Two things about ripple

[LongRunner]

Note 1: The ferrite coils used in output filters have to have capacitors on both sides (electrically speaking) of them to reduce the ripple voltage substantially. If you only have one cap after the ferrite coil, you'll gain little compared to not having that coil to begin with (EDIT: This may not be true of all PSUs for a very strange reason - scroll down to post #5 for the explanation), as you'll just end up with the ripple current going through the coil and capacitor after it, only reduced a bit (as what else is there to complete the loop the ripple current takes through the circuit???). (The ferrite coils are much smaller than the mag-amps in inductance as well as physical size.) Having a complete filter indeed keeps the output ripple voltage much lower, but that's because the cap before the coil completes the ripple current loop, so the coil and the capacitor after it only draw a small ripple current, giving a very low final ripple voltage with good caps.
(I haven't actually sourced this statement, but it would seem to me that the unimpressive +12V ripple result on that Rexpower unit is indicative. You can see in the photos that +12V there has a decent coil, but only the one cap. My recommended solution would be to use higher grade replacement caps in filtering stages with only one cap and/or no ferrite coil.)

Note 2: On 100/120Hz ripple, from the primary side - having multiple outputs on the same transformer means there will be some level of it on all but one of those. (It can't reach the outputs of the buck converters used for +3.3V and +5V in modern units, nor those of the linear regulators sometimes used for the negative rails.) It just needs to be kept at a safe level, which is the sole responsibility of the PSU engineers. The secondary-side components won't have any appreciable effect on that frequency, unless you install some ludicrously huge (in µF) caps.

[Wester547]

I noticed in Delta/Newton power supplies that they almost always have no ferrite coil for the +12V rail, just two or three low ESR capacitors in parallel. Judging by the reviews, it doesn't seem to hurt ripple suppression much (except for the EA-500D, but that power supply has a forest of wires almost completely covering the secondary side so the ferrite coils and capacitors basically get no airflow, and the +5V rail only has one capacitor and one ferrite coil...), probably because of the very nice toroids they always use for the +5V/+12V rails. Are you also saying that 4700uF capacitors and higher on the output actually have an effect on 100/120Hz ripple... or does it have to be much higher than that? And are you saying that ferrite coils actually do handle ripple current the way electrolytic and polymer capacitors do if used properly in the type of output filter (as well as being something of a "voltage divider" for the capacitors before and after), so that the capacitors after the ferrite coil have less ripple to handle the better the ferrite coil is?

[LongRunner]

Are you also saying that 4700uF capacitors and higher on the output actually have an effect on 100/120Hz ripple... or does it have to be much higher than that?

4700µF would only be starting to suppress 100/120Hz on the -12V (and in old half-bridge PSUs, -5V if applicable) rail. To filter out 100/120Hz ripple on the main rails, you'd need bigger caps than you could fit into the PSU. The only practical change you can make to an existing unit that will help with it is, of course, installing larger primary capacitors.

And are you saying that ferrite coils actually do handle ripple current the way electrolytic and polymer capacitors do if used properly in the type of output filter (as well as being something of a "voltage divider" for the capacitors before and after), so that the capacitors after the ferrite coil have less ripple to handle the better the ferrite coil is?

That's a bit of a confusing question, but as used properly, the coil and the capacitor after it indeed form a "divider" for the ripple voltage going in (and on the capacitor before it). For the math required, see the Wikipedia articles. But even a worst-case scenario example, with a 0.5µH coil, a 2200µF 6.3V 10mm Chemi-con SXE (an extremely old and obsolete series, which is only barely "low ESR") capacitor after it, and a 30kHz (!!!) switching frequency (higher is better here as it allows a smaller coil to provide the same filtering that a larger one would at a lower frequency) reduces the ingoing ripple voltage by 1.41:1 - and improving the parameters helps a lot.

(EDIT: I used the wrong formula at the time. The formula I used was only good for dividers composed solely out of one property — resistances, capacitances, or inductances. As we have a resistance/inductance combination here...)

[Wester547]

Wouldn't a degree of 100/120Hz ripple pass through the +5VSB output as well when the computer is in standby (I guess not much can be done about that?)? And was your example without a capacitor before the ferrite coil, since the ripple current would go straight through the coil and onto the capacitor (my guess is that a filter that has a capacitor before the ferrite coil and none after wouldn't be much better?)? And I was asking if ferrite coils remove ripple current or not so the capacitors after the ferrite coil have to handle less.

