Multi-stage VRM and Presler

[Liquid3D]

Hi all I'm new here please excuse my ignorance.

I'm testing the Gigabyte G1975X which was designed primarily with multi-core CPUs (Presler) in mind. The board utilizes 8-Stage power conversion (or CPU power circuitry) which was touted by the manufacturer to be more stable with such PSUs. Unfortunately the board isn't stable and just like most others may benefit from a Vdroop mods.

Since the advent of multi-stage power conversion some have countered it was merely a cost cutting move since capacitors would have to grow ever larger and the cost would eventually become prohibitive. Yet looking at Intel's D975XBX refrerence board whch uses better mosfett cooling and seemingly more capacitance at "..four 1200 mkF and eight 560 mkF capacitors." I didn't hear about any instabilities in any reviews?

Of course Intel ref boards aren't known as overclocking platforms, nor do they ususally allow BIOS option to support it. Gigabyte on the other hand does, however their "Turbo" cooling alternative new on this board, while effective only cools 50% of the power circuitry? The other half is located so only a Intel Stock cooler or another HSF using secondary cooling onto the socket-area will help to cool it.

Since the board was intended for "Gamers" and "Enthusiasts" whom would most likely employ H20, phase-change or large OEM Tower coolers (with heat-pipes raising the "bulk" of the fins and fan/s away from board devices) it seems their solution wasn't well thought out. Irregraless between Intel's power hungry CPUs which are handicapped by thir own design guidelines (ATX 2.01) capping 12V Rails at 18A Presler looks to be one of the most unstable CPUs on the market (IMHO).

A knowledgeable friend explained the reason multi-stage VRMs are used was because Inductors would have be as large as the Capacitors on power conversion pre-dating 3,4,5,6,8-stages. He also said the reason we're experiencing "Vdroops" are; "...due to the inductor arrangement on these motherboards." and that; "People are pushing the VRMs to insane levels..." So while mobo makers spec the most cost effective parts (least expensive); "There is going to be a simple IR drop across all that wire..."

I need help understanding how these mutli-stage vs older power-conversion designs work. And will our only reprieve be to find PSUs which have very high 40A+ 12V rails? Most of which of course do not meet the ATX 2.01 spec which as I understand hasn't been required in over a year, yet hasn't been "Offcially" dropped?

I need someone to explain this to me like I'm a four year old, because I'm struggling with the issue and reading engineerin papaers is sometimes difficult wihtout the analogies which help the layperson? I hope this isn't too Off-topic for the fourum, and i apologize for the Post length. Thank you...

[Per Hansson]

Re: Multi-stage VRM and Presler

Well, the first thing that comes to my mind is what sort of powersupply do you have and what sort of cooling does it employ?

Because if it is one of the cheaper high-wattage supplies it will only be reated to deliver say 400w at 20°C (which most rooms are hotter than to begin with) and as the temp inside the PSU increases it's efficiency decreases. Say that the PSU is not getting very good cooling at all, it might very well be 60-70°C inside of it, in which case it would not be able to deliver even 200w IMHO

Of course I could be all wrong too but since you make no mention of it I have to ask...

And also are the capacitors in it ok? Because if they are not you might be sending crappy voltage to the CPU's VRM setup, in which case it will have to work much harder and get even hotter to deliver clean current to the CPU...

[gg1978]

Re: Multi-stage VRM and Presler

basically, the more phases you have in your VRM, you can use smaller MOSFET's for each phase for a given design load, and you need less output capacitance to filter the VRM output.

Of course if you need more Vcore amps, the reduction in output capacitance is probably less of an advantage. The more caps you have in parallel for your Vcore output, the higher the amperage available for your CPU.. For example, an Epox 8RDA+ i just recapped has 6 3300uF caps for the VRM output, which is over 18000uF total. The 8RDA+ is a 3 phase design though..

I also believe the inductor value being less for each phase is correct also..

Of course a 8 phase design costs more for inductors and MOSFET's, as you have at least two MOSFET's and one inductor per phase.


The inductor arrangement shouldn't be causing Vdroop faults. Vcore drooping means that the CPU is taking more current than the output caps, MOSFET switching phases, and input caps can supply.


Of course the fact that the VRM input is +12V doesn't help much either. The bottom FET's have a much higher duty cycle compared to the Top FET's, hence why some motherboards have two bottom FET's and one top FET per phase. The bottom FET's do all the heavy lifting, while the top ones don't do much at all.

Of course going to +3.3V for VRM input would be interesting, as you'd have to have a lot of +3.3V wires from the PSU to the motherboard, but at least the VRM MOSFET's would have a more equitable duty cycle.

Do you know what kind of caps Intel is using on the D975 board?? I'd be willing to guess the VRM output caps are either Sanyo OSCON's or Fujitsu polymers. Polymer caps have more current capability under normal operating conditions.

[willawake]

Re: Multi-stage VRM and Presler

intel uses 5 phase. is gigabyte really using 8 phase on this board? doesnt have their 8 phase add on card, looks like 4 phase really???

according to the reviews :

gigabyte : 7x560, 4x1500, 3x1000 = 12920uf
intel : 4x1200, 8x560 = 9280uf

both are using nice oscon sepc for vrm output.

the intel board took a long time to come into production. i think the gigabyte was one of the earliest. the intel board appears to have been released at revision 302 which is curious, i would have expected such a revision number to be of a product months old. Revision 303 will be released in march and will have the following :

1. Updating the BIOS to revision 0442.
2. Adding an alternate SPI (BIOS storage component). The existing SPI
part will continue to be used as a primary component and a new revision
of the SPI part will be used as an alternate if needed.
3. Beginning to phase in the new Intel logo on the PCB silk screen.

i dont see anything there to suggest problems with the reference design.

regarding assisting you to understand vrm in order for you to develop a vdroop mod, unfortunately i would consider such a mod to require more than a basic understanding.

compared to the intel board, the mosfet sinks do look poor on the gigabyte, that could be easily user rectified if they feel quite hot. although it does have that active cooling system which looks crap and noisy, although reminds me of 1u rackmount style cooling. The northbridge hsf looks quite effective though, although greatly smaller than the intel passive solution. better take some temp readings with probes to check if anything needs to be done with those sinks. you could anyway remove them, and apply arctic silver instead of the stock tim

how does the power connectors work? the gigabyte manual was rather bad in explanation. i assume for best effect you plug in the 24pin atx and the 2x4 12v connector like on the intel board. although it appears optional like you could plug in only a 1x4 12v. what method did you use?

i guess that is you posting here
http://www.ocforums.com/showthread.php?t=442132
great photos of Wyoming btw.

are you using the PCPower&Cooling TurboCool 850SLI in your sig or similar on that board? if you are then i guess you meet Per's requirements for a decent psu lol

personally if you are using a good psu and the cooling is good then probably that board has a bad vrm design. i would watch out for a later revision and request that via RMA if possible. regarding modding, looks like 2 caps are missing in vrm and could be added.

[Liquid3D]

Re: Multi-stage VRM and Presler

Sorry to revive this old thread just wanted to thank everyone whom respeonded. I thought it was unfair of me to visit this forum register ask one question and not even have the courtesy to respond.

the odd thing is while Googling mosfett H20 coolers this thread got tagged in the search, it reminded me of how much of an "ass" I can be sometimes. With that apology I'll try to visit more often, since I do find this forum valuable.