Skip to main content
Topic: Which is the best lossless codec? (Read 373572 times) previous topic - next topic
0 Members and 9 Guests are viewing this topic.

Which is the best lossless codec?

Reply #350
Just to put in perspective what fast encoding is about.
NIN The Downward spiral CD1, i5-3570@4.4, GTX-970

453.066.564 Bytes ~3sec. flacCL (open source) -6
453.725.634 Bytes ~13sec. tta


...as you wrote above, it can be true only "on capable hardware". I think is not relevant to codecs comparison.

All modern CPU/GPU combinations support OpenCL meanwhile. Even my integrated HD4000 graphics does nicely with flacCL. I have no benchmarks though.
But you are right with flacCL there are many more factors to consider especially when going multicore.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!



Which is the best lossless codec?

Reply #353
It more appropriate to list features that are only common to a very small number of subjects under the dedicated sections for those subjects.

If you want to make exclusive features stand out like the way products are often marketed, do it on your own site.

FWIW, I've held this position on other articles as well.
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

Which is the best lossless codec?

Reply #354
Below is newest results of comparison of TTA codec and TAK (Note that I have mentioned before, that I prefer to comparison open source codecs only).
The album has been chosen randomly. I have found good hardware for this test: Intel celeron 2.40Ghz RAM 1Gb. OS Windows 7, swap file is disabled. Here is shown three results of testing.

I still don't get it: Are you interested in the maximum encoding speed? Then why don't you test TAK's fastest preset -p0 ? Or are you looking for the best compression at a similar speed? Then the compression ratio is missing. Anyhow, results of a single album don't mean much.

And indeed: You have found good hardware for this test! Im sure the celeron is a Pentium-4-class-type. There have been several reports (here at hydrogen) that TAK's optimizations are counterproductive for this microarchitecture. A heavy speed penalty is the result. I never really cared because i always disliked the P4 microarchitecture and didn't want to modify TAK for a dying architecture Intel dropped because of it's inefficency. I don't kow -and never received reports- about another microarchitecture that at least supports the MMX-instruction set (1997) where TAK suffers that much.  Your results are only relevant for people using a P4-type-cpu.

...as you wrote above, it can be true only "on capable hardware". I think is not relevant to codecs comparison.

You yourself used hardware capable to slow down TAK for the comparison. Possibly unintentional. But somehow i have the feeling you are cherry-picking...



Which is the best lossless codec?

Reply #355
Album: Abba "The Visitors" (Deluxe Edition), 16 tracks, 783.392.206 bytes
CPU: i7 4770K @ 3.5GHz

TTA 3.4.1
443.763.826 bytes, ~16 sec, 56.64%

TAK 2.3.0

-p0  448.085.906 bytes,  5.73 sec, 57.20%
-p1  440.034.353 bytes,  7.08 sec, 56.17%
-p2  433.761.465 bytes,  9.75 sec, 55.37%
-p3  431.268.735 bytes, 19.50 sec, 55.05%
-p4  429.712.339 bytes, 32.03 sec, 54.85%

Which is the best lossless codec?

Reply #356
Then why don't you test TAK's fastest preset -p0 ?


This is obvious. I have compared the codecs in modes, which gives same or better compression ratio.

Which is the best lossless codec?

Reply #357
Album: Abba "The Visitors" (Deluxe Edition), 16 tracks, 783.392.206 bytes
CPU: i7 4770K @ 3.5GHz
................

Yes. It can be true. But all details is significant. I have this album and I can reproduce your test on several different hardware platforms (in monday). The results will be very different.

...as you wrote above, it can be true only "on capable hardware". I think is not relevant to codecs comparison.

You yourself used hardware capable to slow down TAK for the comparison. Possibly unintentional. But somehow i have the feeling you are cherry-picking...


I have used old hardware for my tests (Intel celeron 2.40Ghz RAM 1Gb) especially to avoid the effects of compilers optimization (sorry, I have no more neutral hardware). Otherwise you will compare the work of compilers and optimization for "capable hardware", but not the codec algorithms. All what I want to say, that the TAK compiler is more optimized for new processors. That's the reason why hardware and OS for codec comparison must be specially selected and the procedure must be documented in details to get independent results.
I'm not against the comparison of codecs optimization for selected hardware platform, but it's a different tests.

Which is the best lossless codec?

Reply #358
What is the "most neutral software"? Pentium Pro without MMX/SSE support?

P.S. I think that in comparison tests we compare encoder performance anyway, and not codec algorithms (whatever it means).

Which is the best lossless codec?

Reply #359
I have used old hardware for my tests (Intel celeron 2.40Ghz RAM 1Gb) especially to avoid the effects of compilers optimization (sorry, I have no more neutral hardware). Otherwise you will compare the work of compilers and optimization for "capable hardware", but not the codec algorithms.


As far as I know from following the development of the codec here at hydrogenaudio over the years, the codec does use lots of assembly code paths. So your "avoid the effects of compilers optimization" goes way beyond your expected outcome if the codec can't use the code that has been specifically written to use the better hardware. 
And why does that matter??? Because the development did imply decisions where an improvement in compression would cause a tradeoff with speed, so it prompted an optimization of the code path for speed.

