VRM phases: voodoo?

All about them.

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
BrianF
Posts: 151
Joined: Sat Apr 16, 2005 2:59 am

VRM phases: voodoo?

Post by BrianF » Wed May 09, 2018 3:08 pm

I remember a looong time ago, back when Intel still made and sold their own branded boards (how I long for those), they had a white paper which basically talked about how a board with a pile of VRM phases was not necessarily a good thing. Some cheap boards they said, use an abundance of them to overcome the fact that they suck. Better to have few well spec'ed and built phases they said.

Then I recently read something Steve said a long time ago about about a mother board choice: "I don't want a board with zillions of VRM phases - it helps for overvolting, but impacts idle/low power efficiency."

Got me thinking...

How are we supposed to disseminate between boards with a handful of VRMs (and maybe no heat sink) and ones with a dozen (with as much heat sink as some low end CPUs). Either could just as likely suck.... or be the paragon of electrical engineering. Who knows?

Does counting VRM phases really tell us ANYTHING about a motherboard anymore?

Food for conversation... or just thought as case may be.

nagi
Posts: 125
Joined: Tue Aug 31, 2010 6:48 pm
Location: Outside the box

Re: VRM phases: voodoo?

Post by nagi » Thu May 10, 2018 6:24 am

Here's the thing: VRMs are not created equal. There are high-end ones that even if you have a dozen of them, will have smaller losses in total than the cheap-o solutions that some low-end and even medium quality boards sport. Yes, if two boards would use the exact same components and differ only in VRM count, more would mean a higher idle draw. But there are no boards like that. The boards with more VRM phases typically have more other extra features too.

The whitepaper is of course still right - more VRMs mean higher power draw, and also more stable output. This means more stable OC, and also more means higher potential power draw for the CPU. Also more surface area to dissipate the VRM losses on, though that's rarely an issue that could be solved with adding just extra 2 cm2 area. And yes, you could take a higher number of shitty VRMs to produce a semi-usable output.

Also, many boards use phase doublers or sometimes even multiple chokes per phase, meaning counting those is, in itself, of no use. You have to look at the controller, the layout, etc.


A youtube channel to check for this is Buildzoid's Actually Hardcore Overclocking. Though his intentions are far from our (liquid nitrogen / other sub-ambient cooling records), and until recently, he mostly reviewed videocards, one could learn a lot from his videos. I know I did.

CA_Steve
Moderator
Posts: 7650
Joined: Thu Oct 06, 2005 4:36 am
Location: St. Louis, MO

Re: VRM phases: voodoo?

Post by CA_Steve » Thu May 10, 2018 11:33 am

As Nagi pointed out, every mobo might use different parts/VRM circuitry and bottom of the SKU stack boards may go with lesser efficiency circuitry. There are still a few sites out there that collect lots of motherboards and compare them. So, you can at least get an inkling of idle power draw.

BrianF
Posts: 151
Joined: Sat Apr 16, 2005 2:59 am

Re: VRM phases: voodoo?

Post by BrianF » Fri May 11, 2018 3:29 am

Thank you both for the resources. I'll have to make time to sift through Buildzoid's channel (overwhelming amount of interesting material).
CA_Steve wrote:There are still a few sites out there that collect lots of motherboards and compare them. So, you can at least get an inkling of idle power draw.
Forgive the remedial electrical engineering question, but what is the implication we draw from that? I normally would take a low power draw at idle to mean "efficient", but isnt it still a crap shoot as to whether such a design can "step up" when needed under load?

Clouding my understanding is the whole "doubling" of phases thing. For example I looked at two different boards, one had 4 phases in a 2high/2low config, the other had 8 phases in a 1high/1low config...so at the end of the day the same total number of MOSFETs and presumably (all else being equal) similar total capacity? So whats the advantage of the latter (presumably more expensive) configuration? Stability?

CA_Steve
Moderator
Posts: 7650
Joined: Thu Oct 06, 2005 4:36 am
Location: St. Louis, MO

Re: VRM phases: voodoo?

Post by CA_Steve » Fri May 11, 2018 9:00 am

BrianF wrote:Forgive the remedial electrical engineering question, but what is the implication we draw from that? I normally would take a low power draw at idle to mean "efficient", but isnt it still a crap shoot as to whether such a design can "step up" when needed under load?
That's why you also look at the load power graph.

BrianF
Posts: 151
Joined: Sat Apr 16, 2005 2:59 am

Re: VRM phases: voodoo?

Post by BrianF » Fri May 11, 2018 9:31 am

CA_Steve wrote:That's why you also look at the load power graph.
Just so I understand, what we like to see is low consumption both at idle and under load because that would be a strong indication of an efficient design, yes?

But is it still possible that a given mainboard exhibits a low consumption under load not because it is efficient, but because it sucks and is shortchanging the CPU?

CA_Steve
Moderator
Posts: 7650
Joined: Thu Oct 06, 2005 4:36 am
Location: St. Louis, MO

Re: VRM phases: voodoo?

Post by CA_Steve » Fri May 11, 2018 9:52 am

That's why you look at the reviewer's CPU benchmarks for stock and OC performance. :)

It's a multi-variable world. You look at a set of benchmarks and use them to make an informed decision about what product best meets your needs...which will be different than the next guy.

BrianF
Posts: 151
Joined: Sat Apr 16, 2005 2:59 am

Re: VRM phases: voodoo?

Post by BrianF » Wed May 16, 2018 3:50 am

Found this article which has some interesting information, but perhaps outdated?

One thing he mentions is that a phase will always be made up of two mosfets, one-high one-low, with either a driver IC OR a third mosfet acting as the driver.

I dont know if this holds true anymore because I see a lot of board, even some rather high-end ones, using three mosfets per phase but not because one is being used as a cheap driver. Rather they are in a one-high two-low configuration (which I gather is a way to use cheaper fets in tandem to get a lower RDS(on) on the low side?). I've even seen boards with 4 mosfets per phase in what is apparently a two-high two-low configuration.

All else being equal, if we assume the same nominal output, are 3 or 4 fets per phase designs indicative of lower quality compared to a "traditional" 2-fet design?

Also, what happened to the "integrated" chips which incorporate the 2 mosfets and the driver in a single chip? I haven't seen anything like that lately. I was surprised when I examined my 6 year old Intel DH77KC and found exactly that.
Last edited by BrianF on Wed May 16, 2018 7:39 am, edited 1 time in total.

BrianF
Posts: 151
Joined: Sat Apr 16, 2005 2:59 am

Re: VRM phases: voodoo?

Post by BrianF » Wed May 16, 2018 7:06 am

Just for conversation...

From left to right (in terms of CPU vCore): 8 phase (2 fets per) / 4 phase (4 fets per) / 4 phase (2 fets per). There are all non-OC boards (H370/B360).
VRMCompare.jpg
Now the really interesting thing here is that the PWM chip, High fets, and Low fets of the first two are identical. Make you wonder if the first really is 8 phase or if they just double up on chokes.
You do not have the required permissions to view the files attached to this post.

Post Reply