64-bit Mix Engine Question

I am confused. You didn't say if you are working in 32 bit or 64 bit Sonar! If you are working in 32 bit, the old sales pitch from Cake said that you would have some benefits with the 64 bit engine checked. If I remember correctly, the noticeable differences would be in the reverb tails being smoother, delay trails being smoother and clearer, but no real discernable difference in the core 32 bit audio.
Since I switched to 64 bit, I can't even get some of my projects to load in 32 because I'm using past the 32 bit RAM limit. Plug in instruments have gotten bigger these days! Even drum programs have become huge with the advent of 64 bit.
I hope that I am remembering correctly and that helps you out.

SonicExplorer
Asked another way, I assume I'm hearing 32-bit during playback and am not experiencing the benefits of the 64-bit mix engine, is that right? 

 
Not really.
 
When playing back what you hear is dictated by your interface, the majority of which operate at 24 bit and in the real world, this can mean the effective bit depth is around 20
 

SonicExplorer
 
As for a difference in audio quality, I could have sworn when I ran preliminary A/B tests when S5 was first released that I could hear more clarity and smoothness.  Just an overall better sounding result with the 64-bit mix engine. I want to say the result was smoother and punchier, just overall higher-end professional sounding.  It was subtle but if you are a picky person it was definitely noticeable. But I simply can't recall back that far to be honest.
 

If you have a single track that started in the analog domain (including sample libraries!), it's inherent noise will absolutely bury the level at which 32 bit calculation errors start to appear. 
 
Since in this case the errors do not accumulate over time and do not accumulate rapidly, there are simply not enough calculations done on a single track in the mix engine* to raise the error level enough to be anywhere close to the noise already in the tracks, much less audible.
 
And obviously adding multiple tracks like this together will not make calculation errors add faster than the noise from the individual tracks.
 
In fact most of the 32 bit calculation errors won't even make it into 24 bit audio output, much less be above the noise floor due to other causes, much less be audible.
 
You will find people who believe they hear a difference, but inevitably they are not using a reliable method of comparison and what they claim to hear and the circumstances where they claim to hear it simply do not match what the calculation errors would sound like if they were audible.
 
 
* there are circumstances where 64 bit calculations are necessary due to the extreme number of calculations done, but this is not one of them.

SonicExplorer
After pondering this further, I *think* "mix engine" may not be a good term?  Audio engine maybe is better? In that context, I would be hearing the 32 bit engine if the selection is unchecked, and 64 bit otherwise. 

Actually "mix engine" is the better term, because word length only comes into play when mixing two or more tracks. If you're just playing back a single track with no modification, then there is no benefit to greater mathematical precision. It's only when modifying the audio by summing tracks or using a multiplier to alter their levels (e.g. faders, compressors, EQ) that those calculations can affect the outcome.
 
Mathematically, it's actually quite similar to how your property taxes and bank interests are calculated: when real dollar amounts are multiplied by some rate multiplier, the results are stored in imaginary currency (millicents, or thousandths of a cent). These are intermediate values that will never appear on a tax bill or bank statement. That extra precision reduces the overall rounding error, so that when those numbers are ultimately expressed in real currency the cumulative error will be less than a penny.
 
32- or 64-bit calculations do that for audio, making errors insignificant to the final result. As noted above, by the time it gets to your audio interface it's essentially 20- or 16-bit resolution anyway. Given that even 32-bit audio is overkill, there isn't even a compelling reason to use the 64-bit engine. You will not hear the difference, any more than your bank statement would be different if the bank used microcents instead of millicents.
 
 

Hmmm...  I did a lot of reading on the whole 32 vs. 64 bit mix engine last night and still came away not knowing what to believe.  On paper it seems not realistic to hear a difference in most scenarios, but in practice some people claim to hear it.  
 
If enough things are rounded, eventually the decimals can add to pennies which can add to dollars, it all depends on how you group and round, and how many times that transpires.  Being an old school analog relic I just don't know enough about the inner workings of this digital stuff.
 
