Is there really that much difference between 20 and 24 bits?

[cassembler]

many if not most 24 bit converters are not true 24 bits at all. they may be 21 or 22 and simply fill the remaining with a null value. when designers are trying to create an electronic device with 16,777,216 different voltage step values, they are basically at the physical limits of what electronics can do. therefore, all those bits don't always make it. don't sweat it.

[might be full of shit, but I don't think so] Actually, I think the highest measured bit depth of ANY converter (at room temperature)is limited to about 22 bits or so. Notice how converters stopped at 24 bits. Anyone ever seen a 32 bit converter? But they're sure pushing those enormous sample rates... [/might be full of shit, but I don't think so]

Bit depth is only one of many factors that make a converter good. IIRC, a great way to "hear" how good (or bad) a converter is is by taking a sine wave, at a frequency NOT easily divisable by the sample rate (11,025 Hz is a bad choice) and fade it in and out from -infinity to about -60dBfs or so. You'll have to be really careful and crank the mains, but believe me, you'll hear quite quickly how your converters really work.

Try 887 Hz, I think that one was recommended.

[chikkenguy]

if youre recording at 16 or 24 bits with the same unit and listening to those recorded files, you probably wont hear a difference. the difference is mainly during the processing stages of your mixdowns, bounces, whatever. go to www.digido.com and look under articles and read the one about dither.

[mrc]

The real difference I've noticed is in the headroom, during tracking. With 16 bit, it was really important to push the levels right up close to 0db. I added an Adat to the cassette/8track/16track step ladder, and was shocked by how much was lost, even 3 or 4 db off the 0db limit. Jumping to 24 bit from 16, I was able to stop using outboard compression, except as an effect. It's no longer necessary to push the levels right to the brink, to get fine resolution. Tracking vocals, very fast peak limiting (less than a .05 ms attack, -10 or so threshold and 500 or less ms release) and it can be helpful with drums as well. Peaks that go by so fast they don't meter can be difficult to control with regular compression, and can cause overs in the mix that are real problems. I've found that using the limiting and backing off 0db by 1 or 2 db's @ 24 bit, I don't really miss using tape anymore. The problem is how cheap all the tape gear is on ebay, now Some of the newer versions of Daws allow using a software limiter during tracking, so a hardware limiter compressor may be an antique as well. A lot of the all in one channels have compressor/limiters that are way too slow for this purpose.
mrc

[jayrope]

i second the "yeah: 4 bits (more)" option. all the rest means using your ears. use pow-r dithering to get to 16bit in the end in any case. don#t just cut the overflowing 8bit off.