Is it worth the extra load on the CPU to use 64 Bit mixing in AP?

I was just wondering what the consensus was on using the 64 Bit mixer option in Auria Pro, in your opinion is it worth the extra load it puts on the CPU?

Comments

  • I am interested, too, re 16 v 24 in Cubasis. Maybe an explanation as to what bit rate means, please.
    Also, does it make a difference in laying down tracks as well as mixing?

  • @LinearLineman said:
    I am interested, too, re 16 v 24 in Cubasis. Maybe an explanation as to what bit rate means, please.
    Also, does it make a difference in laying down tracks as well as mixing?

    My current understanding of what the bit rate is, and I absolutely could be mistaken, is that it’s actually bit depth, which is the amount of detail that is recorded, again, I could be totally wrong, but that’s how I understand it. In Reaper on pc, I record with a bit depth (as it is referred to in settings) of 64, which may or may not be a total waste of space, but I just want the highest level of quality I can get.

  • @LinearLineman said:
    I am interested, too, re 16 v 24 in Cubasis. Maybe an explanation as to what bit rate means, please.
    Also, does it make a difference in laying down tracks as well as mixing?

    16, 24, 32 and 64 in these cases refers to bit depth not bit rate.

    You will find plenty of good articles about this d on the web.

    Short version: uncompressed digital audio (technology pcm encoded audio: https://en.m.wikipedia.org/wiki/Pulse-code_modulation) represents audio as a series of numbers. The bit depth refers to the number of bits used for those numbers. 16 bit numbers use 16 bits. 24-bit numbers use 24-bits, etc.

    Theoretically, the more bits you have, the better the fidelity in certain respects. CD audio is 16-bit.

    Higher bit-depth buys some extra headroom.

    Most people can't tell the difference between 16 and 24 bit audio even on good equipment. Obviously, some can. Mixing and processing at higher bit depths does improve headroom and reduces the likelihood of a number of kinds of distortion that can happen when processing signals.

    So, even when mixing down to 16-bit audio and using 16-bit sources, it makes sense for apps to do their processing at higher bit depths--which comes at a price since 32 bit numbers use twice the memory as 16-bit numbers. Btw, because of the way computer memory works, 24-bit numbers use the same amount of ram as 32-bit numbers (though when stored as files that isn't true)

    Bit rate refers to something different which you should look up. It refers to the number of bits per second needed to stream a file. And is consequential for decisions about encoding (compressing) and streaming your files.

    As to whether the added resources of 64-bit mixing is worth it: probably not for most people. You would probably already know if you are someone for whom it would be consequential.

  • For the original question: IMO no, the 32bit engine is absolutely adequate for most mixing needs.

    @LinearLineman said:
    I am interested, too, re 16 v 24 in Cubasis. Maybe an explanation as to what bit rate means, please.
    Also, does it make a difference in laying down tracks as well as mixing?

    For recording acoustic instruments a bit depth of 24 is considered to be beneficial because it lowers the noise floor substantially, meaning there will be less noise in the recording and far more headroom for making adjustments later.

    When recording digital sources such as soft synths it's not really an issue since there will be no noise to begin with.

  • Thank you very much for your replies, and cleaning up this issue, I appreciate it, as I’m very much still learning.

  • @espiegel123, indeed, I don't know, therefore I am not! But now I do. If it keeps noise down (even tho there is none and more headroom (who doesn't need more?) then I will keep touching 24 bit depth for mixdown. Perhaps post your cogent remarks to the wiki?

    Thanks @richardyot, always value your talent and knowledge.

  • I believe at 16 bits you get clipping if you exceed 0db. At 24 or 32 (as long as the app handles it properly), you should not get clipping.

  • @richardyot said:
    When recording digital sources such as soft synths it's not really an issue since there will be no noise to begin with.

    That's not quite accurate as you can still get digital (quantize) noise when you do processing, and it also gives you more headroom. But it's certainly less of an issue.

  • @Ripper7620 said:
    I was just wondering what the consensus was on using the 64 Bit mixer option in Auria Pro, in your opinion is it worth the extra load it puts on the CPU?

    No.

  • @wim said:
    I believe at 16 bits you get clipping if you exceed 0db. At 24 or 32 (as long as the app handles it properly), you should not get clipping.