What this really mean is that if you end in this road, you are comparing a codec (yours) optimized with a 10 years old compiler optimized for a 10 year old hardware, versus a codec (TAK) optimized (by writing code explicitely) with current compilers for current hardware.  There is no way to pretend a fully fair comparison if you go that route.  The only way would be to rewrite both in a common language , compile them with the same compiler and run them in the same hardware... Not feasible.

Which is the best lossless codec?

Reply #360
I have used old hardware for my tests (Intel celeron 2.40Ghz RAM 1Gb) especially to avoid the effects of compilers optimization (sorry, I have no more neutral hardware). Otherwise you will compare the work of compilers and optimization for "capable hardware", but not the codec algorithms.

End-users don't use "algorithms", they use codecs. Codec = specific implementation of some algorithm. It may be efficient or not, it may have some optimisations for modern hardware or not have it. If it doesn't have ones, it's a "problem" of this codec and it's developer, who is unable or too lazy to implement it.

Quote
All what I want to say, that the TAK compiler is more optimized for new processors.

Ok, I re-tested TAK without MMX and SSSE3 optimisations (command line key -cpuNone). Just plain old Delphi code. The album and it's size are the same.

-p0 -cpuNone 448.085.906 bytes, 8.06 sec, 57.20%
-p1 -cpuNone 440.034.353 bytes, 13.13 sec, 56.17%
-p2 -cpuNone 433.761.465 bytes, 21.99 sec, 55.37%

Even with all "modern hardware" optimisations off TAK -p1 wins in speed and compression level. And again - common end-user usually uses default settings where all optimisations are switched on.

Which is the best lossless codec?

Reply #361
I have used old hardware for my tests (Intel celeron 2.40Ghz RAM 1Gb) especially to avoid the effects of compilers optimization (sorry, I have no more neutral hardware)

Okay, sure. What about this, it's an old test (with an older version of TAK) I did on my AMD Turion 64 ML-34, which was introduced in March 2005, two years before the last release of TTA. That seems neutral (meaning: not giving TAK advantages through the instruction sets that were introduced after the last TTA release) to me. Tests were done from RAMdisk, and the CPU was clocked down to 800MHz to make sure caching and other hard disk stuff do have as little influence as possible.



This is neutral, still TAK is a clear winner here.


Music: sounds arranged such that they construct feelings.

Which is the best lossless codec?

Reply #362
once more, someone has been changing stuff in the Wiki without discussing. Changes

- Changed table: WavPack hardware support from limited to very good
- Changed table: WavPack encoding speed from fast to very fast
- Changed text according to changes in table

I really don't think that Wavpack hardware support can be classified as very good. Furthermore, Wavpack isn't classed as 'very fast' when checked with the criteria mentioned as comments in the table markup.
Quote
- Encoding speed is very fast if > 150x, fast if >75x, average if >40x, slow if >20x, very slow if <20x.
- For decoding speed thresholds are doubled, i.e., very fast if >300x, fast if >150x etc.
- Thresholds for compression are at 56% and 58%
- Speed and Compression are based on each encoder's default settings and are taken from the this comparison

Should I undo those changes?
Music: sounds arranged such that they construct feelings.

Which is the best lossless codec?

Reply #363
Yes.
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

Which is the best lossless codec?

Reply #364
It's all is Ok, but can we close the first question. I did not see any objection about placing of "Password protection" line into comparison table. Can I restore this line?


If you seriously didn't see any objection  ... am I the only one who thinks this just underlines the argument that the dev is too biased to judge? 

Good that you asked, in the very least.  But let us wait for consensus instead I think.


I went on record as objecting and ktf obviously objects.


If explicitly going on record as objecting is what it takes, then count me in.


And I think it is safe to revert the changes regarding WavPack too.  Maybe we could discuss what should be the baseline for "very fast" encoding speed and so forth, but I do not see the point now when there will surely flow around accusations of moving goalposts.
High Voltage socket-nose-avatar

Which is the best lossless codec?

Reply #365
For me, there's only one level of compression (why should I care about ~5% difference) and only two levels of encoding speed: "fast enough" and "too slow".

Which is the best lossless codec?

Reply #366
The album has been chosen randomly.
[...]
Deep Purple, Machine Head. 2ch/16bit/44.1KHz


Randomly? I bet it has been chosen according to your (good!) taste :-)


Look, there is a good reason or three to let neutral people do the testing. Even when you and TBeck and Bryant are presenting completely honest tests, they are still likely to be greatly influenced by what you have developed your software for and optimized it on. I take it that TBeck doesn't listen much to Japanese noise music when you see results like http://www.hydrogenaud.io/forums/index.php...st&p=837834 . Letting the developer choose the material does lead to cherry picking of data.

