What's the use of 20 and 24 bit DACs with CDs?

The typical DAC chip of a typical mid-fi CD player does not actually out-put a 16 bit word length. Even the good DACs often only achieve 14 bit words at best-- this is with a 16 bit DAC chip. But when 20 to 24 bit chips are used, the DAC can more closely approach or actually achieve true 16 bit words, and thus produce more "information" or detail from a 16 bit encoded CD.

The "word" length in bits has nothing to do with the sampling rate. Even a typical inexpensive CD player uses 8X oversampling, which yields an actual sample rate of 352.8 KHZ. Over-sampling and up-sampling are essentially the same thing, but recently the term "up-sampling" has become a buzzword to help sell allegedly better CD players.

"Better" CD players are better (and more expensive) because of the use of the highest quality parts, ie DAC chip sets and other parts, and excellent engineering, design, and implementation. Levinson also uses 8X oversampling, but uses the best quality parts and implementation. I am not an expert in this but took a big interest in it (and did a lot of research) prior to purchase of the expensive Levinson 360S DAC. The 360S is an absolutely great DA Converter, but there are others. Cheers. Craig

Great response Craig! It's the same as a car. The speed limit is 65, so why get the Turbo V6, when the standard 4 cylinder will get you there.

A couple of points more here. When a 16 bit word is read from a cd, some form of processing, typically rate conversion, is needed to produce a 24 bit word from the 16 bit word. The rate conversion can be done inside the D/A chip, or it can be done in a processor preceding the D/A. The 24 bit word does not increase the information in the original signal, but just represents with higher precision the mathematical product of the 16 bit data word with the rate conversion filter coefficients. So Cjcerny is correct in assuming some form of processing to get from 16 bits to 24 bits.
Secondly, upsampling (typically done in a processor) can be significantly better than oversampling, 8x or otherwise, done inside the D/A chip because the quality of the upsampling filter can be much better than that used in most D/A chips - so there can be a significant difference in upsampling vs oversampling depending on where and how the rate conversion is done.
Thirdly, Cjcerny is correct in using the term 'bandwidth'. Bandwidth is frequently used in digital processing to mean total bits per unit time (e.g. the increased bandwidth of DVD), and not just to mean spectral bandwidth.