DFI NF4 Ultra-D vdimm mod: come inside.

[Malves]

560pF if I remember right. Should give you an extra 5mhz in o/c. Too much hassle for so little gain.

[sluggo]

What is the Vtt regulator using for gate drive? If it's 3.3V (with the jumper in standard position) and 3.3 gets pulled down momentarily or gets noisy, it could drop below Vin briefly. The Vdimm reg might ignore it - this one might not. Just a maybe.

sluggo

[PORTUGAL]

this problem possible cold be caused by autooscilation in the integrated circuit,wath about trying adicional eletrolitic capacitors in the input and the output from the regulator.

[dippyskoodlez]

560pF if I remember right. Should give you an extra 5mhz in o/c. Too much hassle for so little gain.

something like this, or am I way off?

http://www.mouser.com/index.cfm?hand..._pcodeid=58114

[persivore]

I'm a bit busy atm to work on a fix for Vtt atm. I should be able to get it done by the end of next week. If anyone wants to work on it instead of me doing it, feel free to do so .
What I have in mind is to use an opamp between the reference voltage pin on the Vtt regulator and the voltage divider which is used to make Vtt follow Vdimm, and use the opamp to provide feedback to correct the voltage. This will involve lifting one of the pins of the Vtt regulator off the board, however.

[Malves]

This will involve lifting one of the pins of the Vtt regulator off the board, however.

Piece of cake.

something like this, or am I way off?

http://www.mouser.com/index.cfm?hand..._pcodeid=58114

Yup, but size 0603. According to EMC2.

[craig588]

Lifting a leg is easy, lifting the whole thing would be hard.


Another thing, that seems like a very small capacitance, I have caps rated for thousands of microfarads in a drawer right next to me. Why can't I just use them? I thought I saw multiple farads somewhere which would have been pointless to try to reach, but 560 picofarads is really easy to reach.

[tictac]

this one?
http://www.mouser.com/index.cfm?hand..._pcodeid=58113

[PORTUGAL]

can someone post an image with the exat points to soldering the 560pf capacitor?

[OzSnoal]

Normally pf caps are used to filter noise.

Maybe this has been posted before, nice software I came across:
http://www.kemet.com/page/spicesoft

It shows a graph of impedance vs frequency for your selected capacitor.

Select your Chip Type/dielectric for your capacitor and it will show you a graph of impedance vs frequency, a 560pf COG 0603 has its lowest impedance at approx 166MHz. It's a nice toy if you want to pick a cap for a specific frequency. And an 0805 of the same part shows the lowest impedance at 190MHz.

[EMC2]

--- stealth17 ---
Go read the 2nd page of my post on the Zippy/PCPnC fix thread for why you see the fluctuations you do with those settings

--- OzSnoal ---
Vtt and Vref are two different signals, depending on where you look at them.

Vref - used for what it's name implies, a reference signal to determine the mid-point of the memory signals. It's an input to the Vtt regulator and is shown in the scope pics.
Vtt - this is the signal termination voltage used to supply the termination networks on the signal lines to/from the memory. If you look at your MB, you'll see nice long rows of termination networks. The loading occurs as a result of the current through these termination networks to the memory signals.

If you want to read information on what it's all about, go look at design guides from memory manufacturer's...

For where to look at the signals in general, they are shown in this post of mine: http://xtremesystems.org/forums/show...1&postcount=10

--- uwackme ---
bunky? Yes peaches Vdimm stays rock stable... look at the scope pics again, it's all there

--- Malves (and all) ---
The 560pF caps are for the main frequency components of the noise that is prevalent on Vtt and Vmem - they are NOT for the problem shown above in the scope pics and being discussed. And just one won't help ...

--- sluggo ---
The regulator is using Vmem for both it's input source and it's gate drive... the point instrumented in the pics for all 3 signals is at the Vtt regulator. It is not an input signal droop issue.

--- Portugal ---
One possibility... and it may be being set off by transients... work in progress

--- persivore ---
That is actually one of the possibilities I mentioned to Malves... but to be done right need a small PCB or protoboard. All connections can be made at the Vtt regulator btw

If it is a pure load issue, you could possibly parallel another regulator... would require a small HS and good soldering...
Or another alternative is replacing the existing regulator with a different circuit

--- everyone ---
Patience

Peace

ps. Oz - set it to NPO dielectric and 25V rating... it's 300Mhz centered for an 0603... an 0805 will have a lower frequency center than an 0603 due to the packages higher inductance And I use AVX caps... hate kemet

[stealth17]

--- stealth17 ---
Go read the 2nd page of my post on the Zippy/PCPnC fix thread for why you see the fluctuations you do with those settings

what thread?

http://www.xtremesystems.org/forums/...earchid=662302

[EMC2]

The sticky... Zippy and 850SSI

[stealth17]

The sticky... Zippy and 850SSI

it didnt help

vtt is still everywhere

[sluggo]

Okay, so assuming this part (the 9173A) is used successfully on other motherboards, then this is either a bad lot of parts or a bad implementation/layout on the DFI boards. If it's a bad lot of parts then mod results may be unpredictable. If it's a bad layout (my choice), it's usually because they've got the bypass caps too far from the part. Wish I could help but I can''t put my board on the bench right now

sluggo

*edit* Sorry - I'm jumping to conclusions here. Obviously, there's many more things that it COULD be, and I'm just shooting from the hip.

