DFI NF4 Ultra-D vdimm mod: come inside.

Quote:

---

Originally Posted by JNav89GT

time to go off to get my 3rd degree, this time as EE so I can understand this stuff.

Maybe I speak in medicalease

---

It's easy to understand John... just imagine that Vtt is your blood pressure and an equivalent event would be your BP dropping to 70/40 for a few moments when you first see a really fine bikini bendover due to increased blood demand ;)

--- Craig --- More than I would spend for personal use...

Coming up with fixes is easy... doing one that is affordable and can be done by most is a bikini of a different color :lol: Until then, my suggestion is to use Memtest T8 to fine tune your drive strengths ;)

Peace :toast:

I'm currently running on a sio mod, 3.5 volts. But still planning on doing the op-amp mod.

emc2, am I reading your last scope right ...1.38V *2 = 2.76 vdimm - so it looks like you are using tccd (or other lower voltage mem).

1. Would you expect that vx, at 3.38 vdimm types of voltages will also show as dramatic of load dips?

2. were you lowering or raising drive/data strength to reduce the fluctionations. Any suggestions for vx types of voltages I've been running 8/3 just reduced to 7/2 and running test 8 for the night to see how it acts.

Thanks for the information/work on this

Chad

p.s.

3. when analyzing voltages at this level of detail ... do orange/yellow slots make any difference?

4. exactly what brand/model of memory are you using for this test?

yeh i am kinda wondering too... TCCD pulls more amps. thats why it uses less volts. what ya think?

--- Chad --- Ummm... look at what C2 is labelled as ;) Nominal Vtt is 1.455 for that test run... and yes, for that test I was using TCCD.

1) Yes

2) This weekend I'll be back on high volt mem, let you know then regarding #s. From what I've seen to date, lower drive strength (based on info about which way the numbers run on DFI) lessens magnitude of the drop... but too far of a drop and you loose Mhz from edge rates slowing down... it's a balancing act with stock Vtt config. Oh... and think about what kind of memory needs higher drive strength ;)

--- Malves --- based on some other info I uncovered today, this board definitely could benefit from better high frequency decoupling on Vmem and Vtt...

Peace :toast:

ps :p:
3) Not as far as this Vtt issue goes...

4) I'll post that elsewhere later ;)

Does anyone want me to try to design a fix for the Vtt droop problem?

Quote:

---

Originally Posted by Malves

Trying to find where I read about board being damaged if selecting vdimm in BIOS > 3.3v rail. Can't find it, thou.:(

---

Your 100% right.. I cant find where I read it either..

But I think it was one of the OCZ guys that said if your using the BigToe mod and raise the vdimm past what your 3.3v rail is set to it will cause major problems and might damage the mobo..

on vx, drive=7, data=2 tested well over night - #8 no errors (I was 8/3 and was getting about 5 #8 errors in a 6 hour period)

Quote:

---

Originally Posted by EMC2

--- Malves --- based on some other info I uncovered today, this board definitely could benefit from better high frequency decoupling on Vmem and Vtt...

---

Waiting your PM with more details.:D

Quote:

---

Originally Posted by persivore

Does anyone want me to try to design a fix for the Vtt droop problem?

---

Plz do.
What more mods are there for this board, if any at all ?

Quote:

---

Originally Posted by THunDA

Your 100% right.. I cant find where I read it either..

But I think it was one of the OCZ guys that said if your using the BigToe mod and raise the vdimm past what your 3.3v rail is set to it will cause major problems and might damage the mobo..

---

but the highest option in the bios is 3.3v when using the bigtoe mod :stick:

Nope, I have 4V avalible. I have also used the 4V option while running off of the 3.3 line just to figure out how much voltage is lost from the 3.3 line to the memory, my board is undamaged and my 3.3 line was only at 3.8V. (Just if you are wondering, it's about .2V less than the 3.3 line)

Well... after another round of the Java dance I found that the Vtt issue is a bit more Xtreme than I thought...

Ran a little Spi1M and saw this:
http://img146.imageshack.us/img146/9...gspi32m4cr.png
Then during the initial ramp-up of OCCT, I saw this, about same magnitude as the Memtest T8 drops, but shorter duration (like the Spi1M dips too):
http://img8.imageshack.us/img8/7164/...cctdips5lo.png
Then this is what showed up just as OCCT failed:
http://img146.imageshack.us/img146/7984/occtdip18sn.png
After giving it another go I got this "beauty" after about 5 mins:
http://img146.imageshack.us/img146/4841/occtdip21ek.png
Funny part... this was at sub-standard clocks speeds :lol:

Notice the captures regarding Vtt during OCCT failures are much worse than the MemTest8 ones. Those are about a 17% deviation from what Vtt should be. The part used is rated for 3% regulation. Hmmmmmmmm.......

