DFI NF4 Ultra-D vdimm mod: come inside.

[EMC2]

"might go off a bit" is putting it mildly

--- Chad --- Vtt is used by the memory and the CPU as a reference signal to tell when a signal is high or low. It's also used in the termination networks for the signals on the MB to...errrmmm...in "gumbie" as you put it... to make the signals better. What I was telling Malves is that under certain conditions the Vtt fluctuates a lot causing errors. What he was saying is he doesn't have a CPU to put in the MB to check out what I told him.

--- Malves --- Vtt is rock solid stable until certain burst data patterns are used... then it fluctuates quite a bit... enough to cause ~25Mhz OC difference It appears to be a load issue, as it's relatively long term fluctuations (you can see them with a DMM). You can run T5 (pretty much T1->T7) and it's rock stable... T8 and it's all over the place during the course of the test. You can pretty much predict when errors will pop up by monitoring where Vtt is. Same thing happens when in Windows... and when running most benchies. Note that both the input rail and the reference to the VTT regulator remain rock solid... it's only the VTT output of the regulator that fluctuates.

Peace

[Malves]

Interesting. So, maybe a VTT mod could help?

Also, I haven't mentioned before, but I knew the LM328 just uses 2nd stage for both 3.3v and 5v. The switch between both should be via jumper, not the stages. So, with that said, I got vdimm up to 4.5v via trimmer, using JP17 @ 2-3.

Those that want more control over their vdimm and those that are brave enough for 4+vdimm, this is the mod.

[stealth17]

wow cool stuff

so malves, think you could make up a vtt mod?

so all we do is put the jp-17 in and we can get higher than 4v with the trimmer in the same place right?

[G H Z]

Nope, haven't seen that board yet, bro. I don't like vio mods, sorry.

Why don't you like Vio mods?

-sorry for more OT

[EMC2]

Malves - ygpm

[Malves]

Thanx for the heads up, bro.

Well, EMC2 mentioned in his PM, it looks like under load the 1st stage of LM358 limits the 2nd stage to ~4.2v. So, under load, vdimm @ 4.5v may not be possible.
Unfortunately, I don't have the right equipment and not enough of it, too, to test all of this, but I guess EMC2 has been there.

Vio mods don't give you any control over vdimm, that's why.

[timpanogos]

It's info like this that has me now watching this site/forum closely ... thanks all great info

[Waus-mod]

Why a vdimm mod if you can use the bigtoe mod? the only thing is to fake the 5v jumper. no soldering. Why should i still prefer this mod?

[timpanogos]

I believe the "bigtoe mod" is to slide the block connector down one row, put it on the 3.3v side and single jumper the exposed top pins from the center to 5V side.

This is the jumper that allows the bios to set voltages over 3.2. However, you are now on teh 3.3 volt rail, and being able to set actual voltages above 3.2 will be based on if that voltage is available on the 3.3v rail. With supplies like the OCZ the supply allows you to tweak this, on others, you'll need to vmod.

However, Malves has warned me that setting the bios higher than what is available on the rail could damage the mobo. So you might want to be careful with the BigToe mod, making sure your rail is over what the bios is going to try and put on your memory. Also be aware that my bios/mobo/etc. seems to overvolt the mem by about 0.08, so I assume you best have your rail overvolted by at least that same amount from what you plan on setting in your bios

[Malves]

However, Malves has warned me that setting the bios higher than what is available on the rail could damage the mobo.

I remember reading something about this, but I can't remember where. So, don't take my word for it.

[stealth17]

the bios will only go up to 3.3v and then you just adjust the rail and nothing more in the bios. there isnt an option for a higher voltage so unless you 3.3v rail is less than 3.3v then there will not be any problems

[Waus-mod]

It really works great on my board. setting my ocz to 3.8v and i gain a stable 3.7vdimm. Works truly great!

are there more mods for the nf4 ultra-d?

[Malves]

It really works great on my board. setting my ocz to 3.8v and i gain a stable 3.7vdimm. Works truly great!

are there more mods for the nf4 ultra-d?

Which one? Bigotes'? There's vcore mod, but i don't think anyone will need it.

[Waus-mod]

And the vtt?

[stealth17]

wow vtt does suck. specs in my sig, i was running 275*8 at 3.6v (bios) and the actualy vdimm was 3.7v the vtt was all over the place. it jumped from 1.83v up to 1.87v! ouch! it was anything but close to stable but the 3.7v was satble although i did see it jump to 3.71v for a second a time or two.

anyway, can we fix the vtt atleast somehow?

[EMC2]

I've recorded fluctuations of ~80mV when nominal VTT was 1.455... most of it in the negative (lower V) direction... about 3x what it should be.

Here's some pics from an o'scope... sorry for the graininess... frames from a video capture with a digi-cam These are taking running T8 of memtest, the large steps are when the test pattern changes between mostly 0 patterns to mostly 1 patterns:







Peace

[Malves]

EMC2, do you think replacing/adding the "VTT caps" would help with the fluctuations?

[timpanogos]

quick explination of vio please ... I know you said you did not like/use it, just wondering what it is ..

Thanks

Chad

[Malves]

vio = 3.3v rail of your PSU

[EMC2]

Malves... slightly, for helping some with the relatively shorter term fluctuations like the small bumps in the 3rd scope pic that occur before the step change... but not for the longer term changes that start/end at the step points. Note the scale on those pics is 5ms/div... and the larger step changes last for upwards of seconds (as long as the offending pattern is being burst). Basically the VTT is drooping due to the current demand and the internal FETs RdsOn. Can think of 3 fixes... 1 that positively would work... 1 that has a 90%+ chance... and one that is 50/50 (which I'm not willing to sacrifice my MB for).

Peace

[timpanogos]

sounds like I did the sio mod that you do not care for. I am interesting in going ahead and doing the op-amp mod that was shown as the topic of this thread. However, I'm still confused, without the sio mod, how can the opamp mod help, on my vx that want 3.3+ volts?

I would be interested in your "droop" mod also .. any info is appreciated ...

Chad

p.s. can you elaborate on why you do not like the sio mod

[EMC2]

Just a little update on the Vtt side of things... was looking at it and a few other things with better equipment and got a nice capture of a Vtt step. Did a rough calc on the capacitance needed for this... can't get there from here, it comes out several farads worth However, I did find that drive strength has an effect on the magnitude of the step, gained some Mhz, along with enough info to base fixes on Now to find time and parts

Capture of Vtt step on DFI nF4 - this one occured during transition from an FFFFFFFF to 00000000 based modulo:
[IMG][/IMG]

Peace

[JNav89GT]

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

Maybe I speak in medicalease so I can feel smart.

[stealth17]

definatly a big problem. hope you can help us EMC2!

[craig588]

How much does a fancy scope like that run?

Hmm, back on topic, I would love to perform any mods required to fix this problem. Unfortunately the only thing I know how to do is look up chip specs and reduce resistance on the sense leg, which is probably where pretty much everyone else here also hits there limit too; we are all betting on you EMC.