Pros and Cons of Digital PWM?

[OSKAR_WU]

Yeah, I know how Vdroop works, and I really don´t want it. Stable voltage is best for us XS members. Asus, DFI, Gigabyte... must implement an option in the BIOS to let us rock stable voltage under load, or a max of 0.01v~0.02v, like CPU Voltage Damper does in lastest Asus boards.

I measured it with the Smartguardian utility, didn´t know how to read it with a multimeter. Anyway, SM confirms what I saw in my tests: with the same CPU, Asus and my current Intel board behaviour is the same. Both measured with a well calibrated multimeter, after vdroop it needs 1,19v to a stable 3.3Ghz, a little OC. With the iCFX3200, 1.35v in BIOS resulted in a 1,28v idle (no problems here, typical Vdrop), but at load, it dropped to about 1,20v. At that voltage it should be stable like it was on Asus and Intel, but it wasn´t, 1,25v or 1,30v didn´t fix it. Even at that low OC, that huge voltage fluctuations kill the stability. I played with the iCFX for about a month, trying to understand what I was doing wrong, but honestly, I think that the problem was a faulty board or something, not me. I really liked all the BIOS options, great board, but the voltage related stability issue forced me to get rid of it.

There is Vdroop disable circuit in some of our P35 board , we will put this circuit in all future board with Digital VRM(except AMD board , AMD does not define clear Vdroop design guideline in their power spec) ...

SG utility is not calibrated unlike your reading in bios ... We have ask ITE to give us SG source code , we are working on it and try to see if we can make it more close to BIOS reading(which is calibrated) ...

About how much voltage to have stable over clocking result , I think it's vary from single board to another(even the same model) , so I can not comment too much what's the real reason you can get stable oc as other board unless I can get them and test by myself to see what's the possible root cause ...

[STaRGaZeR]

Oh I see, thanks

Yeah, stable voltage changes in single motherboard to another, but in this case uping it waaay up didn´t help.

I´m curious about what DFI is going to bring us related to X38, maybe if it is on time and voltage Vdroop disable option is present, I´ll give it a chance. Maybe you should put it clear in the package and specifications, I´m sure many folks will like it.

And please, no more green crap in the LanParty series, orange rocks. Keep us far away the heatpipe madness too, with the no transpiper version abailable since board launch date

[Jodiuh]

Wow...really good info here guys. Thanks Oskar...can't wait to get my P35 all UVed up.

[mesyn191]

Yea, great info. Oskar thanks for the help.

[humeyboy]

The Voltage Droop is in the Intel mobos as Intel makes the vendors add it.

Asus mobos always overvolted slightly on AMD.

[MJRL]

There are a lot of vantages for the digital pwm, besides of voltage control.
One of the essencials for me its a clean cpu socket zone

Compare this to that

And for the heat there is always solution

The guys who said there is a vdroop on PWM Digital should try DFI P35 and see it, now with PWM Dig 8 Phase i've never seen a board so stable out of the box.

[Raja@ASUS]

8 . In my personal opinion , with proper design in heat dissipation , the only cons is the price too high ... The other problem is the supply is not too good compare to traditional VRM , maybe this is why 1st tier player can not use it , they can not use a design which has supply problem ...

Oskar here in number 8 are you referring to parts availability for tier 1 who need big bulk orders?


regards
Raja

[bypolar]

The Digital PM on the DFI Boards get so hot with liquid cooler they shut the boards down. a fan is need to provide some airflow around the area to keep things in check. We also add some heat sinks to the chips.( The little Blue ones next to the Block)
[/IMG]

[Taner@XS]

My baby has a high temperature.
Blood Iron, CPU E2160

[Jodiuh]

That looks like a lack of airflow in the case...or high ambient temp.

And the pics comparing the Asus P5E3 Deluxe to the DFI...hilarious. The Asus looks like it's got warts. In contrast, the DFI's socket's gotta be one of the cleanest I've every played w/.

EDIT: fixed

[jimmyz]

nice to have your input on this oskar. i use phase change cooling so the digital is far easier for me to insulate and the temps are very reasonable. my ICFX3200 actually stayed around 0c according to sg.

off topic: Oskar you have made a simply stellar board with the DFI LP UT P35 T2R i am impressed even more every day. as i have said in other threads it is probably the finest intel board on the market. i can't wait for the x38 version also.

ot 2: oskar when you release a board like the x38 lp could you possibly post up an overveiw of the bios parameters and tweaks. I know Andre and others work really hard to get us enthusiasts the info we crave but it would be awesome to have the engineer's input.

