Bits and Pieces

Let's talk about bits, and more specific, 16bits. This blog was on my to-do list for quite some time, but I finally started and, more importantly, finished it! 🥳

Even though it really is 2025, I see people reverting back to the year 1995 where 16bit was pretty much the only way to go and so, the standard. I get it. I’ve seen some articles pop-up recently about 16bit being more than enough and 24bit, or 32bit is just a waste of diskspace and no one will hear the difference. Depending on the situation this is indeed (partly) true, let's find out why.

As a final medium for CD for example, 16bit it’s perfectly fine, but that is because CD works at 16bit. For listening to lossless audio, 16bit is also perfectly fine and you will not hear any difference between a 16bit and 24bit master. But wait, why do I want to write a blog about this then anyway? Maybe you noticed, I was talking about the ‘final medium’, ‘the master’, being perfectly fine in 16bit. That is where things go wrong and a lot of people assume that 16bit is always fine. OK, let’s dive into this in a simple way without getting too technical why you really need to stop using 16bit.

Assumption is the mother of all fuckups

Recently I had someone send me a mix in 16bit because all of his samples where in 16bit so he assumed it didn’t matter at all if the mix was also rendered in 16bit. He was told (by YouTube. Duhh..) that it will remain 16bit anyway. WRONG! All DAW’s (and plugins) work internally at 32bit floating point (some even at 64bit floating point). The moment you play back one 16bit sample or audio file without doing anything like levels, EQ or whatever, the audio remains 16bit indeed. But as soon as you lower the gain by let's say, 0.1dB or add 0.28dB of high shelve EQ or whatever, the magic happens! Now all of a sudden, we are working at 32bit floating point and the outcome is not 16bit anymore, it’s really 32bit!

Let me briefly explain how bitrates work without getting to technical and yes, I will miss some details here and there but I will do that on purpose because otherwise it might get too technical and that’s what I really want to avoid here. I think that a lot of misinformation comes from misinterpreted, complex technical information. Simplicity is key!

The simple rule for bits

There is a simple rule for digital audio which is easy to remember, every bit is 6dB of gain. Yes fellow nerds, I know this is not true, it’s 6.02dB and you can not call it gain/headroom or whatever but I will do this anyway! It’s my blog right? I want to keep it simple so please be quiet. No offence.😃 Where was I? OK, once again: every bit is 6dB of gain (or headroom).
16bits x 6dB = 96dB, so a 16bit audio file has 96dB of headroom between 0dB and -96dB. Plenty of headroom right? More then the human hearing can perceive! Hmm…  right.. let’s get back to that later.
24bits x 6dB is = 144dB, that’s a 'bit' (pun intended) more right? What about 32bit floating point then? 32bit floating point has 1500db of headroom. Wait, what? That’s a mistake, right? Hmmm.. yes, you are correct, it’s actually 1528dB. I will not get into detail, but the name floating point is where the magic happens. Simply said, it ‘shifts the bits to where they are needed’, even above 0dB (up to about +770dB if you really want to know).
This 32bit floating-point system is used in every single plugin and DAW (except for the ones at 64bit). As said earlier, as soon as you ‘change the audio’, being it gain, eq, whatever your 16bit audiofile becomes in fact 32bit audio. I think you can now also imagine that if your render that precious 32bit mix to a 16bit file, you are degrading the audio. And now it get’s even more interesting because that degrade to 16bit is actually doing more harm than you might expect. Let me explain.

Truncation, now the fun really begins

I think you can imagine that what you are basically doing is chopping of the lower bits when going from 32bit to 16bit. You will lose a lot of precious information. Let’s assume your mix is peaking at -6dB, this means that at 16bit you will only have 90dB of headroom left of those 96dB you have available. But that’s still plenty of headroom right? Hmm.. yes, in the ideal world it would be but again it’s a bit more complicated than that. When you simply chop those bits off, you literally truncate those bits, you throw them away. This is where things go wrong because that information which is now gone is transformed back into distortion because of truncation/quantization errors. Sounds like fun right? OK, but you can use dither! Oeh.. Dither, now the fun begins! Where’s that can of worms? Dither will indeed mask the truncation/quantization errors (it's job) by adding a layer of noise. But wait, did you just say noise? Yes, I said noise because that’s what dither basically is. Nerds will tell you it’s way more complicated and it is, but I said it before, simplicity is key and I will stick to that and if you really want to know more about dither, google it. It will soon get noisy, watch out. 😅Good luck! Dither.. the thing is, you wan't to use dither only once, for the final medium and we are not talking about the final medium here, we are talking about 16bit rendered mixes which are going to be mastered, in case you forgot.

Some prove is in the image

By now you might be convinced already that using 16bit, even with dither, might not be ideal when you make an export and send that to the mastering-engineer. If you still have doubts, let me show some images I made zooming into a reverbtail and fade-in showing the difference between 24bit, 16bit truncated and 16bit dithered.


This is an image showing a reverbtail fade-out zoomed in. You can cleary see that the truncated 16bit version misses the 'chopped off bits' which shows the truncations/quantization errors. The dithered version in the bottom also shows the dither noise profile. (I used a standard flat dither in this case) I think it's easy to imagine what the truncation sounds like right? 

This image shows the intro/start of a track which is slightly louder then the fadeout from above. Notice the truncation errors and missing details on the left side of the image? Dither masks this a bit but still it's not the best result and 24bit is clearly better.

What to do now?

I hope I convinced you by now to stop using 16bit for your final mix renders and start using 32bits or 24bits. Of course, when rendering to 24bits you are in fact also downgrading the mix, but in practice there will not be any noticeable difference. If you want to do things by the book, you should use 24bit dither when rendering but even that is questionable.

One big advantage of using 32bits floating point is that if you happen to have some peaks shooting over 0dB, that will be fine since floating-point will capture those and internally all information will remain there without digital clipping. Gain staging always remains important though, but that’s of course a totally different topic.

Yes but the files are a lot bigger in 24 or 32bit! You serious? 🤨Maybe it's time to ditch that 1gb harddrive and do some upgrades first.

What else? Be careful with your settings and always double check before rendering. Quite recently there was a Logic update and after that update the render settings went ‘back to default’ for a lot of people, which turned out to be 16bit truncated! Also make sure that you render to PCM and not some lossy format. And dither? That should just be used once, for the final medium and only when you lower the bitrate to let's say, 16bit. But we just agreed that we stopped using 16bit didn't we? So forget about dither please.

It's 2025, 16bit move over and go back to 1995, that’s where you belong!

This Blog was written on a laptop, at the diner table, by a real person (me) and no AI was involved. OK, this is not fully true, I used chatGPT because I wasn't sure if the headroom of 32bit floating point was 1524dB or 1528dB.