CPU Power Phase Count

yojimbo

Limp Gawd
Joined
Nov 2, 2004
Messages
218
[H] Readers, Reviewers & Editors,

I'm going to get straight to the point: Why do I need 16, 24, or 32 power phases for my CPU? Seriously, why? With all of the P55 motherboards coming out of the woodworks sporting larger multiples of CPU power chokes and MOSFETs than the board that preceded it, I can't help but wonder. I'm not an engineer by any stretch of the imagination, which is why I pose my question here.

Here are my arguments:

1. Additional power phases do not aid in overclocking or stability.

[H] has reviewed six Asus X58 motherboards in the last year: P6T, P6T Deluxe, P6T6 WS Revolution, P6T7 WS SuperComputer, Rampage II Extreme, and Rampage II Gene. All six performed nearly equal at stock clocks, and they all overclocked to frequencies of 3.8 - 4.0GHz [I don't consider 200MHz significant and is within the tolerance of review sample variation]. The P6T and the Rampage II Gene are outfitted with 8 power phases to the CPU, while the other four have 16. If doubling the number of CPU power phases helps to stabilize power for better overclocking, then why is it that both the P6T (8 phase) and the P6T6 (16 phase) were able to both reach 4.0 GHz with equal stability using similar CPU voltage? Why is it the $250 board performed equally as well as the $400 board?

Now, it could be said that the additional power phases are useful for people who use extreme methods for overclocking, and this could possibly be true. The P6T6 WS Revolution allegedly holds the record for a Core i7 overclock, but I haven't heard a.) Anything about how stable that system was for doing anything but benchmarking, b.) What ridiculous kind of cooling apparatus it took to keep the CPU stable, or c.) How much the power circuitry played into stability. In the end, a power phase advantage does little for the majority of consumers. I, for one, prefer my system inside a case without having to constantly feed it liquid nitrogen to keep it from BSOD'ing on me.

2. Additional power phases do not help save energy.

While [H] may not include a look at power consumption as part of their motherboard review process (which I find intriguing, considering power consumption is an essential part of the video card and CPU review procedure), other review sites do, especially since 'going green' is considered en vogue these days. Asus, Gigabyte, and MSI all have their power saving tricks, and most of these seem related to regulating power phases around the CPU. Mind you, none of these seem to work properly while overclocking, which would mean that this 'feature' claim is aimed at the mainstream enthusiast.

There seem to be a multitude of ways to measure the power draw of the motherboard, but I don't think that matters much, so long as one review site measures power draw in a consistent fashion. With that in mind, I present to you part of a review from Bit-Tech. I purposely chose this review as it includes all three of the aforementioned motherboard companies and their implemented power-saving features on X58 motherboards. Despite the X58 chipset being an elite enthusiast chipset, it should clearly prove my point regardless of who power-saving features are marketed to.

Looking at Bit-Tech's chart, it would seem that the MSI boards have the lowest power consumption in any comparable setting of their power-saving program. Also worth mentioning, the MSI X58 Pro features 5 phases to the CPU and the X58 Eclipse has 6. The next nearest in low power consumption is Gigabyte's X58 UD3R, which features 8 power phases. Asus's P6T Deluxe, using 16 phases, consumes more energy with it's power saving features turned on than the MSI X58 Pro uses without any power saving feature running.

Could this be a poor implementation of a power saving algorithm? Maybe. Could it be related to the MOSFETs and chokes used? Probably, though I don't know by how much. It's also interesting to point out the Rampage II Gene, a mATX board, pulls more than most of the full ATX boards, and the Foxconn boards, both featuring 8 or 16 phase designs, manage to pull more power than either MSI, Gigabyte, or Asus. Go figure. I can at least summarize by saying that if more phases meant better energy efficiency, the boards with the better power phase distribution would have the lowest power consumption and not the opposite way around.


In conclusion, I have to ask again: Can anybody explain why I need more CPU power phases? Without any engineering experience, I have, for the most part, logically explained away (or, at the very least, poked significant holes in) any marketing fluff about the benefits. Can my fellow [H] readers, or perhaps the columnists or editors, help me figure out why Asus and Gigabyte are having a pissing contest over CPU phases? (E-peen stroking aside, that is...) Can anybody come up with a real-world benefit, if not help support my theory that there's nothing more to this phenomena than a bunch of PR BS?

Thoughtful answers and replies would be appreciated, BTW.

Thanks in advance.
 
Last edited:
Yeah, we figured that out a while back too. The reason G and A do it is that the vast majority of people that buy their boards do not know what you and the rest of us at [H] know.


There is no inherent benefit to the end user with more than 8 phases (and an argument can be made that 4 are sufficient) all other things being equal.

For the manuf its about "more phases" = smaller power handling ratings for the devices used thus allowing solid state inductors and smaller filtering capacitors. Devices that handle less power usually are less prone to failures but that is offset somewhat by there being more of them to fail. They are usually significantly cheaper. And of course there is the marketing angle I don't even want to get into.

