I tried to fit the question into the title and description, but let me explain it more clearly.
I understand the basic principles of how the closer to 0db a recording is, the louder it plays, and the less headroom there is for peaks. I've got all my albums MP3gain treated to 89db, but of course there are some tracks with higher volume levels, I only apply album leveling, not track leveling.
I forget the dynamic range of a 16bit playback device, I think it's 100db, but I'm not positive. Now say I'm playing a file in winamp with peaks at -3db, so that gives me 3db of digital headroom before the file starts getting clipped by the playback process. I can tell just by listening that some tracks are getting digitally clipped by my use of winamp's equalizer. I've got some diy speakers that don't have the best bass, so I like to boost the lower ranges by about 6db or so, and during the beats I hear it clip unless I turn the preamp setting way down.
I've got an nforce2 sound chip, and I know I should upgrade it just for the improves S/N ratio and those types of things, but would having a 24bit soundcard avoid the clipping I'm getting in winamp from the eq boosting sounds above the limit of 16bits? I would get an extra 8 decibels of headroom, so pretty much no matter how much I bass boost the music, it wouldn't be boosted above the 24db range. If it does work like this, are there any special plugins to fully exploit the 24db range while playing back processed 16bit audio? I know of MAD, but anything else?
Oh, I want to get a maya 5.1 mkii, but I can't find any for sale, nor can I get any good reviews of the mkii version.

Thanks,
Keith D

Hi
your music will clip on 24bit hardware too, because 16 bit signals (audio cd etc) will be mapped to 24 bit by adding 8 bits to the lower end of each sample value. like this:
1011 0110 1110 1001 -> 1011 0110 1110 1001 0000 0000
You will have to reduce the global gain of your equalizer to avoid clipping

I think you're confusing bits with decibels.

1 bit resolution = approx. 6dB dynamic range

16 bit = 96dB dynamic range
24 bit = 144dB dynamic range

The problem is that the range is added at the bottom, i.e. what clips at 16 bit also clips at 24 bit.

The only solution would be to use 32 bits float but I don't know converter really capable of this.
If you reduce the level, 24 bits might help you because you keep 8 bits of accuracy below 16 bits.
In the Sunhillow example, if you decrease the level of 6db 1011 0110 1110 1001 0000 0000 becomes 0101 1011 0111 0100 1000 0000. With 16 bits it becomes 0101 1011 0111 0100, so you have a quantisation error. Dithering can partially solve this problem but at a price (added noise). This is theoretical, in the real world, this is very subtle.

I’d say, don’t use too much equalisation, use level reduction with 24 bit accuracy (or with a dithering in 16 bits).

In case of 24 bits, you have to be sure that the path of the sound thru the player, the mixer and the soundcard is 24 bit all the way.

24 bit over 16 bit just means a lower noise floor, not a higher gain before clipping. What you can do is turn down the volume, up to 48 dB, and still have a lower noise floor than 16 bit.

Thanks everyone, I get it now. It doesn't seem so important anymore for me to get a 24bit soundcard, I'm probably just going to wait until I get an nforce3 mobo with 24bit audio built in.

I recommend against motherboard sound cards. There is usually much more electrical noise and lower quality of the DACs.

A link to RMAA results would be interesting.