Motherboard Terminology Question

[BobRoss]

I am reading a review of the ASUS Rampage Extreme motherboard and they mention that there is a 16 phase supply for the cpu, a 2 phase supply for the dimms and a three phase supply for the north bridge.

Can someone clarify what that means? I understand what the term three phase means in terms of AC power but i am pretty sure that it does not apply to DC power in the same way. So I am lost.

[Gizmo]

Actually, believe it or not, it's pretty much the same concept, but applied to PWM controllers instead of AC.

Basically, each of your PWM 'phases' runs out of phase with the others (but at the same frequency), in order to provide better regulation. I've got to suspect that the Asus mobo claiming 16-phase CPU power is some kind of over-clever marketing, as anything over 4-phase on the cpu strikes me as overkill, not to mention that it would have to be heinously hard to implement (I mean, really, 32 output FETS, 16 output inductors, and 16 or 32 filter capacitors!?!?)

[BobRoss]

Oh i see. So is that is how the motherboard regulates voltage? through pwm? interesting. I was just wondering about that the other day.

Thanks Gizmo!

[Gizmo]

It actually depends on the motherboard, and what voltages you are talking about.

The CPU is done on PWM, and has been since the P4 days. However, mobos may implement any combination of linear and PWM regulators for the RAM, AGP/Video, and chipset voltages.

Generally speaking, linear regulators provide 'cleaner' (i.e. less ripple and noise) power, use fewer components, and are easier to design than PWM regulators, but PWM regulators tend to be considerably more efficient, and at the higher end of the power spectrum can also be made more compact.

[Gizmo]

I've been researching the Asus 16-phase power a bit more. Aside from claims of higher power efficiency, and the controller switching from 16-phase at high loads to 4-phase at low loads, I've been finding precious little in the way of real information regarding the implementation. I've also seen references to 8-phase and 12-phase boards, although again I've seen nil in the way of real nitty-gritty details on implementation.

So, color me skeptical of this being anything more than a marketing gimmick until I can get some real information to sink my teeth into.

My gut instinct is to suspect that, because most people are used to counting FETs and calling each FET a 'phase', they are actually implementing a 4-phase controller, and dynamically switching between one and four FETs per actual phase, depending on the load. The marketing guys are then calling that '16-phase' switching. If that is the case, then it isn't anything new; high-output switchers have been doing that for a while, and I <THINK> that EPoX were actually doing that at least a couple years ago with their last mobo versions.

[BobRoss]

Does it stand to reason that if the sine waves of the phases cross at a higher point due to more phases you may be able to have less voltage droop? Though I guess that the caps should be smoothing it out anyway right?

[Gizmo]

In principle, yes, the more phases you have, the less droop you have between each phase. However, a larger capacitor is much less expensive than implementing an entire additional phase, and will usually work just as effectively. In addition, increasing the frequency of switching will also achieve the same effect.

Of course, there are practical limits to both switching frequency and capacitance, but there are also practical limits to the number of phases you can implement. As with everything else in engineering, there are always trade-offs to be made.

[BobRoss]

interesting,

Thanks Gizmo!