24 or 32 bit Recording...

I have a couple questions that are probably Jack related:

  1. Does Jack automatically try to open the audio interface as 24 or 32 bit audio, unless ‘force 16 bits’ is selected?
  2. Can I make Ardour record 24 bits instead of 32 bit files. I see no benefit in recording to 32 bit…it just takes up more hard drive space!

mm0

hmmmm…where did my sig go?

JACK attempts to configure the device for 24 bit audio by default unless told not to. In practice, 24 bit audio generally means moving 32 bit data back and forth between the computer and audio interface. I don’t know of any audio interfaces that provide 32 bit integer format, though recent RME devices have had the option of delivery 32 bit floating point. JACK does not attempt to ask for this format.

To change data-on-disk formats, go to the options menu and see “Audio File Format”. Its a matter for debate whether or not a particular format is ideal; Just keep in mind that there is a tradeoff between CPU cost (to convert data-on-disk into the 32 bit floating point format used inside ardour) and disk space. How this tradeoff resolves will be different for different people, which is why ardour offers you a choice.

Thanks Paul, My projects are generally small, so I like to use 24bit to disk. But typically even with larger projects (24+ tracks), I don’t usually have any problems with CPU…I just haven’t done any project in Ardour yet that are that big…
mm0

Just FYI (and for everyone’s), I did a project a few months back whith 12 different sessions, each containing at least 28 tracks, some of them with upwards of 50… all at 32-bit float, and many of the tracks with effects (and the rest bussed together, with fx on the busses). It pushed Ardour and my dual-athlon-mp machine (5+ years old) to what I thought would be the limit, but it chugged right along. Only occasionally would I get “disk couldn’t keep up” messages, usually around areas with heavy edits.

I’m of the opinion that there is no practical improvement in terms of fidelity when it comes to the difference between 24 and 32-bit-floating bitrates. The difference between 16 and 24-bit is significant, but most of us don’t know what a true 24-bit rendered signal sounds like, due to the power requirements of converters, wordclock sync/beating clocks, and a bunch of other factors I barely understand at this point. Most 24-bit signals end up playing back at around 20 bits at best. This is more than adequate, as only a percentage of that is usable dynamic range for most music.

Some time ago, I ran a few tests on several different systems. 24-bit/48K was the highest rate I could detect a noticeable improvement on…anything higher simply seemed to increase processing/memory requirements, and lower rates suffered in terms of fidelity…especially with quieter tracks. I couldn’t see much of a difference between dithering shapes, but could hear an absence of dithering during quiet tracks, etc. Of course, a bit of this is all system specific.

However, at an ill-educated guess, running plugin processing at 32-bit-float might be a good idea, as it might help maintain some fidelity. Maybe someone more knowledgeable could chime in with some references on the subject…

yes, I totally agree. The point where is does make a difference is internally…I haven’t looked, but hopefully ardour is internaly 32 bit float once the audio is being processed…this allows plugins to make calculations without clipping, and will definately help the signal path, even though it will be converted back to 24 bits on the way out.
mm0