32 bit Float in Samplitude: please explain how this works???

[Alex Wyler]

I am confused about what 32 Bit Float is, why it may be important to run my sessions this way, how to dither down to 16 bit from 32 Bit Float and what the benefits are?

When I start up a new project in Samplitude, i have two major options: the bit depth and the Sample rate. The choices are-

Sample rates: 44.1, 48, 88.2, 96, 192

Bit Depth: 16, 24, 32 Bit Float


Well, i understand what dithering is when going from 24bit to 16 bit and i understand sample rate conversion when going down to the standard 44.1 final sample rate from a higher one like 88.2 that many people like to record at these days to get a bit higher resolution from the get go.

But I do not understand what 32 bit float is and what things I need to know in order to get my final mix audio files to be 16 bits.

Please educate me as to what the advantages of running at 32 Bit Float over running at 24 bits is when recording all my audio files in my session?

What things do i need to know when using 32 Bit Float that will require some different procedures and steps taken that do not exist when running standard 16 or 24 bit projects?

Thanks for any help on this. You time helping me is very much appreciated!

[fifthcircle]

Internally, the summing engine in Samp works at 32 bit float (with some processes going way above that). The 32 bit float option is pretty much a legacy setting from the days before they could record a 24 bit fixed file.

without going into a large amount of detail, you'll get better fidelity recording at 24 bits rather than 16. Mix and then dither as the last step (ie in the bounce or burning the CD). Use 32 bit float only if you are going to be doing destructive effects and saving the settings. Otherwise, use 24 bits.

--Ben

[sleepwalker]

I wouldn't consider 32bit float a legacy option. If I'm not mistaken, it's still the preferred bit depth for Nuendo.

Originally Posted by fifthcircle

I
nternally, the summing engine in Samp works at 32 bit float (with some processes going way above that). The 32 bit float option is pretty much a legacy setting from the days before they could record a 24 bit fixed file.

without going into a large amount of detail, you'll get better fidelity recording at 24 bits rather than 16. Mix and then dither as the last step (ie in the bounce or burning the CD). Use 32 bit float only if you are going to be doing destructive effects and saving the settings. Otherwise, use 24 bits.

--Ben

[adzski]

Originally Posted by fifthcircle

Internally, the summing engine in Samp works at 32 bit float (with some processes going way above that). The 32 bit float option is pretty much a legacy setting from the days before they could record a 24 bit fixed file.

without going into a large amount of detail, you'll get better fidelity recording at 24 bits rather than 16. Mix and then dither as the last step (ie in the bounce or burning the CD). Use 32 bit float only if you are going to be doing destructive effects and saving the settings. Otherwise, use 24 bits.

--Ben

I understood that recording files at 32bit may provide better cpu performance as no realtime conversion needs to take place when using plugins.....of course that may well be hocus pocus

[ghellquist]

Originally Posted by adzski

I understood that recording files at 32bit may provide better cpu performance as no realtime conversion needs to take place when using plugins.....of course that may well be hocus pocus

I think you will be hard pressed to notice any difference in performance. Very hard pressed.

I fully agree with Ben, set it at 24 bits and let it stay there.

The only reason I can see (am a bit shortsighted though) in doing 32bit float is if you want to share the raw files directly with some application that cannot read 24 bit files.

Gunnar

[alfonso]

Working at 32bits float doesn't change anything regarding the dithering to 16 bits, the same algos can be applied.

I prefer using 32 float to 24, it gives more headroom when recording and processing and you don't use more CPU, a.f.a.i.k. the opposite.

my 0,02 €

[captain caveman]

If you use plug-ins, then 32 bit float is the way to go.

Not only does it give you more headroom than 24 bit, but it's impossible to clip your plug-ins (unless you are using a Powercore which uses 48 bit fixed).

So 32 bit float is better than 16 bit, 24 bit and 48 bit fixed for this reason.

