Should I choose 32 Bit float instead of 24 bit in Cubase in 'project setup'?

[cocell]

Yes. The more bits the better accuracy you get mathematically.

[MadforBrad]

32 bit float and 24 bit have the same level of accuracy.

The 8 xtra bits serve as a mantissa to raise the available headroom but the actual accuracy is the same.

But ya, i would chose the 32 bit float.

[DAW PLUS]

Choose 24 bit:
-the mantissa is "added" during processing, no difference between a 24 bit and 32 bit file.
-24 bit takes less space
-24 bit files can be opened by any other application, while 32 bit files can't.
-there is no measurable extra processing power used for the 24 bit to 32 bit "translation".

[LittleJohn]

Originally Posted by MadforBrad

32 bit float and 24 bit have the same level of accuracy.

The 8 xtra bits serve as a mantissa to raise the available headroom but the actual accuracy is the same.

But ya, i would chose the 32 bit float.

I dont think so, I think best to choose 24 at the project level. the intenal engine will run 32 float regardless of the project setting, and 24 is better for file size and compatibility... in other words there is nothing to gain from setting the project file size to 32.
USE 24.

[MadforBrad]

Originally Posted by LittleJohn

I dont think so, I think best to choose 24 at the project level. the intenal engine will run 32 float regardless of the project setting, and 24 is better for file size and compatibility... in other words there is nothing to gain from setting the project file size to 32.
USE 24.

well for people that don't want to have to worry about red lining a channel, 32 bit float will take care of that. I don't know for certain if every DAW will always run at 32 bit float. Surely some offer the option which is why someone is asking this question. He is clearly referring to project setup. Running at 24 bit will make gain staging crucial.

Since the accuracy is the same, why not. I suppose it is a tad lazy but at least it is one less thing to worry about. And as far as file size, not really a concern is it ?

For the dude who said 32-bit float can't be opened in other applications...

Sorry you are wrong.
Here's a tiny improvised list of a few programs that can open 32-bit float:

Audacity
Wavosaur
Energy XT
Goldwave

DDclip Pro
MultitrackStudio Pro Plus
Cool Edit 2000 (now Adobe Audition)

I'm sure there are many more.
I'm sure any DAW worth it's weight can open them. I'm guessing that Reaper could probably open them too.

Also, the range for 32-bit float is many many many many more decibals greater than even 24-bit and 32-bit int. Go with 32-bit float if you have the disk space.

[Sk106]

Cubase operates at 32 bit floating point internally, no matter what the project standard is set to. The project standard is more about what resolution the audio is stored in and played back in. I've found that it doesn't affect computer performance much, although Cubase would theoretically have to convert alot realtime, since it operates in 32 bit floating point internally.

The basic difference between 32 bit float and 24 bit fixed is that 32 have a higher potential accuracy for small detail, since its mantissa is dynamic. For example, if you drag a sample point manually, moving it in 24 bit fixed you'd see the samplepoint 'snap' to the fixed grid, jumping stepwise from gridline to gridline. With 32 bit floating point it won't snap, but glide, meaning you can position the point almost wherever you please. It does snap to a grid, but a grid much more flexible than 24 bit, which is fixed. It will adapt itself dynamically to the input, theoretically allowing for a more accurate reproduction of the dynamic spectrum.

Other than that, 24 bit fixed point and 32 bit floating point is pretty much evenly matched in benefits and disadvantages. They both offer about 140dB range for example, and the output resolution is equally a matter of interface operation and driver. 24 bit fixed is the industry standard though, and if you're slightly ahead of that you need to remember that you are; you need to adapt to that, rather than it to you.

If you were on Protools I'd say use 24, since Protools works at 48 bit fixed point internally (double 24), but you're on Cubase, so I say go 32. With cubase, you won't exchange entire project files anyway, and you can downsample files if you deliver to someone else who can't read 32 (just use a damn good SRC).

[-tc-]

Originally Posted by astroidmist

I'm sure any DAW worth it's weight can open them. I'm guessing that Reaper could probably open them too.

Yep it can, and also 64-bit float ... it's Pro Tools and Logic that require 24-bit fixed.

However, I wouldn't be surprised if Logic's freeze files are 32-bit float.