[LongRunner]

Wouldn't a degree of 100/120Hz ripple pass through the +5VSB output as well when the computer is in standby (I guess not much can be done about that?)?

No, as the control loop cancels it out. It can't, however, provide perfect cancellation of 100/120Hz ripple on multiple outputs from one transformer - only the one the PSU engineer chooses to track. This limitation can be sidestepped by, in a modern PSU, making the +12V as the only output from the main transformer itself, using buck converters for +3.3V and +5V, and using either a linear regulator (7912) or a buck-boost converter (as I suggested for the dream PSU; that converter topology is commonly used on HDDs to provide -5V for the head preamplifier. It can output any voltage you want as long as its polarity is the opposite of the input's) for -12V.

And was your example without a capacitor before the ferrite coil, since the ripple current would go straight through the coil and onto the capacitor (my guess is that a filter that has a capacitor before the ferrite coil and none after wouldn't be much better?)?

I wasn't referring to a filter with only the one capacitor. I was just referring to the ripple voltage filtering ratio provided by the coil and the capacitor after it - there is a cap before the coil, I just wasn't referencing its specs. What makes you think I was referring to a single-cap filter, when I already stated that the coil-and-single-cap arrangement provides little benefit???

If you don't have a capacitor straight after the mag-amp, the ripple voltage at that point is "undefined", so my statement wouldn't have been valid in that context.

And I was asking if ferrite coils remove ripple current or not so the capacitors after the ferrite coil have to handle less.

In a complete filter, the cap after the coil takes little ripple current (as long as the cap before the coil remains intact), but not if you only have the one capacitor after the coil.

EDIT: Now that I take a better look at the scope shots on this site, I see that some PSUs without the ferrite coils have sharp "spikes" at 100Hz as part of the ripple waveform, which the coils in conjunction with the capacitors after them seemingly help to damp. It's confusing, but less so when you consider that the waveform is nothing like a sine wave (which is what reactance is based on). I'm quite sure the design of the PSU is the root cause, though.

EDIT 2: Okay, this unit does have (slight) 100Hz ripple on +5VSB, but it seems to be an outlier.

So it seems that the only place adding a ferrite coil before the first capacitor is of use is in group regulation (the half-bridge and forward setups where a single large mag-amp is shared between +5V, +12V, and the negative rails).

[Wester547]

Do ferrite coils on the output help with the "escaping" 100/120Hz ripple, then? And adding a ferrite coil before a capacitor means better crossloading, you mean?

[LongRunner]

Do ferrite coils on the output help with the "escaping" 100/120Hz ripple, then?

Only with the sharp spikes that appear in the ripple waveform of some PSUs, at that frequency. I don't know the exact way in which said spikes are produced, though.

The Hairong in the 2011 el-cheapo round-up is a pretty good example of this phenomenon. You can see that, if those spikes didn't appear, ripple suppression on +5V and +12V would actually be decent, possibly even "good". (I noticed also that none of the main rails there are clear of 100Hz.)

It remains that these coils have no effect on the fundamental frequency of 100/120Hz. The spikes, on the other hand, are very short, and easily absorbed by the coils used.

[Wester547]

This might sound like a rudimentary question but the primary capacitors are there to smooth the 100/120Hz ripple, right? Do they help the "escaping" 100/120Hz ripple and those spikes at all? Since they seem awfully small capacity wise to help such a thing, though it was to my knowledge that they are there to smooth that low frequency ripple and provide enough current to feed to the main switcher(s) to chop and deliver to the transformer, but I take it the higher the capacitance in the voltage doubler, the lesser the 100/120Hz ripple?

[LongRunner]

This might sound like a rudimentary question but the primary capacitors are there to smooth the 100/120Hz ripple, right?

They do smooth it, but only to a level that the control loop can work with. If the control loop couldn't keep up, you would still get an unacceptable amount of it on the output.

Larger primary capacitors should help.

[Wester547]

I figured the following topic of discusssion would be best to bring up in this thread, but does an increase in 100/120Hz ripple mean a noticeable increase in the ripple voltages on each rail (considering the fact that the ripple frequency on the secondary side is much higher, often as high as 100KHz or more)? I believe LongRunner said that even with the higher grade polymer capacitors used in the Hipro polymod test that the ripple voltage didn't change much because the bulk storage and control loop (both of which handle the 100/120Hz ripple) and the main magamp circuits (which handle a majority of the ripple current on the secondary side), as well as the type of output filters, remained the same (despite the great decrease in capacitance by way of using polymers, but the huge increase in ripple rating and decrease in ESR rating).