Those "OC's like a formula 1 car today, won't run at stock next day" comments are starting to make sense...

Oh well, it's Java time :toast:

Quote:

---

Originally Posted by craig588

Nope, I have 4V avalible. I have also used the 4V option while running off of the 3.3 line just to figure out how much voltage is lost from the 3.3 line to the memory, my board is undamaged and my 3.3 line was only at 3.8V. (Just if you are wondering, it's about .2V less than the 3.3 line)

---

opps i guess there is a higher option, sorry my bad :rolleyes:

i tried having my 3.3v rail at 3.87v and set the bios to 3.7v and i got 3.7v but the fluctation was HORRIBLE. the vdimm went down to 3.56v in memtest #5 :( this is on a OCZ Powerstream 600watt...it didnt fluctuate though when i have it at 3.8v and set in bios to 3.6v. i saw .1v fluctuation but thats about standard with the crappy circuitry

-----------
@ EMC2
-----------

WOW! its sad that the vtt sucks that bad. thats just amazing how much it drops in OCCT! hmmmm and a cure would be------ :slapass:

Great findings, actually, not so great findings.

A question, Vtt where is it connected to on the CPU and RAM (a pin number would be nice)? I want to read on it.

I'm guessing it is pin 1 of the DIMM, otherwise known as Vref (not to be confused with your Vref indicated in your measuring points). Am I correct?
Well, if someone could point me in the right direction...

What I don't understand, if it is 'just a reference' for the CPU/RAM to 'know' what is a logic 0 or 1, what is loading it so heavily? What is the expected current draw from the VTT source?

You can pick either leg circled in yellow, for VTT reading.

Quote:

---

Originally Posted by Malves

You can pick either leg circled in yellow, for VTT reading.

---

pic doesnt show up....no access permission

did some intense re-tunes the last few days running on the 623-2 bios.

I'm superpi, memtest, 3dmark05, 3dmark03, 3dmark2001, aquamark3d, S&M (long), OCCT and Prime95 stable --- at stock 200*13 1.5.2.2.5 (all others tight)-- I've also got an alternate set of loose timings which also pass all benches -- 200*13 2.3.3.10 (all others loose)

I'm 240*12 1.5.2.2.5 stable for all of the above -- except OCCT/Prime95 and S&M (cpu ok, mem fails 2nd test)
Same for 240*12 2.3.3.10, 250*11 1.5.2.2.5 and 250*11 2.3.3.10

I tried many things to get prime95 stable at the OC's to no avail!!

After seeing your OCCT scoped vtt - I'll bet prime95 does the same thing!!

If you experts get/have any mod's please pm me, I will try a mod.

Chad

p.s.

you can see my exact settings for these benches here

http://www.dfi-street.com/forum/show...&postcount=116

p.s.s sorry about the dumb comment on the pm me ... just willing to try something, even if it's not ready to publish.... like everyone else here ...

lol we all want a mod. havent you read the thread? if they pm you one, we cant see it :p

Good catch bunky.

Not great news, but a good reason why higher than 3.4V makes not improvemnt beyond 250Mhz on my 2 DFI SLI boards with either UTT-CH or UTT-BH or OLD-BH5. If vtt is falling apart above a certain point, sheesh.

Esp points out why the New-BH (ValueVX) were having Test #8 errors with 1T _and_ bankinterleave enabled.

Simple experiment to generate further data. Anyone with a OCZ booster...which drives Vtt/Vref as well as Vdimm and might take up the slack and provide stable Vtt. Checking the voltages/stability with the booster installed would give as more data points.

Guess we'll have to collaberate on a DFI NF4 Vtt project, like we did for the NF7-S last year.

THINKING CAPS ON :cool:

I have a new booster, but no o-scope - willing to mail it off to emc2 for this test

p.s. I'll go tear down my beautiful overclock and stick the booster in to see if I can get a ottc pass at 250*11 -- report back later

Chad

VTT does not need a mod. It needs a fix. It tracks vdimm, but really bad.

Picture in my previous post above has been fixed.

Thats what I mean Malves. The booster will over-ride the Vtt from the NF4 regulator as it is also a regulator. So hopefully it will provide enough extra current to keep Vtt stable. The result should be a high stable overclock.

So Vdimm stays rock solid through all this, just Vtt gets the shaft?

EDIT: Never mind about below, found it in the AMD datasheet - I should've looked before asking ;-)

Malves, I would like to know what VTT is connecting to, i.e. to which pin on the processor, to which pin on the DIMM slot. Got the pin, I got the details to look things up on.

From a technical point of view I want to read about how and where VTT is used, and looking at datasheets without knowing that, well, is like looking for a needle in a haystack.

Thanks

what about changing out the vtt regulation ic with a better quality one? or adding caps to help stabilize the signal.

EMC said it would take a few farads of capacitence to fix the fluxuations so extra capacitors are a no.