is released.
Bear in mind that MP3 is ten years old now. Modern audio codecs like HE AAC are definitely better than Vorbis, technically.
If Theora gets market share, it'll be because of its openness, not because of any price or quality advantages. Windows Media 9 is free-as-in-beer for most uses, and is today a lot better than Theora could possibly be in a year.
Actually, the corporate world, especially European companies, are looking to MPEG-4 as their future format of choice. The cable industry has already agreed to switch to MPEG-4 for digital cable (in the vague future). MPEG-4 is becoming the standard format for cell phones via 3GPP.
A lot of this has to do with maturity. You can actually buy interoperable, commerical MPEG-4 solutions from a variety of vendors today. Also, MPEG-4 supports real-time streaming over lossy networks. And it has profiles for everything from cell phones to HD. There's at least 100x more work into MPEG-4 than Theora.
Of course, Microsoft's Windows Media 9 is even farther along in maturity in many ways, and certainly has strong technical advantages over MPEG-4 if Windows 98 or higher is the exclusive playback platform. It has better compression efficinecy, and much better scalability over real-world consumer internet access.
It really depends on what you're trying to accomplish.
What cost savings will content creators have by using Theora? It's not like you have to pay a license fee to distribute digital video files in the propritary formats. There are some cases where you have to pay content fees with MPEG-4, but they don't apply to most users (like you get the first 50,000 users a year free, and don't have to pay for marketing content, just content that gives you revenue).
When you're doing a lossy compression like MP3 or AAC, you're going to convert from PCM to a frequency domain, and then back again. Going from 16-bit PCM to 16-bit frequency and back to 16-bit PCM introduces rounding errors. Starting with 24-bit in the first step will effectively eliminate those rounding errors, reducing noise slightly. So even if playback is via 16-bit in the end, keeping it 24-bit as late into the process as possible will pay off.
Oh, it certainly won't be a night and day difference. But it is a factor. When doing a FFT or any transfer into frequency domain, rounding errors are always a significant factor when delivering with the same precision as the source. I imagine this would matter more for content with a high dynamic range like classical, and less for typical pop/rock content.
This is of course an emperical question. A good test would be to take a 24-bit source file, and encode it to AAC-LC in QuickTime in Better and then Best modes. Better will treat the source as 16-bit, and Best mode will use the extra precision of the 24-bit source. In fact, the ONLY difference between Better and Best is the use of more than 16-bits of information in the signal path.
Over 160? Maybe you've been to too many Megadeth concerts or something, but 160 Kbps is quite audibly lossy in my experience. Now, I'm fussy about encoding artifacts, but 160 is the lowest I'll use for listening to on headphones on an airplane. It has to be at least 192 for me to not find artifacts distracting while listening on a good stereo (Paradigm Active/20 reference monitors attached to my video editing rig, in my case, self-powered with all XLR signal routing from the jukebox machine. I grant this is overkill for the casual listener).
Personally, I encode my library at 320 Kbps Normal Stereo without any filtering. This is overkill for listening, but that's enough data that I can recompress to another, more portable format like AAC on an iPod without windup up with a audible multigeneration artifacts.
All things being equal, I'd use FLAC, but I really really like the iTunes interface, and 320 MP3 is the best format it has historically supported. It now does 320 AAC, and I'm toying with switching to that (although I haven't yet, since the files won't be quite as widely interoperable).
It looks like they're working with 16-bit source, not the 24-bit source that most of the iTunes AAC files are ripped from. So this test, while certainly very interesting, won't be useful to determine the iTunes music store quality.
Also, I didn't mean that to be a criticism of the original test. 6.0 was the current version of QuickTime when they did the test, so it looks like a fair test for the state of the technologies at the time.
I note that in the 64 Kbps test, they used the AAC-LC encoder from QuickTime 6.0. This was a pretty darn lousy one, lacking any ability to specify a sample rate at a given data rate, and had poor quality. The current version of QuickTime 6.3 (for Windows and MacOS X), has a much improved, more flexible AAC-LC encoder, so if they did that test today AAC would likely rank higher.
If using the Apple encoder, encode in "Better" mode with 16-bit source, and in "Best" mode with source that's more than 16-bits per sample (and hence isn't a CD rip). Support for mastering from 24-bit when running in "Best" is one of the reasons why the AAC-LC files as part of iTunes sound so good.
No, it'd be like playing X-rated segments from Girls Gone Wild during the Power Puff Girls.
There's spam out there with graphic pictures of women having sex with donkeys. Someone randomly sending those out without the faintest hint or intention to filter kids out of the list doesn't merit a "cavet emptor" defense!
Said spammer does, however, certainly deserve to be the model in their next donkyphilic photmontage.
Of course, if it's an indie band, than you can't use the "the Big Five is keeping me down" excuse. It's up to each band to decide to release their stuff for file sharing. Screwing a major label is one thing, but screwing a band who owns their own copyright, and hasn't released stuff for sharing is another. If anything, the moral imperative to pay is greater for indies, since they don't have those big advances to live off of.
Well, pitch, speed, etcetera could all be trivially obscured if a ripper knew what to look for, but a perceptual codec itself would likely keep those paramaters alone. Of course, not a lot of syntax could be put in there.
And yes, this isn't at all specific to MP3. The reason why formats like WMA and Apple's secure MPEG-4 use encryption+metadata instead of watermarking is exactly this. One could certainly apply a propritary extension to MP3 to do the same thing.
Personally, I've never been convinced by Digimarc. I've yet to see a convincing demo with content I've chosen that kept the watermark sufficiently robust.
Sure, you can stuff data into the audio file in any number of ways. But not in a way that'll survive any kind of recompression, or analog playback. A watermark is something that has to handle lossy reproduction by definition. Persitant metadata doesn't.
Others have sufficiently trashed the parent on the other topics, so I'll foucs on the watermarking issue.
Modern audio codecs use psychoacoustics, which encode the sounds the human ear and brain can hear over the ones that we can't.
Watermarking works by putting imperceptible sound in the signal that can't be heard, but can later be extractable by computer.
See the problem?
A codec at a "good enough" data rate (where no apparent artifacts are heard), won't be at high enough data rate to encode a robust watermark.