[fifthcircle]

I can tell you for certain as a long time samplitude/Sequoia user... Recording at 24 versus 32 bit will make no difference in the sonic outcome of your product. It has been awhile since I've tested, but as of Version 6 when 24 bit was introduced, the difference between processing power required was miniscule- the biggest problem being with high track count projects, 32 bit causing disc throughput issues (it has 1.5 times the data rate).

Remember, even the best A-D converters have 22 bits of real resolution. Any more than that is just representing the sound of electrons bumping into each other. 122dB is beyond the threshold of hearing and is a theoretical number.

As I said before, the place where you'll get an advantage with processing is when you are working with offline processes as it requires you to write the processed file. Even then, the difference is pretty minor.

Now, please be aware the recorded file bit depth is completely different than the summing engine's bit depth. That bit depth is *not* user-selectable. Samplitude and Sequoia work internally at whatever it takes for the program to sound as good as possible. You can specify an output bit depth of a trackbounce or mix, but nothing else...

--Ben

[alfonso]

Originally Posted by captain caveman

So 32 bit float is better than 16 bit, 24 bit and 48 bit fixed for this reason.

I don't agree with the last part of this statement, I don't know Powercore directly, but 48 bits fixed and even 32 bit fixed have higher resolution than 32 float and are theoretically better, i.m.o. practically too.

As it has been said, higher bitrates are expecially helpful in processing and editing.

[captain caveman]

Originally Posted by alfonso

I don't agree with the last part of this statement, I don't know Powercore directly, but 48 bits fixed and even 32 bit fixed have higher resolution than 32 float and are theoretically better, i.m.o. practically too.

As it has been said, higher bitrates are expecially helpful in processing and editing.

If you are talking ITB though, then 32 bit float allows you to go over 0dbFs, way over - 100s dB. So you'll pretty much never ever clip your plug ins, although you lose 1 bit of resolution for ever 6dB above 0dB.

I am no expert in the maths behind it all, and I am just repeating the conclusions of other discussions but really when it comes to anything equal to or above 24 bits then there has to be another reason other than AD headroom.

No clipping plug-ins, little bit less to think about = all good.

[mogWai]

The problem with 32float is all that headroom. Your ADDA will only ever output (and record) 24bits max, so, if you want to be anal, your mix bus needs to be dithered before hitting the DA.

Some people dither their output to 24bits. I used to, but gave up after really honestly not being able to tell the difference between 32bit dithered and plain old 24bit.

I use Cubase. It will always run internally at 32bit Float - so all the extra headroom needed to process plug-ins is there, whether you need it or not. I always record 24bit and the project is always set to 24bit, I always mixdown to 24bit.

It has been shown and proved elsewhere on the net by Nika Aldritch that floating point is not necessarily the best way to handle the large maths, and that floating point is still prone to quantization errors.

I think he showed that fixed point is always better, but you need 48bit fixed point - like in Pro Tools - to achieve the best headroom without quantization errors.

[Alex Wyler]

So, is the concensus that I should run my Samplitude sessions at 32bit-Float or at 24 bits when loading a new project?

I'm still confused...there are some confliciting opinions here?

For those of you saying 24 bit, can you direct me to some published research that proves that 24 bit operation is overall better?

[fifthcircle]

Alex- Look at my original posts... Everything is explained there. You are asking what bitrate to record in. Unless you are doing destructive processing, there is NO advantage to recording at 32 bit float.

We are talking Samplitude here... All this stuff about other programs is pointless information to your question. In Samplitude, the 32 bit float file is 24 bits with a bunch of zeros at the end. Those zeros get you nothing in terms of plugin processing. You cannot choose what bitrate to process in Samp. Period... You can only choose input and output bitrates. (The output choice affecting dither for the file/stream)

--Ben

[mogWai]

Find out if Samplitude does all its internal calculations at 32bit regardless of what you set your project to.

If it does (I suspect it will) then just set your project at 24bit. This will mean you record 24bit files and thats the best you can do.