Originally Posted by Sk106

24 bit fixed point and 32 bit floating point is pretty much evenly matched in benefits and disadvantages. They both offer about 140dB range

This is incorrect. Check your research.
Here's one site: Bores Signal Processing - Introduction to DSP - precision and dynamic range
It has the math to show the dynamic range of 32-bit floating point to be in the thousands of decibels. Othere websites corroborate this info. Look it up.

For floating point format, though, the binary point is moved automatically to accommodate larger numbers: so the dynamic range is determined by the size of the exponent. For an 8 bit exponent, the dynamic range is close to 1,500 dB: So the dynamic range of a floating point format is enormously larger than for a fixed point format:

The on-screen display range of 32-bit float may be about the same as 32-bit integer, but the actual data of the file can go beyond 0 dbfs of 32-bit integer. For this reason, if the software supports it, (as Cool Edit 2000 does), you can normalize a 32-bit float file DOWNWARDS to 0 dBfs if the peak goes beyond 0 dBfs. I have done this many many times as importing my 32-bit float DAW export files into CoolEdit 2000 is a regular habitual activity for me. It's also written about extensively in the Cool Edit 2000 .hlp file. If the software doesn't support full 32-bit float, the data just clips at 0 dBfs and can't go any higher. This is the case for 32-bit and 24-bit int.

[Fredrik]

In practical terms it doesn't matter though, since the internal representation is still (at least) 32 bit float anyway and the precision is the same. The only part where it would matter is if you wanted to save an intermediate state, such as when freezing audio.

[EastWest Lurker]

[QUOTE=timlloyd;6695817]

However, I wouldn't be surprised if Logic's freeze files are 32-bit float.[/QUOTE

They are.

[DAW PLUS]

Originally Posted by astroidmist

For the dude who said 32-bit float can't be opened in other applications...

Ugh, I meant not by all applications, thanks for catching that. There are some hosts which have issues with 32 bit float, especially older ones. Also, standard players like WMP and Quicktime can't open them.

Since 24 bit files are still processed in 32 bit float, it doesn't matter. The range remark you made is simply not true. 32 bit float = 24 bit + 8 bit mantissa.

[dubrichie]

let it go Leon.

you tried, fair play to you again, but you simply cannot take responsibility for the education of the masses of GS members who don't actually know what they are talking about. (and often don't listen either)

to the OP:

if it makes you feel better, record your audio as 32-Bit files, but there is no practical advantage to doing so.

[-tc-]

For recording, no, there isn't. For interim processing stages such as freeze files, stems etc... there can be advantages.

[robertshaw]

Originally Posted by dubrichie

there is no practical advantage to doing so.

With regards to plugins that is untrue.

[Lights]

A while back Steinberg released a mixing tutorial video that featured Cubase 4. They don't sell it anymore and haven't updated it. However, the Steinberg presenter specifically recommend recording in 32bit float. I won't pretend to understand why, but that was Steinberg's own video.

To make things less clear, Steinberg's own default empty project is 24bit.

[serga]

Originally Posted by Sk106

Cubase operates at 32 bit floating point internally, no matter what the project standard is set to. The project standard is more about what resolution the audio is stored in and played back in. I've found that it doesn't affect computer performance much, although Cubase would theoretically have to convert alot realtime, since it operates in 32 bit floating point internally.

You`re right. This is what happens exactly. Cub have to convert 24 bits to 32f
everytime you push the play button and it eats cpu not much but...

Originally Posted by DAW PLUS

Since 24 bit files are still processed in 32 bit float, it doesn't matter. The range remark you made is simply not true. 32 bit float = 24 bit + 8 bit mantissa.

No, sorry it's not *my* "range remark". I didn't invent the stuff. I'm just reporting what is documented and it matches with my experience.

You are talking about the data representation format and not the effect that the mantissa has upon the rest of the bytes. The mantissa isn't inconsequential as you are implying. You need to go back and check your facts. Wherever you are getting your information from, it's clearly not enough.

The dynamic range of 32-bit is much more than 24-bit or 32-bit. It's the difference between counting in whole number integers and counting int whole integer numbers AND counting tons of decimal places between them. That's why it's called "floating point". It's a decimal point that moves from value to value allowing a greater dynamic range. Go back and check your sources against other sources on the internet and in audio and computer science books.

That is why 32-bit float has an advantage over the other formats.

[DAW PLUS]

Originally Posted by astroidmist

No, sorry it's not *my* "range remark". I didn't invent the stuff. I'm just reporting what is documented and it matches with my experience.