Bottom line, the number of phases is basically meaningless. It is much more important as to how well the circuitry is designed and the actual devices used. Unfortunately that detailed information is almost impossible to find as board manuf do not disclose those specs. If you are based in electronics you can determine the pwm mosfet controller chips part number and the mosfets parts numbers from the board and make good generalizations, the same with the inductors and caps if you can find the data sheets.

So forget about it, you already have figured it out. A good 8 phase is better than a crappy 12 phase etc. etc. What you really need to know is which/if the CPU voltage regulation circuity meets or exceeds Intel's latest specifications for VRD (Voltage Regulator Down) circuits. And of course that is something else that is rare as hen's teeth.

Here is an example of what good motherboard specs would include if the manuf could get their heads out of their .... don't hold your breath. Intel latest VRD specs which describe the minimum requirements of motherboard CPU power:
http://www.intel.com/Assets/PDF/designguide/321736.pdf


Even just a "board complies with Intel VRD 11,xxx standard. " would be nice, not compatible, compliant and certainly not "supports". I have seen it a couple of times but on server boards if memory serves. Ha ha look at this.

http://www.gigabyte.us/FileList/WebPage/mb_080811_vrd111/tech_080811_vrd111.htm

Wait Wait, in the interest of balance. Asus "supports" it too on some models (fair enough) .
http://www.asus.com/search.aspx?searchitem=1&searchkey=p5q

WTF does "support" mean. Does it fully meet (aka compliant) Intel specs or not ? At least they do mention the latest VRD version so at least they read the thing. I hope.



Likely does not apply to the i5 i7 but I have not checked.

(If you are wondering if you hit one of my rant buttons - yep).


Oops got me so excited I let you slide on something.

2. Additional power phases do not help save energy.

True on the face but not expressed properly. It is true that supplies with a higher number of phases will provide smoother power when you shut some of them down when the cpu is under light load. So I would say it is true that additional power phases provide the opportunity for designers to implement circuity and algorithms that have the potential to save energy in some usage patterns.

http://www.intel.com/technology/itj/2008/v12i3/7-paper/6-voltage.htm

I fully agree that to date the focus is saving power at light loads when the cpu is not using much power anyway. One has to wonder if the cost and manufacturing and R and D that went into it will ever be recouped by any savings. I just put on on/off button in the OFF postion. eh, technology marches onward.
 
Last edited:
Thanks for your reply. I was kind of hoping I'd hit someone's rant button, and I'm glad you replied with as much enthusiasm as you did thoughtfulness. However, your response begs more questions:

In light of a recent AnandTech article from Mr. Shimpi himself regarding the cost of P55 boards and the cost of components, which is obviously not very detailed but gives a general idea, I would think more power phases means driving up the cost of a motherboard. Let's assume a good quality power delivery circuit (which I imagine refers to a set of MOSFETs + choke, by Anand's description; correct me if this sounds wrong) costs $3 each and okay ones cost $2. That would mean, in theory, a 'good' 8 phase power distribution would cost as much as an 'okay' 12 phase power distribution, and the only benefit to 12 over 8 would be 'smoother transition' as load increases/decreases. While this is just my opinion, I'd imagine the difference in rate of failure would be a wash. 12 crappier-but-more-evenly-distributed phases is just as likely to fail as 8-decent-quality-and-built-for-more-abuse phases. In theory, that is. I guess the bottom line is: Assuming a good working power phase switching algorithm is implemented for either, would an end user ever notice the difference between the two power designs?

Tipping off from that question, what also begs to be answered is why a mainstream chipset is being used as a vehicle for deploying ridiculously overwrought power designs? I would have thought a lower-powered chips such as Lynnefield and the P55 would need less power circuitry, and companies vying for a customer's money would have found a way of wracking as much money out of a good performance mainstream board that didn't cost so much to make as a DQ6 Extreme, ROG, or Classified board would. As a consumer, it's almost off-putting to see a mainstream chipset dolled up as if it were a top-of-the-line offering.

From an energy saving standpoint, I have to wonder what MSI is doing right and everyone else is doing wrong when it comes to energy consumption. They make off with less power phases on their X58 boards, provide similar features for people to play around with, and somehow consume anywhere from 8 to 75 watts less than Gigabyte's or Asus's similarly designed boards (depending on energy saving settings, of course). The performance is pretty much the same, give or take a few MHz on the overclock. Is it their power design, their energy saving algorithm, both, neither?
 
Last edited:
[H] Readers, Reviewers & Editors,

1. Additional power phases do not aid in overclocking or stability.

Yes, you're correct. Additional power phases doesn't directly improve oc ability. They exist for two reasons:

First, they help reduce ripples. Ripples are harmful to the stability of the system, especially for high frequency circuits.

ripple.png


Secondly, currents are now distributed more evenly so that no mosfet will overload or produce too much heat.