[hipro5]

When the cpu load changes , no matter how fast the VRM can respond , there is always a very short period you need to use the Choke and Capacitor to maintain the voltage stable ... If we can design unlimited capacitor on the board , we can remove the droop design when the capacitor can handle all these transient state , but that's not possible ...

From the schematics you have uploaded, users can see how the power circuit responds from heavy load to light load (idle) and visa-versa.....

a. Digital OR traditional VRM with ALL the Mos-Fets in ONE LINE (straight) and cooling them with ONLY ONE heatsink, IS WRONG.....
Why?......EVEN if a user place the DEFAULT Intel heatsink onto the CPU, the motherboard is NEVER going to be in a "straight line" - take a look at it from it's side - .....
By this, the CENTRED Mos-Fets are getting WAY hotter than the sided ones....
Imagine now that a user buys another CPU heatsink and he starts to squeeze it much more.....
BIG mistake putting ALL Mos-Fets in one line with ONE heatsink on them.....
Do you want to talk now about the enthousiasts with the cascades/LN2s?....You get an "S" shape motherboard.....

I'm NOT talking about DFI here......I'm talking about ALL.....


b. When using a Digital VRM Mos-IC-Fet, you encode two problems:
1. TOO tiny surface to cool down with a heatsink and
2. in case that some user wants to takes off it's heatsink and place another one on them, they are TOO delicate and they break VERY EASY....(Even a corner of them if it breaks you're s(crewed....


c. NO MATTER how high switching frequency the Digital VRM you make to work for efficiency/etc., you CAN'T "filter" the "Voltage spikes" from idle to full load of a CPU.....
By NOT using electrolytic capacitors ALSO ADDED on their output circuit so to "catch" (filter) better those voltage spikes and "walking" on the reference design of Voltera (for example), you're doing NOTHING....


My thoughts:

a. If you want to use Digital it's OK BUT place those Mos-IC-Fet by TWO (and only OR by 4 in a square) with a quite big heatsink on them (forget heatpipes and crap like this)....
b. ADD some elecrolytic capacitors on their outputs...(SMDs like ASUS uses in some of their designs "catches" faster those voltage spikes - Vdroop of ASUS is another thing though and it has nothing to do with them - )
c. Use AT LEAST 8 Digitals so by placing them in square, you will need only two quite big heatsinks which MUST NOT TOUCH each other with heatpipe or anything else in case the motherboard gets stretched by the CPU heatsink or so....
d. DON'T use the "lined" chokes for the digital......when a user screws his mobo with the heatsink/LN2 pot/cascade, some of them break their ceramic cover and they become useless.....

.

[JoeBar]

Very nice read, as always, George!

[Pete]

True bloody facts

Very very nicely put there.

Now why are all the MOBO MFGS focusing on having people like Oska oc'ing the willys off the bord when what they need in the first place is someone like Hipro that knows the facts both at the engiering aspect but end user.

Hipro maybe you should get some mobo made of your own desgin, id pay for it

[hipro5]

Very nice read, as always, George!

Very very nicely put there.

Now why are all the MOBO MFGS focusing on having people like Oska oc'ing the willys off the bord when what they need in the first place is someone like Hipro that knows the facts both at the engiering aspect but end user.

Hipro maybe you should get some mobo made of your own desgin, id pay for it

Thanks guys, but that's just facts....

Oskar is a hell of technician you guys.....DON'T underestimate of him...

Have you guys ever thought that he's NOT alone?...He just CAN'T do whatever he wishes.....He's an employ.....So in some cases/designs sometimes "other people" may tell him what to put on or off of a mobo.....what parts for example.....cheaper, etc......

IF "they" leave him alone, I bet he could do miracles...

Another thing I whould also like to add :
Manufacturers GET R.F. technicians TOO......We are NOT anymore at 33MHz....We are WAY HIGHER than that.....
This is going (pointing) to anyone who can understand of what I'm talking about....

[dinos22]

hipro it's simple

make a little pot for the mosfets and stick some dice on it during the runs

[hipro5]

hipro it's simple

make a little pot for the mosfets and stick some dice on it during the runs

NOT that simple.....It won't "touch" ALL of them when mobo gets squeezed and changed it's "lined shape"...

Take a look at the below photo and HOW a mobo could become to fully understand of what I'm talking about.....



