Proper Supply Voltage needed for DFI nF4 MB for stability

[EMC2]

Why You Want to Keep Your Input Rail to the Vmem Regulator Circuit 250mV, or more, above your Actual Measured Vmem Voltage
(Subtitle: Excerpts from a Looney OCer about Vmem Supply Voltages)

For several months now I've been suggested to people to make sure that the supply rail to their memory circuit on the DFI nF4 MB is several hundred milli-volts above whatever they plan to run their memory voltage at... and peeps are probably tired of hearing it by now

Well, I sat down this evening with some UTT memory, a good scope, and a BIG cup of java to look into another subject and along the way the java jumped up outta my cup and wacked me up side the head , said "go ahead, show 'em why you've been prattled on about that dude." (I knew I made that pot of coffee too strong! )

So along the way to doing some other things I grabbed a few o'scope shots that illustrate the point rather well and decided to pop up :worm: tonight and share. I'll agree that they aren't anywhere as near as interesting as a lot of other pics I've seen in these forums (especially the "scw thread" ), but what the hey

Here we have Mr. Stability throwing out the first pitch with Vmem set at 3.0V in the BIOS:


Batting 1st, we have Vmem set at 3.0V +0.03V in the BIOS, playing MemTest T8 (left field):
3.0V +0.03V during Memtest T8


2nd up, we have Vmem set at 3.1V in the BIOS, playing MemTest T8 (center field):
3.1V during Memtest T8


3rd up, we have Vmem set at 3.1V +0.03V in the BIOS, playing MemTest T8 (right field):
3.1V +0.03V during Memtest T8


And batting cleanup, we have Vmem set at 3.2V in the BIOS, first in MemTest T8 (designated hitter):
3.2V during Memtest T8


and here wiffing at the ball trying to knock it out of the park, playing Windows:


What does it mean (besides that I'm loony as a bedbug )... well there are two morals to this story.

The first one is make sure you feed your MB properly:
If you use a "standard" 3.3V rail to supply your MB, don't go past 3.0V in the BIOS for Vmem.
If you use a "tweaked" 3.3V rail and the jumper trick to get higher voltages, keep the 3.3V rail at least 250mV higher than what you have Vmemset for in the BIOS. For example, if you are going to set Vmem to 3.2V, your 3.3V rail needs to be 3.55V or better.

And the second moral to the story is don't make your java too strong

We now return you to your regularly scheduled OC'ing session.

Peace

[Ugly n Grey]

Nice.

[bachus_anonym]

Man... I just love that scope

Nice stuff there, Daniel! This is the very explanation that many were looking for. Nothing better than getting down-and-dirty and bringing real-life testing.

Kudos

[flexy]

Man... I just love that scope

Nice stuff there, Daniel! This is the very explanation that many were looking for. Nothing better than getting down-and-dirty and bringing real-life testing.

Kudos

yes indeed a nice scope.

But the vdimm issue itself - you dont need a scope, not even a DMM

Just use mbm w/ the NF4 data file (posted elsewhere) and check your DRAM voltage.

You can also config eg. OCCT so it shows the DRAM voltage in any of its dsiplays (like i do with MBM set up right).

Running off 3.3 rail (since TCCD) and up to 2.9 super stable - VDIMM rail is steady....3.0V SLIGHT fluctuations....and at 3.1 VDIMM volatge in OCCT shows 3.14 and fluctuates from 3.00 ---- 3.14 all over.worst i've seen was even 2,99.

Not sure what thread it was (i think even a EMC thread ) where they explained how the voltage regulator craps out once you get REALLY close to the limit of the 3.3rail.

And, after all, THIS is one of the main reasons people want to volt mod their rails so they can shoot higher AND get more stable VDIMM.

As said, just use OCCT or MBM with the nf4 datafile and watch DRAM fluctuate

[mikead_99]

I agree that you don't need the 'scope pictures to figure this one out, but it is nice to see it charted in this way. Makes it clearer as to what we face when we hit that < .25V threshold. Nice work EMC2.