And not to derail my own thread, but I also read a comment last night that said XP sounds better than W2K.  Which really left me scratching my head as I was unaware one OS version from another by itself could change the quality of sound.  Huh??

SonicExplorer
Hmmm...  I did a lot of reading on the whole 32 vs. 64 bit mix engine last night and still came away not knowing what to believe.  On paper it seems not realistic to hear a difference in most scenarios, but in practice some people claim to hear it.

 
In some practice people claim to hear the differences in cables, even though we can prove in blind testing they can't (and prove they are the same on a 'scope). People claim all kinds of stuff they can't substantiate. They want to hear a difference, so they do. 
 

If enough things are rounded, eventually the decimals can add to pennies which can add to dollars, it all depends on how you group and round, and how many times that transpires.  Being an old school analog relic I just don't know enough about the inner workings of this digital stuff.

 
That is the reason for having something like a 64-bit engine: Insurance against rounding. With 32-bit, rounding errors will occur. 32-bit FP is actuall 8-bits exponent, 24-bits significand. That means it has 24-bits of precision over a range of 8-bits, more or less. So when working at normal audio levels, assuming 24-bit input and output, you do get minor rounding errors. However they are going to be on the order or 1, maybe 2-bit at most in all likelihood. So you don't worry about it. Even the best DACs don't resolve more than maybe 21-22bits of signal and other kinds of noise will be WAY higher.
 
However in theory, enough operations, particularly ones that involve things like multiplication or division in sequence, could increase the rounding error. It doesn't really happen in practice, but it could in theory. Well we can easily prevent that just by going double precision. Now we have an 11-bit fraction and 53-bit of significand. That means we have so much precision we will never have any rounding error ever reach the 24-bit output signal, even in edge cases. Since modern processors kill at double precision performance, no real reason not to use it.
 
If you want to see a case where it does matter, look at graphics. Most monitors and video formats are 8-bits per channel output. However you need to do processing in 32-bits per channel (floating point) if you want truly smooth colour gradients. Do it in 16-bit half precision and you can see errors sometimes.
 

And not to derail my own thread, but I also read a comment last night that said XP sounds better than W2K.  Which really left me scratching my head as I was unaware one OS version from another by itself could change the quality of sound.  Huh??

 
For sound going through the OS layer? Sure. Remember that unless you use something like ASIO to talk straight to the soundcard, the OS is modifying the sound. It runs it through a mixer and resampling engine to allow for more than one program to play sound at once. Sometimes does other stuff too. How good or bad it does that can vary. In general newer OSes are better than older ones since there is more CPU power these days so they use better algorithms.

Nothing really goes "straight to the hardware", but it doesn't matter. Extra layers add to overall latency, but shouldn't affect harshness or clarity. Wish I could tell you what your problem really is, but all I can say is that whatever issues you're experiencing they can't be ascribed to the drivers or the O/S. 

The actual sound changed in Windows 2000 to XP which was it became WASAPI which means it operates at a different application level, giving you direct access to the hardware before it passes information through the application layer. Something Technet-ish and geeky anyway.
 

@drewfx1, wow, thanks. So I will avoid any A/B testing then as I don't need to be chasing my tail with something that won't be a fair/stable test.  
 
@interpolated, about the W2K/XP thing, this is rather important as I plan to soon build one last backup system and am on the fence about which to choose for OS.   Just because something bypasses the app layer doesn't mean the app layer did something to alter the sound, other than possibly more latency.  I'd need to understand more about WASAPI and any other thing that changed to learn if XP tends to produce superior audio quality over W2K.  I'll try searching around on the forum and see if I can learn anything about it.....
 
BTW, will Sonar 5 and 6 be able to utilize both CPU cores on XP (Core 2 Duo) ?  