2. Additional power phases do not help save energy.

You're correct again. Additional power phases doesn't help save energy. On the contrary, they consume more energy. This is why mobo manufacturers want to "shut down" some phases when a system is in light load. They'll show you their mobos can "save energy" compare to the same mobos under full phases on modes. But since you cannot compare the effectiveness across different manufacturers, or even different mobos from the same manufacturer, you can not tell if these programs are actually helpful.



Finally, many mobo manufacturers and reviewers are wrong when counting power phases. For example, the Gigabyte GA-P55-UD6 is not a true 24 phases design. The number of phases is not determined by the number of chokes, capacitors or pair of mosfets. They're actually determined by the number of mosfet driver chips. The difficulty is, however, that some chips control 3 or more groups of mosfets while some chips control only one group of mofsets. To make matters worse, some mosfet drivers are integrated into the PWM chip. With all these mosfet driver chips and PWM chips in different package sizes on a real motherboard, it really require a lot of knowledge and time to get things correctly.
 
Well bottom line we do not have enough information to really answer the cost question. More phases means each phase handles less power for a shorter period of time. This allows much smaller (as far a power handling goes) devices throughout that phase circuit. You will notice high phase count boards do not have the "can" type capacitors or the "cube" inductors but instead an IC flat pack kinda looking thing (solid state inductor) and rows of tiny rectangles which are a bunch of surface mounted caps in parallel so that the smaller capacatance values add making up for the small size. A good quality can type cap will cost 30 cents, ten of the surface mount ones with cost a penny each and are more suited to automation in board placement and other manufacturing concerns. Much cheaper, same for the inductors. So yes there are a lot more parts but in general they ususally would come out cheaper and with significant reductions in manuf costs. If even one inductor or cap has to be hand placed because of the thru hole technology (leads protrude through the board thus requiring robotic or manual component insertion) it costs more in labor to install that one part than to have the onsetter machine place the surface mount components for most of the board. (you buy the machine and that it, load reels of parts and stacks of boards and it can run almost 24x7 and does not get overtime pay or medical benifits.

So it really is less expensive to use more parts that handle less power and at the volumes these guys buy parts doing a design that requires a high volume of fairly inexpensive parts is way cheaper to buy than a medium volume of medium expensive parts.

All the the above are generalizations based on my fairly vast (but somewhat dated now- this was 10 years ago) experience in design support of a manufactuing line of DC to DC power converters for a telco.


The new i5 and i7 have more complicated power requirements than the C2Ds. However your point is not completely off base. To some extent it is a marketing game. It always has been a battle between the engineers and marketing and accounting/purchasing. The engineers if they could would design the best solution within some reasonable cost parameters, marketing tells them what they have to design regardless of engineering needs and no matter what accounting thinks it costs too much.

Here is some info on the i5 and i7 that is not too technical and highlights the differences in the cpu power needs.

http://www.lostcircuits.com/mambo//index.php?option=com_content&task=view&id=44&Itemid=42
 
Well bottom line we do not have enough information to really answer the cost question. More phases means each phase handles less power for a shorter period of time. This allows much smaller (as far a power handling goes) devices throughout that phase circuit. You will notice high phase count boards do not have the "can" type capacitors or the "cube" inductors but instead an IC flat pack kinda looking thing (solid state inductor) and rows of tiny rectangles which are a bunch of surface mounted caps in parallel so that the smaller capacatance values add making up for the small size. A good quality can type cap will cost 30 cents, ten of the surface mount ones with cost a penny each and are more suited to automation in board placement and other manufacturing concerns. Much cheaper, same for the inductors. So yes there are a lot more parts but in general they ususally would come out cheaper and with significant reductions in manuf costs. If even one inductor or cap has to be hand placed because of the thru hole technology (leads protrude through the board thus requiring robotic or manual component insertion) it costs more in labor to install that one part than to have the onsetter machine place the surface mount components for most of the board. (you buy the machine and that it, load reels of parts and stacks of boards and it can run almost 24x7 and does not get overtime pay or medical benifits.

So it really is less expensive to use more parts that handle less power and at the volumes these guys buy parts doing a design that requires a high volume of fairly inexpensive parts is way cheaper to buy than a medium volume of medium expensive parts.

I think the smaller magnetics/lower capacitance are the result of a higher switching frequency design, which is more difficult to implement due to pcb design issues, among others.
 
Last edited:
I think the smaller magnetics/lower capacitance are the result of a higher switching frequency design,

Excellent point and one I missed. The values of inductance and capacitance to filter out a higher frequency "noise and ripple" are smaller. Not too sure about the difficulity of implementation, the new digital PWM controller chips are pretty sweet and the frequencies are nowhere near cpu clock/buss speeds but still, I cannot argue that they certainly are more complex, require more board layout work etc. etc. so that point is also a good one.
 
I read that using lots of phases also helps when the load varies widely and frequently, as it does when aggressive power-saving is used.
 
Back
Top