Gain distribution is an interesting subject. Behind it there's a
fundamental tradeoff between
noise on the bottom and
nonlinearity (up to clipping) on the top end of the level scale. Power consumption also plays in.
What makes the subject so tricky in practice is the issue of
flexibility and
practicality. Your average 45 dB of voltage gain found in typical integrated amps came about because people wanted to interface phono preamps (2.5..5 mV input, 40 dB gain @ 1 kHz) with the kind of output levels needed for, say, 80..100 wpc, with 6..10 dB of gain to spare. Your average MP3 player is in the same ballpark or a bit louder.
If you have all of that gain behind the (only) volume pot, the input will be extremely insensitive to overload, which makes the amp very foolproof.
At the same time, however, the volume pot defines your input impedance and as such must be sufficiently high in value.
That in turn increases source impedance and thus voltage noise, or for an equivalent view, available signal power for a given setting drops.
Now SNR always is the ratio of signal power to noise power, so whichever way you see it, it is obvious that you may get a noise problem. (Not to mention related input nonlinearity.) Many amps built as outlined tend to be quiet enough for medium-sensitivity speakers, but definitely won't be dead silent once we get into 95..100 dB SPL / 2.83 V / 1 m territory. The problem is even more readily apparent with the large range of sensitivity found in headphones.
In this situation, adding an input buffer may already benefit output noise levels considerably. A buffer does, in fact, not have "no gain". In fact, it's got plenty -
current gain, that is. You can easily drive a 10k pot with it (or a value even lower, limited only by nonlinearity or allowable power dissipation) if you had a 50k one before. If the following amp has low voltage noise, that'll reduce noise levels by a ratio of up to sqrt(5) (7 dB). (You can't drop voltage noise levels indefinitely though, it gets very hard once you get into 1-2 nV/sqrt(Hz) territory.)
Of course you have already given up some input overload capacity at this point (supply voltages are obviously finite), though it wouldn't be an issue in practice. As you shift gain from after to before the volume pot, output noise levels drop fairly proportionally, but input overload becomes more and more critical, as well as general nonlinearity. Given usual input circuitry supplies and CD player output levels, roughly 10 dB is about it.
That's why top-notch preamps and integrated amps have tended to use a 4-gang pot, i.e. two coupled pots with the second one at the very front. That potentially aggravates channel imbalance, but can yield very good results in terms of noise. It doesn't have to be a second pot, that merely was found the most convenient to use.
What you may end up with could be: Input buffer - pot 1 / switched attenuator - +16 dB gain - pot 2 - +29 dB gain. Which in fact isn't terribly dissimilar to the setup with CD player headphone out connected to power amp input as suggested above.
All this analysis is based on looking at signal levels, noise levels and limits imposed by nonlinearity for each stage. Voltage noise densities are determined, integrated (20 kHz = factor of 141) and compared to signal levels (ratio expressed in dB) after multiplying/dividing by gain if necessary. At least that's what I do.
As it happens,
I wrote an article on the subject of noise in amplifiers myself last year, not being aware of this one. A bit of redundancy certainly can't hurt.