[1] I've got one 32-bit device already and looking to get a soundcard that happens to be 32-bit as well, just wondering if the 32-bit part is mostly marketing gimmicks or the future in SQ.
[2] Also wondering what happens to the sound when your DAP is 32-bit and you connect it to a 24-bit AMP/DAC?
[3] "Of course, there's a healthy debate in audio circles as to the merits of 32-bit. [3a] One side cynically says it's a waste of money and space, and well beyond what humans can hear. [3b] The other side argues what are ye, deaf?
[4] There is a clear difference, especially when lower-resolution audio is upsampled and resampled. CreativeLabs , in fact, says the 32-bit DAC gives the AE-5 "extra headroom" for upsampling and resampling multi-channel audio sources in gaming."
1. It's not "mostly" a marketing gimmick, it's ENTIRELY a marketing gimmick! (see 3 below).
2. No one knows in practice, because there are no 24bit AMPs/DACs. As mentioned by others, about 20bit is the maximum. In reality, the data beyond 20bits in the audio file you are reproducing is almost certainly nothing but noise and even if there were a tiny bit of useful data above the noise, it would be below the noise floor of the DAC itself. All of this is irrelevant though because even 18bits of useful data is almost certainly below the noise floor of your HPs or speakers anyway.
3. This is a lie! There might be an unhealthy debate in audiophile circles but there is NO debate in audio circles, particularly in pro audio circles. Let's be absolutely clear here, 32bit is 24bit!! When you hear about 32bit, they're talking about 32bit float. A 32bit float word essentially comprises of 8bits reserved for the exponent and the remaining 24bits is the mantissa, where the actual audio data is stored. A 24bit fixed word does not have any bits reserved for an exponent and it's all effectively audio data. Therefore, both the 32bit float format and the 24bit fixed format contain 24bits of audio data (although again, at least the last 4+bits are just going to be noise anyway)! The ONLY potential merit of 32bit float is that if an audio clip occurs (0dBFS is exceeded) it can be recovered with 32bit float, whereas it cannot be with 24bit fixed.
3a. It's got nothing to do with what humans can hear. 32bit float and 24bit fixed are in effect exactly the same thing and we can't reproduce anything beyond about 18bits anyway.
3b. As 32bit float and 24bit fixed are effectively the exact same thing, there is no rational alternative but to argue "what are ye, completely gullible and deluded?"!!
4. Marketing BS which takes advantage of common misunderstanding. (see next response).
But for advanced digital processing (inside the machine) 32 bit can be a good thing.
We have to be very careful here! We have to be careful to separate the processing environment from the audio file format, they are entirely separate and different things. This fact appears to be deliberately overlooked/confused by audiophile marketing. Let's use a 16bit example to look at what the marketing is implying: At it's maximum, 16bit gives us about 94dB of dynamic range. If we apply some processing to this 16bit file we are going to get some error, which in effect is going to be noise down at the -94dB level, which is no problem. However, if we apply lots of different processes, say 10 or so, we'll have far more error/noise, which will probably be audible. With 24bit (or 32bit) the error/noise starts at about -140dB and even the accumulated noise (error) from say 10 processors will still end up being well below the noise floor of any DAC and therefore be inaudible. So, 24 or 32bit would be obviously better in the case of several successive digital processing steps. ... Err "NO", this is NOT how it works in practice, this is not how it has ever worked!
In practice, we have what is called a "mix environment". By the mid/late 1990's all commercial DAWs, audio editors and digital mixers had an internal mix environment of either 32bit float or 48bit fixed. Our 16bit file is loaded (or "played") into this mix environment and all processing occurs within this mix environment. The level of error/noise is therefore defined by the least significant bit of the mix environment, NOT the file format and in the case of say 48bit fixed, that's somewhere down around the -280dB level! So even the accumulated error/noise from dozens of successive processing steps remains way below audibility. In effect then, regardless of how many processing steps we have in the mix environment, what we end up with is just one processing step at 16bit, the step of coming out of the mix environment and into the 16bit file format ("bouncing the mix down"). Furthermore, it's standard practice to achieve this final step with noise shaped dither, which results in that error/noise being down at around -120dB.
BTW, from the late 1980's to the mid 1990's digital mix environments were 20bit, which was still sufficient in some cases but almost all commercial mixing was done in the analogue domain during this period (and earlier). Today, commercial mix environment are virtually all 64bit float, which theoretically allows thousands of processing steps before the error/noise reaches the noise floor of even the best DACs. Obviously that's a bit ridiculous but there are some workflows which might require hundreds of processing steps (film/TV rather than music production) and therefore a 64bit mix environment *might*, in certain very rare circumstances, have some benefit.
@Vilhelm taking all the above into account, there might be some advantage in the SoundBlaster card creating a 32bit float mix environment for performing processing (they mentioned resampling for example) but there is none whatsoever in the DAC chip itself being 32bit and no benefit in a 32bit file format.
G