[stealth17]

i know this is a tiny bit off topic but not really.



look at the position of the jumpers i put in that pic. they are the big black squares. notice there isnt one on row B. here is my thinking:

i was doing some testing with no dimms and power ac off, then switch it on and measure the dimm voltage. with rob B jumper on 1->2 i get 2.4v something. with it in 2->3 i get 2.75v, higher than in the other side. if the vdimm controll is messed up, its obvious you will have voltage to the dimms as both pins (1,3) feed from the 5vsb rail. so i was thinking, what if you dont put a jumper there at all and connect the regular 5v rail to B2 (the blue block)? it wouldnt get voltage untill the mobo is acutally on, thus bypass caps are not charged.

i tried the cold boot find steps with no jumper on the B block and i didnt get any voltage to the dimms with the ac power on.

what do you guys think?

[OSKAR_WU]

memtest #8 loop , fx-57 , 300 * 9 , gskill 512mb tccd * 2 , 2.87V mem , 1T

Scope Channel 1 : Pin5 in RT9173A VTT regulator
Scope Channel 2 : VTT rail behind DRAM slot
Scope Bandwidth 20mhz

http://oskarwu.myweb.hinet.net/1.mpg
http://oskarwu.myweb.hinet.net/TEK00000.gif
http://oskarwu.myweb.hinet.net/TEK00001.gif
http://oskarwu.myweb.hinet.net/TEK00002.gif
http://oskarwu.myweb.hinet.net/TEK00003.gif

[Ubermann]

God just entered the building.

[timpanogos]

Oskar, I've always been able to tweak memtest #5 and #8 to pass ok ... it's OCCT and Prime95 that I could never get to pass.

Please test/provide shots with OCCT running so we are looking at the same thing as emc2.

I'd also be very interested in seeing prime95 shots.

try right off with like 250*11 @2-2-2-8 on some vx type memory for your run, and capture the time frame right were it fails.

Chad

p.s.
I've been working hard to make the dfi oc database, I was stuck at stock (could not prime) until I put the ocz booster in - you can look here for detailed settings/benchmark screens.

going for a Johnnylee 24 hour run to to be able to post here ....

http://www.dfi-street.com/forum/show...782#post152782

[EMC2]

Hi Oskar

Winnie @ 9x295
Memory TCCD on BP0815 PCB, 512MB per stick,
running @ 295Mhz, 2.5-3-6-3-9-12-2-2-2-3-4708, 1T, 2.91V
BIOS 510-3fixed, DDS=3 ,DS=7 ,Tref=4708, Interleave ON, 256 Idle cycle count

Scope Channel 1 : Pin5, RT9173A VTT regulator
Scope Channel 2 : Pin3, RT9173A VTT regulator
Scope Bandwidth Limit set to 20mhz

Here's MemTest8 with ONE 512MB DIMM - Vtt as it should be
http://img301.imageshack.us/img301/3...by512mb8jw.png

Here's MemTest8 with 2x512MB DIMM, TCCD, BP0815 PCB - considerable change due to load (>6x) I see higher than you... RT9173A lot difs or CPU?
http://img301.imageshack.us/img301/4...by512mb8pt.png

Here's what you get from a dynamic load change in windows, Windows MemTest, much worse:
http://img301.imageshack.us/img301/2...eststop0hi.png

Here's one from during OCCT, again, bad stuff:
http://img301.imageshack.us/img301/2875/occtstop1ew.png

I have more pics from the session (including equipment) and AVI videos of the captures (low quality, but hey). However, I need either 150MB of web space or an FTP site at DFI to upload them to if you want to see them.

Peace

[EMC2]

--- Stealth ---
You only have voltage on the DIMMs with AC ON and MB OFF until you boot up one time. After that, when you shut down (MB OFF) the power is removed from the DIMM circuits. The moral is, don't change DIMMs with the Vmem power LED on Oh... and I think this would be better discussed in the PCPnC/Zippy Fix thread

Back to the topic at hand...

[EMC2]

Ooooooops... forgot another case/data point from Friday...

Same setup as above, but during windows boot-up...
W2K, just after initial screen with scrolling bar at bottom, when the video gets switched over to the desktop resolution (going from VGA to 1280x1024) and video memory loaded (video is momentarily blanked during the switch) you get something a tad worse than the OCCT and Memtest screens above.

[OSKAR_WU]

The worst case will be all 0/all 1/transitient from all 0/1 to all 1/0 ... The minimal RTT total equivalent resistance in dual channel mode will be around 0.7 OHM ... If the VTT is set as 1.45V , maximum IL in both sink/source mode will be around 2A ... Still within the range of the spec of RT9173A ... RT9173A can control the VTT output within 2% of variation if the IL from -3A to 3A or so ... What you measured should not be a typical result from the design spec ...

[EMC2]

I agree... based on the spec for the RT9173A it should not change that much if max load current is 2A. If it is 2A then... bad batch of RT9173A's? Or maybe because VCtrl = Vin instead of VCtrl > Vin. The spec'd output variation is with Vctrl=3.3V, Vin=2.5V.

Note: Not a temp issue either... 120mm fan over DIMMs and covering Vtt/Vmem regulator area... Vtt regulator is cool to the touch.

[flexy]

hey, i just found this thread. Whow...

I am having the same issue that i can run memtest #5 and #8 for hours.....but i cr*p out with OCCT in Windows in 10 seconds at a certain, given dimm.

I used to run my TCCD at 2.9V....and (dont ask me why)....there is an issue with either )the board )PSU )memory ...that my sticks now need MORE and mroe voltage otherwise i cant be stable in OCCT at all.

2.9V is now a total no go so is 3.0V ...i need to push my TCCD (week 37) with 3.1V so i can get stable in OCCT ! (yes, TCCD !!!)

It looks you guys are sitting on the source of MOST instability problems in windows/OCCT....looking at that readings makes me shudder !

I hope you can come up with a solution ! Great reads and insights here !!!