You only need the extra bits of 32bit float if you are making ANY changes to the recorded file in the mix bus - ie. changing the level fader, adding EQ etc.

You cannot record any more than 24bit, but when performing any process on the file, the wordlength (bit depth) will increase. So as soon as you move a fader you need 32bits to store the result. Samplitude should do this automatically, although like I said before, it still delivers a 32bit word to the DA converters.

Export you mix at 32bit float to capture all that extra processing.

[6strings]

Originally Posted by mogWai

Find out if Samplitude does all its internal calculations at 32bit regardless of what you set your project to.

That's the way Cubase SX does it as well. The only reason for recording @ 32 bit is if you want to track with plug in effects (so says the manual).

[Alex Wyler]

Samplitude Users:

Do you record at 32-bit float or 24-bit?

And Why?

Thanks!

[doubledecker]

It's very simple.
If you gonna do all your work at home ,work with 32 bits.
If you gonna take it elsewhere work with 24 bits.

[tvck]

Originally Posted by Alex Wyler

Samplitude Users:

Do you record at 32-bit float or 24-bit?

And Why?

Thanks!

Ben is correct - there is NO point in recording 32-bit float files, unless you are doing destructive editing.

The bit depth at which you choose to record, has nothing to do with the fact that, internally, ALL proccessing is done at a minimum of 32bit floating point.
Even if you record your WAV files at 16-bit, they are automatically run internally at 32-bit f.p. because that's what the sound 'engine' works at. ( as in Nuendo and others, also some use 64-bit f.p. now)

Do not confuse your recording bit-depth with the inernal processing bit-depth.

People record at 24 bit for a couple of reasons:
Firstly, compared to 16-bit, it gives a much bigger usable range of input level - you don't have to worry about whether or not the the signal level is 'hot' enough, so you can have the input peaking at only -20dB and still have 120dB of dynamic range left.

Secondly, a lot of people think 24-bit sounds better than 16-bit, but then again some others don't think it makes a difference to the sound.

Reasons to record at 32-bit floating point? as stated above, if you do destructive changes to the files you may be better off using 32-bit f.p. for your files. ( in effect, you are 'preserving' the files as they are used internally, i.e. 32-bit float)

---

[Alex Wyler]

So you are saying then that I will get the best sonic results by selecting 32-bit float as the recording bit depth for my sessions then correct?

You have confused the heck out of me. I am lost. Can someone please spell this out for me in lamens terms. I'm a bit slow..I know but I really want to get it?

My apogee Rosetta only goes as high as 24 bits so how the heck can Samplitude record 32 bits if the Rosetta only produces 24 bits?

And i still don't understand the advantage of 32 bit but i really want to.

Help!

[jjdpro]

Originally Posted by sleepwalker

I wouldn't consider 32bit float a legacy option. If I'm not mistaken, it's still the preferred bit depth for Nuendo.

Hence it's (nuendo) summing problem..Steinberg needs to adress this issue big time. 32 bit float is old..Older daws like ensoniq Paris uses a 56bit floating mixer(summing bus)..

[Farview]

Originally Posted by Alex Wyler

So you are saying then that I will get the best sonic results by selecting 32-bit float as the recording bit depth for my sessions then correct?

Only if you plan on doing destructive editing. Even then, it doesn't make much difference.

Originally Posted by Alex Wyler

My apogee Rosetta only goes as high as 24 bits so how the heck can Samplitude record 32 bits if the Rosetta only produces 24 bits?

You are recording 24 bits of information in a 32 bit long word. It just adds zeros as placeholders. Once you do some destructive offline processing, those extra eight bits will become usefull.

Originally Posted by Alex Wyler

And i still don't understand the advantage of 32 bit but i really want to.

There is no such thing as a 32 bit converter, so you can never record or listen to 32 bit audio. It is only usefull for digital processing. The idea is to get all the math errors down in the least significant bits where the get thrown away.