Now, about that vtt.....

[QuikSilver]

Would it be wise to use the 5v rail for the ram after 3v then? I am running my ram @ 3.2V on the 3.3v line and it is somewhat unstable...

EDIT: yes, it fixed it! running 3.2v on the 5v line makes it rock solid under load thank you for making this thread

[bachus_anonym]

@flexy

Yes, I know one doesn't need a scope to do that. I can see very slight VTT fluctuations with a DMM, but what I love about that scope is logging readings over time Cool stuff... You just hook channels up, start it, sit and watch and take a screenie at the end I kind of stopped looking at soft reading a while ago... I mostly use DMM these times Even with NF4 config file, MBM is about .1 -.05 of on all voltages in my case, compared to DMM readings...

I knew about 0.2-0.3 overhead since NF2 times, but EMC2 gave excellent explanation that was needed for some to understand it

[flexy]

I agree that you don't need the 'scope pictures to figure this one out, but it is nice to see it charted in this way. Makes it clearer as to what we face when we hit that < .25V threshold. Nice work EMC2.

Now, about that vtt.....

yep That's *my* concern too.

[LowRun]

Very nice. Man, i wish you could get your hands on most of the PSUs around and review them with the appropriate knowledge for us the ignorant crowd. Thank you.

[Gogar]

Does any other hardware run on the 3.3v rail? It can't hurt to run it at 3.5v right?

So you proved that a high vdimm on the standard 3.3v rail causes instability.. But does raising the 3.3v rail actually solve the problem?
I mean you conclude that you should run it higher to get stability, but i don't see that proven in the data.
You're probably right though sorry to whine lol.. but i like to be sure.. for all i know raising the 3.3v rail to 3.5v could introduce instability as well.
I sure am whining tonight... someone should come and whack me over the head... coz i just realised that with the vdimm jumper you put 5v on it.. and here i am ing whether 3.5 could do harm...hahaha sorry!
Great work dude! ::

[EMC2]

I can see very slight VTT fluctuations with a DMM, but what I love about that scope is logging readings over time

A DMM is too slow to see a lot of things Regarding the scope... it's nice, but there's another one...

You just hook channels up, start it, sit and watch and take a screenie at the end

You make it sound soooooo easy... 1/3rd of the battle is properly instrumenting for accurate readings at the proper place... 1/3rd is a suitable trigger (tho in this case the trigger was easy... and most of the other 1/3rd is divided up between the conditions to recreate what your searching for and cross-checking the results Now if only I could free up the analyzer from work for a bit

Even with NF4 config file, MBM is about .1 -.05 of on all voltages in my case, compared to DMM readings...

Wish there was the capability to add a correction factor to MBM.

--- Flexy ---
Did you mean this post?

--- Gogar ---
Yes, quite a bit of other stuff runs off the 3.3V rail... but most of the time as long as you stay within 10% of nominal, you're fine (in this case 3.6). Heat is usually the only "watch out for" if you follow that guideline. And 99.9% of everything is rated to handle at least 5% tolerance on the rails (which would get you to 3.45V).

In regards to curing the "instability"... the drops in Vmem shown are caused by simple IR drops (the current going through components causes voltage to be "dropped" across them, reducing the voltage available down stream). By keeping the source voltage greater than those drops, you prevent them from showing up on the output (in this case the Vmem level).

Oh... and

--- LowRun ---
Actually, I did look at one PS on here... go look in the Hardware, PS subforum


All... thanks Just thought a few pics would help people understand

[STEvil]

There is a correction factor for MBM, at least for the temperatures.

I would not trust the onboard sensors anyways, use a DMM/Scope or dont bother.

[LowRun]

--- LowRun ---
Actually, I did look at one PS on here... go look in the Hardware, PS subforum

Yeah, i've read your PCPnC 850W review some time ago, great reading but i don't have the kind of money it takes to get this beast home and i can't even find one here. I'd like the same kind of review but for more mainstream PSUs (Antec, Enermax, Hyper, Fortron...), that would be priceless.