That does not require any sorts of dishonesty. I do not need to accuse you of pulling a Microsoft (i.e. doing all sorts of tests in-house and then hiring someone "independent" (as if!) to do precisely the test for which Microsoft knows they will come out winning), I just need to "accuse" you of choosing what you think is relevant to test, and that having a positive correlation with what you think is relevant to actually compress.

Edit: quotation marks for ... uh, "clarity".
High Voltage socket-nose-avatar

Which is the best lossless codec?

Reply #367
For me, there's only one level of compression (why should I care about ~5% difference) and only two levels of encoding speed: "fast enough" and "too slow".


For me, there are not only more than one level of compression, there is also more than one dimension of interest; like, "what do I need" and "what impresses me".

TAK impresses me. Really. But I am not using it, and I doubt if it gets a position where I will migrate to it.
High Voltage socket-nose-avatar

Which is the best lossless codec?

Reply #368
Maybe we could discuss what should be the baseline for "very fast" encoding speed and so forth, but I do not see the point now when there will surely flow around accusations of moving goalposts.


And also what the baseline for "hardware support" is.  Rockbox gave wavpak support to the overwhelming vast majority of all DAPs ever sold (at least in the western world (did "MP4" player numbers ever exceed iPods?)) and in today's era the overwhelming majority of music reproduction devices can, through apps, play the overwhelming majority of formats.

Are only stand-alone-no-third-party-software devices like to be counted?  Only stock OS/software capabilities to be counted?  Perhaps toss the whole column in this modern era?
Creature of habit.

Which is the best lossless codec?

Reply #369
and in today's era the overwhelming majority of music reproduction devices can, through apps, play the overwhelming majority of formats.

What about HiFi components, do they too work with such apps these days? (That's a sincere question, I'm not really up to speed) See for example this list: http://xiph.org/flac/links.html
Music: sounds arranged such that they construct feelings.

Which is the best lossless codec?

Reply #370
What about HiFi components, do they too work with such apps these days? (That's a sincere question, I'm not really up to speed) See for example this list: http://xiph.org/flac/links.html


And that's a legitimate question I don't have the answer to.

It is along the line of what I was trying to get to, though.  Are we talking "hardware support" or "one specific class of hardware we probably should specify" support?  Because, outside an ever-shrinking niche of (at least) the home-audio world, there is no de facto or de jure hardware limitations when it comes to formats.

Creature of habit.

Which is the best lossless codec?

Reply #371
and in today's era the overwhelming majority of music reproduction devices can, through apps, play the overwhelming majority of formats.

What about HiFi components, do they too work with such apps these days? (That's a sincere question, I'm not really up to speed) See for example this list: http://xiph.org/flac/links.html


Going forward, probably FLAC support ends up in almost everything, since its built into Android, and Android is essentially becoming the defacto embedded operating system for everything from audiophile audio players to televisions.  Even for devices that aren't android, they likely run SOCs designed for Android, and so often come with a flac codec in DSP firmware somewhere.  Of course, many of these chips and systems are rarely redesigned, so we may still be seeing products designed in the mid-2000s around WMA Standard and PlaysForSure DRM for a long time yet.

Which is the best lossless codec?

Reply #372
I fixed the error handling entry for Monkey's Audio.

How was it determined that it doen't support replaygain?  So long as the format supports tagging, doesn't it boil down to the player?
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

Which is the best lossless codec?

Reply #373
It is along the line of what I was trying to get to, though.  Are we talking "hardware support" or "one specific class of hardware we probably should specify" support?  Because, outside an ever-shrinking niche of (at least) the home-audio world, there is no de facto or de jure hardware limitations when it comes to formats.

Well, I think the issue is with the blurring line between hardware and software. I would consider something hardware support when it is embedded in the hard- of firmware. One could argue that vanilla Android is 'firmware' and all apps are 'software', which would make FLAC 'hardware' supported (as support is in the firmware) and other lossless formats 'software' supported. But that again, how 'firm' is the firmware of smartphones these days?

I'd say we see smartphone/tables support as purely software instead, as they are actually taking over many tasks that PCs were used for. Hardware then mainly consists of DAPs, home audio and car audio. There are of course other niches like DJ gear and recording devices.

Being the devil's advocate for the current status quo, one could argue that hardware support means whether a codec was made for low-power processors, i.e. decodes really fast. 
Music: sounds arranged such that they construct feelings.

 

Which is the best lossless codec?

Reply #374
I would hope for Potter Stewart to guide me on the distinction between "hardware" support and "software" support.

From the end-user point of view, it is rather: what can I play without having to climb a learning curve? The answers are likely WAV, and then ALAC-in-MP4 or FLAC (depending on whether you want to touch fruity hardware or not), and then the other of the two latter; I suppose that what determines the goalpost for "very good" would be whether it is "very good for a lossless compressed audio format" or "very good for an audio format". That is not godgiven either, it depends on user base; if lossless audio compression is a niche thing, then lossless compression could be assessed relative to lossless compression. Once hi-rez is about to penetrate the average Joe market segment, then average Joe would think that "very good" means at least "at least not too far from MP3".
High Voltage socket-nose-avatar

 
SimplePortal 1.0.0 RC1 © 2008-2019