You are talking about the data representation format and not the effect that the mantissa has upon the rest of the bytes. The mantissa isn't inconsequential as you are implying. You need to go back and check your facts. Wherever you are getting your information from, it's clearly not enough.

The dynamic range of 32-bit is much more than 24-bit or 32-bit. It's the difference between counting in whole number integers and counting int whole integer numbers AND counting tons of decimal places between them. That's why it's called "floating point". It's a decimal point that moves from value to value allowing a greater dynamic range. Go back and check your sources against other sources on the internet and in audio and computer science books.

That is why 32-bit float has an advantage over the other formats.

No need to explain this to me, what you describe is handled by the 8 bit mantissa. 32 bit float = 24 bit resolution with an 8 bit mantissa. This allows you to work with a huge dynamic range as long as you stay within the 32 bit float format, which is true for all Cubendo internal processing. Now, at some point you will have files, either as a playback/recording source which is treated with effects, levelling and summing, and whatever resolution you use, it will happen in 32 bit float.

Does that mean you have to use 32 bit float files to keep that internal resolution? No, because the mantissa is added or removed as soon as the processing starts. And I couldn't even measure a performance difference even with Cubase VST 5/32, which introduced this capability.

My source? Charlie Steinberg.
Another source, Bob Katz, claims you can even work with 16 bit files, as long as it is sufficient for the original recording to capture the full dynamic range of the instrument(s). And in that case, you have to think about your dithering process.

While theoretically extreme processing might reach 32 bit float limits, i.e. audible limits, as explained by Paul Frindle, I highly doubt anyone is ever going to be in that situation. And I am trying to break DAW's since 1998.

I don't want to push anybody into NOT using 32 bit float files. I've done it for years. But I mentioned a few reasons why 24 bit might be the better file resolution to use.

[Sk106]

Originally Posted by astroidmist

This is incorrect. Check your research.
Here's one site: Bores Signal Processing - Introduction to DSP - precision and dynamic range
It has the math to show the dynamic range of 32-bit floating point to be in the thousands of decibels. Othere websites corroborate this info. Look it up.

Good link.

Hm ... it says "The largest number which can be represented determines the dynamic range of the data format". Not entirely sure I agree. It may have a huge dynamic range, but if so only for a limited portion of the sound. 32 bit floating point with a 8 bit integer still only has about a 23 bit wordlength (plus quantization noise). The fact that the mantissa can move over > 1000 dB dynamic range, is not the same thing in practice as being able to represent that range all at the same time though.

But my understanding of it could be flawed. We're mostly bound by 24 bit converters anyway, but still it's an interesting mindcooker.

[samicide]

Originally Posted by adg21

Should I choose 32 Bit float instead of 24 bit in Cubase in 'project setup If I'm working mainly with software instruments and effects?

Yes.

[jupiter8]

It pretty much doesn't matter. 32 bit float has a slight edge in some rare cases but it puts an additional strain on the hd. Unless you do something incredibly stupid 24 bit is as good as it gets.

[Fredrik]

Yes what is affected here is the file format only, not the internal representation.

[gomfist]

just go with 24 bit - its better - smaller file sizes and so faster performance when it comes to file handling / streaming etc. just dont gain all your stuff into the red - right?

[John The Cut]

Originally Posted by DAW PLUS

No need to explain this to me, what you describe is handled by the 8 bit mantissa. 32 bit float = 24 bit resolution with an 8 bit mantissa. This allows you to work with a huge dynamic range as long as you stay within the 32 bit float format, which is true for all Cubendo internal processing. Now, at some point you will have files, either as a playback/recording source which is treated with effects, levelling and summing, and whatever resolution you use, it will happen in 32 bit float.

Does that mean you have to use 32 bit float files to keep that internal resolution? No, because the mantissa is added or removed as soon as the processing starts. And I couldn't even measure a performance difference even with Cubase VST 5/32, which introduced this capability.

My source? Charlie Steinberg.
Another source, Bob Katz, claims you can even work with 16 bit files, as long as it is sufficient for the original recording to capture the full dynamic range of the instrument(s). And in that case, you have to think about your dithering process.

This is correct.

Use 24bit.

Cubase runs at 32bit regardless.

Audio recorded will be at 24bit resolution. That is all you are setting with the project setup. All AD converters work at 24 bit.

[shanabit]

I think you guys just like saying " Mantissa"

Record at -18dB peak and you wont have to worry about clipping in Cubase. Save the squash for Mastering.