Okay, I've spent quite a few hours searching and reading about this W2K/XP/WASAPI/OS sound situation.  Seems to me, when it comes to quality sound cards, their affiliated drivers and reputable DAW software (such as Sonar) there is generally no sound loss in practice due to OS.  Any notion of a kmixer like situation is typically in those cases where the receiving entities accept a common conduit (and therefore potentially combined or altered) signal path.  Otherwise the data mostly passes through (or around if IOCTL or other mechanism if supported) and goes "direct" to sound card.  I say this with understanding there is an I/O OS stack involved.  Slight latency would/should be the only typical impacts.  Yes, in theory there could be buffering etc, but nothing that should alter the actual sound quality per-se.
 
Anybody disagree with this general conclusion?   And BTW, along these lines I found nothing that provided evidence or explanation of XP being superior in sound over W2K (in context of quality cards/drivers/DAW software such as we are discussing - RME FF, WDM Stream and Sonar more specifically)
 
Sonic

SonicExplorer
Ok, thanks, OS not likely impacting sound quality.
 
I could swear however I recall hearing a subtle but noticeable and important (to me) difference between 32 and 64 bit mix engine.  I also now vaguely recall making two mixes and comparing them.  Will check around and see if I can find those files....will post back if I discover anything definitive....
 
Sonic

Ideally, as anyone would agree, higher resolution is typically better, but other factors may affect this. For example, some A/D converters marketed to convert up to 192Khz may really perform better at 96Khz. So, then, what's the advantage of 192kHz? I would argue, if you can hear a difference, there may be something (plug-in or D/A conversion) that affects the quality of the different bit rates.

When thinking of 64-bit vs. 32-bit, some amazing sounding mixes were done totally in a 32-bit environment. Some amazing recordings have been made from 16-Bit, 44,100 kHz bit/sampling rates. Partially, the reason is that these factors are far less important than good engineering and tracking skills.

Whether or not you're a Yes fan, you should go listen to Yes' Talk album sometime. It was recorded using two Macs in tandem (I read) 1993/94, and they had many issues with the technology, per some Chris Squire interviews. But the album (albeit, sterile) is pristine in quality. That was some old technology. It was likely only 16-Bit, 44Khz as tracked. I'd be surprised they even had dithering technology then, because they stayed at the 16-Bit rate.

When it comes to this subject now with 64-Bit, Sonar includes the 64-Bit Double Precision Engine for projects that contain audio (or all audio) rendered to 64-Bit. As the Sonar documentation alludes, no recording devices really record at anything higher than 24-Bit. Most projects we work on today are 16/24/32-Bit when inside Sonar, but never sampled higher than 24-Bit. So, why do we even need 64-Bit Precision? Only if you're working with a project that was 64-Bit to begin with. This option is disabled by default, because for most purposes, you simply don't need it. 

Cakewalk could simply be future-proofing their engine as perhaps sampling rates for A/D conversion and VSTi's will become that precise and dynamic.

the more itb processing you do the more worthwhile it becomes

drewfx1

pwalpwal
the more itb processing you do the more worthwhile it becomes

So please tell us how much is too much and how you arrived at that particular amount.

it's purely maths - 64 is more accurate (the decimal places go further) - same reason to do itb processing at > 48k SR even if you can't hear it ;-)

The short and the long answer are both no.
 
R

Technical vs. practical is a condition not met in many people's minds--not calling anyone out, just in general populations. If you think of Limits in Calculus I, there are many things we can sample nearly perfectly but never ever reach absolute perfection. When it comes to sampling audio, for example, you can never sample fast enough to make the sampled audio absolutely a perfect match to the source.
 
In the case of 32 vs. 64 bit, sure, a calculation can be infinitely more accurate, but when do you stop hearing the difference? At what point does Pi become easier to round off to be effective? When it comes to tape, as in full analog, if 32-bit audio is audibly better, what's the true benefit of going to 64-bit for most people?

I won't downplay the benefit of 64-Bit (I export at 64-Bit precision), but I don't see the benefit of mixing with the 64-Bit double-precision engine.