Now put me a "straight heatsink" on those (let's suppose) Digitals to cool them down.....


EDIT: BUT I know why you said that!.....You want that OVP mod of the NVIDIA....

.

[dinos22]

lol biatch

nah i'm serious

also in terms of bending there is a solution for that too

i could quote some nice pics MM posted a long time ago of his wooden board and rubber standoffs but CFB looking it up

you get the idea though

let me recap then ...... wooden/metal solid board for benching with good wide and fairly short rubber standoffs and DICE cooling on PWM and you are good to go....you will have no heat in that area of mobo and mobo should be dead straight mounted on the board..............i see no reason why it would get bend that badly

[Ace-a-Rue]

Thanks guys, but that's just facts....

Oskar is a hell of technician you guys.....DON'T underestimate of him...

Have you guys ever thought that he's NOT alone?...He just CAN'T do whatever he wishes.....He's an employ.....So in some cases/designs sometimes "other people" may tell him what to put on or off of a mobo.....what parts for example.....cheaper, etc......

IF "they" leave him alone, I bet he could do miracles...

Another thing I whould also like to add :
Manufacturers GET R.F. technicians TOO......We are NOT anymore at 33MHz....We are WAY HIGHER than that.....
This is going (pointing) to anyone who can understand of what I'm talking about....

that last para flew over my head but those para's above that last read so TRUE to my eyes and ears!

[coolrmaster]

and for all your troubles maybe this board should be given away for free too huh?

Oh I see, thanks

Yeah, stable voltage changes in single motherboard to another, but in this case uping it waaay up didn´t help.

I´m curious about what DFI is going to bring us related to X38, maybe if it is on time and voltage Vdroop disable option is present, I´ll give it a chance. Maybe you should put it clear in the package and specifications, I´m sure many folks will like it.

And please, no more green crap in the LanParty series, orange rocks. Keep us far away the heatpipe madness too, with the no transpiper version abailable since board launch date

[MrToad]

It has been mentioned before in the thread, but I also find digital PWMs a PITA from a mechanical point of view.

With my (defunct) IN9 32X I found myself in a catch 22 situation, which I was never able to overcome, and ultimately led to irreversible damage (I crushed the ceramic encapsulation).

With the default setup, and no OC, chips would shoot to what I understood as unreasonable 80-90 C.

Same setup, mild OC (X6800 @ 3.33 Ghz), they would shoot to 99-108 C under load.

Changing the thermal pad for a different type (the thick light blue one you often find in laptop GPU/VRegs) had no measurable impact on temperatures.

Liquid TIM was a no-go with the default HS, as it had (wisely enough) "legs", which left a gap enough to run freely a green rizla between the HS and the chips. Filling such void with Ceramique or MX-1 was impractical.

I looked at the possibility of making some sort of custom HS, but I didn't have at the time tools or a place to do so.

I finally acquired a water block from mips.de, and decided to give it a try. The problem I faced was... torque.

Too much and the PCB would warp leaving the centre chip with no contact whatsoever, and risk crushing the most external ones.

Not enough and the block would tilt due to the traction/torsion of the tubing.

The difference between "not-enough" and "too-much" was so little that achieving a good mount became a job for a 70 yr old Swiss clock maker.

In the end, my patience run out, and with it my pulse. I heard a sound like someone stepping on wet sand, and there went my MB.

In retrospective, even having used a back plate wouldn't have helped that much. Being the ceramic encapsulation so brittle, and with such tiny surface area, I doubt I would have managed to apply enough torque to leave the block sitting completely parallel to the surface without the same dreadful consequences.

[dinos22]

does DFI get THAT hot as well

[Ace-a-Rue]

digital PWM's, IMO, are fine at default clocking...but...once you start with the overclocking of a Quad core and then "jamming" with P95, you better have upgraded the cooling solution cause most times the current TIM under the HS will not keep up with the excessive temps produced by overclocking a Quad cpu.

[strange|ife]

thank you oskar wu for your info, i enjoyed reading it.

i wasnt exaclty sure what vdroop really was, till i read that. Lets hope for better designs in the future

[Raja@ASUS]

Digital PWM supplies for Memory and Chipsets may hold some merit, especially for DDR3. The lower overall current draw will probably allow for the Dmosfets to supply their full bandwidth. Using DPWM in these areas will undoubtedly increase cost, but may well be worthwhile for outright performance based motherboards.