    I should add that this is really only important for mixdown and for recording. Internally all Hosts I know of are using at least 32 bit, making internal clipping a non issue.

    Again, purely from my understanding, and I’m no expert.

  • edited May 13

    @cian said:

    @Ripper7620 said:
    I was just wondering what the consensus was on using the 64 Bit mixer option in Auria Pro, in your opinion is it worth the extra load it puts on the CPU?

    No.

    A concise, honest, straightforward and scientific answer, love that :)

    32-bit floating point is so ridiculously accurate already that I doubt there's anything besides inertial navigation for a spacecraft that would require more accuracy. (OK, maybe that's pushing it... but so far, I've used 32 bit floats (i.e., float, not double) in absolutely everything, everywhere, and never encountered an issue).

    EDIT: Note that "16-bit" or "24-bit" audio files use INTEGER representation, while any internal mixing uses FLOATING POINT, which in itself is a better format for representing values that can span a large range.

    A good explanation of the accuracy of floating point is here:

    https://www.google.com/amp/s/blog.demofox.org/2017/11/21/floating-point-precision/amp/

  • @SevenSystems said:

    Of course if you deliberately record your stuff at -120 dB you might run into quantization noise, but yeah...

    And even then you'd need to do destructive 'normalise' to be able to hear it ;)
    Too many apps have somewhat 'limited (~20-30db max) gain' forcing one to normalise low-level recordings.

    It would be nice if a DAWs such as Cubasis could do 'non destructive normalise' (Ie. analyse the recording and set the clip-gain(which is also missing) to make it a 'normalised' recording).

    So in order to 'normalise' a super low-level '1-3bit recording' ~96db of gain would be needed...

    The advantage of 24-bit recording is definitely audible if the recording level is less than optimal.
    (ie. record at -55db using both 16 and 24 and then normalise the files).

    I try to stick with 'if it sounds good' there's no need to change anything...

    From what I've heard the 64-bit mixing is mostly needed when channel count increases...(>64 channels or so?).

  • @Samu well yes, every floating point operation will in most cases slightly increase noise, as not every real number can be represented exactly. So yes, the more additions (i.e. mixing), the more noise is introduced. If 64 bit mixing is used, the amount of noise introduced by each added channel is 2**28 (268435456) times less than with 32 bit mixing. But it's already very small with 32 bit to begin with :)

  • 32 bit floating point is already phenomenal for audio computing. The floating point means very quiet sounds are computed with the same accuracy as loud sounds. The only thing the floating point doesn't do is compute super tiny sounds overlaid with a bone crushingly loud sound. So the sound of a mouse rustling next to the sound of a freight train crashing off the tracks may incur some amount of quantization noise.

    Where the goal is human ears enjoying sound, I think the resolution is more than there. For figuring out "why do my recordings sound like ass?" I would look elsewhere...

  • @Processaurus said:
    So the sound of a mouse rustling next to the sound of a freight train crashing off the tracks may incur some amount of quantization noise.

    Best way ever to put it :D

    Yeah it's a bit hard to grasp what floating point means. Essentially, it has always the same "relative accuracy" compared to the magnitude of the number it represents. So it is less accurate for very large numbers, and more accurate for small ones. The problems arise when both "classes" of numbers are mixed.

  • Quantization noise can be an issue at 16 bit. But you'd really have to go out of you way for it to be a problem at 24 bit.

    A lot of plugins/DAWs will use 32 bit for internal processing, as the increased accuracy is probably worth the (slight) increase in CPU load for some operations.

    64 bit is going to significantly hit your CPU though for a bunch of boring technical reasons, and it's highly unlikely to have any effect that you could ever hear.

  • @SevenSystems said:
    Yeah it's a bit hard to grasp what floating point means. Essentially, it has always the same "relative accuracy" compared to the magnitude of the number it represents. So it is less accurate for very large numbers, and more accurate for small ones. The problems arise when both "classes" of numbers are mixed.

    For audio it really only becomes a problem when you get close to 0. Particularly if you do something like make a sound really quite, do some processing and then knock the gain up. With 16 bit this can happen if you're not careful, but with 24 bit you'd really have to go out of your way.

    Also for really small values you get denormal values (numbers that are larger than 0, but to small to represent using floating point) which can do a number on CPU performance if the programmer was not careful. But that's pretty far into the weeds (and incredibly tedious to deal with).

Sign In or Register to comment.