The other advantage to 32 bit float (actually, just floating point math in general) is the headroom. As the signal tries to go above the ceiling, the ceiling gets moved so you don't clip the signal.

This does not mean that you can't clip the signal when recording in 32 bit fp. Remember, your converters are only 24 bit. It's very possible to clip them.

Again, there is only a slight advantage to recording 32 bit fp IF you do a lot of destructive offline processing. If you just use inserts and sends in the mixer to apply all your effects, 24 bit is the way to go.

[taturana]

32 bit ...maybe for cpu reasons... but for sound it really doesn't make much of a difference...... the sound diff comes mostly from how the daw handles the sound...
for sonar i keep it in 24 bit...... i guess the process is happening in 64 bit(or 32 bit float in the vst plugins) ... i always set wavelab for 32 bit float temp files though....

[aussie_techie]

incomming signal from a sound is 24bit max, usion 32bit float files just pads out the file with 0s making it bigger for no real reason. the file bit depth has nothing to do with the mix engine bit depth.

[johnzorzi]

Set your recording bit-rate to 24 bit!

REASONS:
]] AD converters only reach this level

]] Any files created will be easier to convert/recover in case of problems

]] Recording to 32-bit float will only be different to 24 bit (apart from 8 zeros) in particular cases (sub-mix render, recording with plugs etc) and the difference is not worth the disk space anyway

]] I say so. Now, ignore or follow at your own choice

]] Really, mix choices about whether to process or not, how you EQ, what plugs you use will have a MUCH bigger impact on your sound - don't worry about it when it gets to details this fine. Spend the effort considering this question on mic placement instead - as I should be doing, bye now..

[Sequoia Berlin]

]] Any files created will be easier to convert/recover in case of problems

I am afraid this one's inaccurate. The 32bit float format is at least as well specified as the 24bit option.
There are many confusions how to use the fact chunk with 24bit audio files, as one example where 24bit files can cause problems in programme exchange even today. Other issues are the unclear byte alingment in case of using the link/continue chunk.
There is probably no preference as far as future retrievability is concerned.
The majority of developer libraries handles 32bit WAV files, only a few handle 24bit.

Cheers,

Sebastian

[Greg B]

Originally Posted by Alex Wyler

So, is the concensus that I should run my Samplitude sessions at 32bit-Float or at 24 bits when loading a new project?

I'm still confused...there are some confliciting opinions here?

For those of you saying 24 bit, can you direct me to some published research that proves that 24 bit operation is overall better?

The nice thing about Samplitude is that you can choose how you record each individual track. Samplitude supports multiple bit rates within the same project. If you wanted, you could record all the drums and guitars at 24 Bit, and all the Vocals at 32 Bit Float. Depending on how you will be processing the individual tracks, something like that might make sense. I record and export my vocal tracks in 32 bit float and bring them into Melodyne for tuning and other processing. Then bring them back into Samplitude.

I also mix down the final project to 32bit float for mastering (if it will be mastered in Samplitude). If it's going to a ProTools studio for mastering, then I send a 24 bit file.

[d_fu]

Originally Posted by Greg B

The nice thing about Samplitude is that you can choose how you record each individual track.

Unfortunately, you can not do this when recording multiple tracks simultaneously. Such a feature has been on my personal Samp wishlist for years... But they never included it in any update version...

[JMWS]

From the Steinberg´s Internal Mixing tutorial, the benefits of 32bit:

More bits sound better!

No clipping

24bit clips at 0dbfs

Intel and AMD chips run in 32 floating point, require less threads, and therefore they require less cpu power in 32 float as no conversion to 24 or 16 is necessary.

Internal computations in Cubase are 32 float, therefore less cpu hungry.

No dithering required until mastering process.

No rounding errors caused by truncation. i.e. less degradation of the signal through processing.

Because truncation is avoided, precision in the quiet aspects of the sound are greater i.e. in that part of the sound where the spatial information is (reverb) = more spatial sounding mix.