NF7 rev2 VTT mod need it so bad...

[uwackme]

Im doing tests now....

Weird results, at full Gig results on main board worse than before mod.

11,3,3,2.5 still error even at 220Mhz.

BUT...WTF...the 2x 256M sticks, 11,2,2,cas2 cpuint enb at...

drumroll... 245Mhzx10!!!!!! 3.3Vdimm 1.65Vtt... I assume this must mean BH5 or BH6 worst case.

The Vdd right now is 1.65 also, and going higher dont help, and lower will error. Going up to 1.95Vdd doesnt hurt either.

My poor 512M stick will need review.

The ODD thing is on the other NF7-S the full 1Gig was happy at 230Mhz 11,3,3,2.5 so there must be a diff between both these rev2.0 NF7-S's.

Ill keep you posted.

[uwackme]

Well, 67 passes through Memtest86, no errors.

Im typing on the 512M stick at 230Mhzx10 11,3,3,2.5 cpuenb 3.3Vdimm 1.65Vtt on the backup system, also without error.

Ill move this alone into main system, and see where it goes.

Wonder if CH5 and BH5 cant get along well in same system.

[uwackme]

Swap things, the 512M stick is in the main system, 3.3Vdimm, 1.65Vtt, 1.80Vdd, running 230Mhz 11,2,2,cas2.0 cpuint enb.

When all three sticks are installed however, instability/errors ensue.

That is all with bios D10.... now Ill begin testing D23 3Dfire.

Also, now typing on 2x 256M sticks in backup system, 230Mhz 11,2,2,cas2.0 3.3Vdimm, 1.65Vtt, 1.65Vdd.

Maybe mix+match BH5 and CH5 causes some weirdness?
Also, the 512M stick gets alot hotter in operation...16 chips instead of just 8 on a 256M stick.

Beginning to think these heatspredders are insulators instead

The Vtt mod has seriously made improvement in how high an FSB good ram timings can go to.

[tictac]

have u try higher vdimm?

Vdimm @ 3.4v Vbt @ 1.7V?

[uwackme]

Weirder and weirder dept.

On D23 testing of main board...the same 2x 256M sticks....same settings in all things.

230Mhzx10 11,2,2,cas2.0 cpuint enb went 100 passes error free.

235Mhz...errors on all timings. 11,2,2,2 11,3,2,2 11,3,2,2.5 11,3,3,2 11,3,3,2.5

240Mhz errors on all timings. This was dimm#1+3 or dimm#2+3

With the single 512M stick, all same settings.......

240Mhz would not even boot...no post...this is single channel, possible data point there. The 2x 256 would post and run @240Mhz.

235x10 11,3,3,2.5 and 11,3,2,2 worked flawlessly.

230Mhzx10 11,2,2,2 was flawless.

Ok, so something in D23 (and all newer BIOS's code) is different from D10, NVidia/Abit put some changes in that limited our topend regardless of circumstances.

Now for totally WACKO... back to D10, work on the 1Gig issue. Before the mod it was working fine at 220, after the mod...oh oh it was erroring. So I started testing each variable one at a time.

Vdd...1.6-1.95 no effect, still errors, return to 1.8Vdd that was used in all previous tests.

VDimm...3.0-3.35Vdimm....no affect or some worsening. Return to 3.3Vdimm

Vtt...hmm...return to lowest setting, like the bad condition from premod...BINGO! works again, no errors. With pot installed and set to min, new Vtt is still higher at Vtt= 1.545Vtt. This setting yeilds no errors...220Mhzx10 or 230Mhzx10 with 1Gig in place. Though at 230, the rams get hot in the 3 DIMM arrangement, out of the case, I need to rig a fan on them. The word OUCH comes to mind touching them. The world needs a better DDR heatspredder....mounted with ArticCeramique with FINS...instead of those stupid LEDs on the new Corsairs ;-)

TicTac...no for some reason none of these boards does 3.4V comfortably, regardless of 3.3V rail setting...tried 3.6-3.7-3.8 rail, but anything over 3.35V caused weirdness...pot was no longer effective, etc. I think a overvoltage element must be in the VDimm regulator circuit.

Going by the rams, 245x10 11,2,2,2 cpuint enb is plenty fast @3.3V....so I dont think I need 3.4V.

Now Im going to put the 9800pro in on the bench and see if it's presence alters the results at all. Then Ill be off to reassembly and Windowsland.

I guess I should peek under the 512M's heatspredder...is it ch-5 or not. 11,2,2,2 cpuint enb 3.3V @ 233Mhzx10 for a couple hours no errors....hmmm

So either 2x256 or 1x512 are both happiest with 1.65Vtt 3.3Vdimm but together...they want 3.3Vdimm and 1.545Vtt... weird....not 1.55Vtt...but 1.545Vtt

[tictac]

seem the board overheated too much after the vbt mod

my friends fry his vbt n channel mosfet... he didnt put heatsink on those mosfet....

do u think after we did the vbt mod it also effect other voltage than vbt.... ?

[Dirty_Punk]

i have used VDDR up to 3.6-3.65v so VTT up to 1.8-1.825v and no problem with my mod yet.... howether it's very hight voltages and only for test use (max try 8h of stress) so in this case connect direct +3.3v is better and easyer

[tictac]

Originally posted by Dirty_Punk
i have measured on the DIMM socket the VTT, not on the mosfet... so on the dimm socket i measured over 1.7v with over 3.4v of VDDR, but this with power off motherboard (also VDDR can be measured with power off motherboard with this mod, with normal mod not because you measure always aroud 2.85v)

howether with +3.3v directly connected to dimm pin you get a much more stable VDIMM
with both at 3.2v i get about 3-5mhz more with this mod VS normal VDDR mod

Pin 1 is Vref:
Pin 2 is Vtt :
Pin 7 is vdimm :
Pin 15 is Vdimm :

So do u mean Vref is the real Vtt voltage on the Rams

i'm getting lost here...

[Dirty_Punk]

i have never measuered voltages on mosfet or else, always used the Dimm slot because only in the slot you get real voltages...

DDR pin map

[tictac]

if the pin 1 is the real vtt

i think u r probably right

coz i have measure the pin 1 it shown: 1.95-1.98v(half of my vdimm reading)

i/o volt = 3.95v
vdimm = 3.95v (Vdimm = V i/o mod)

and i also measure from the vbt mosfet it shown 1.63V(Vbt=Vdd mod)

vdd= 1.63v

Edited: Now what mod is this for?

[Dirty_Punk]

probably Vbt mosfet doesn't get its voltage from VDDR mosfet and divide by 2 (and this explains why VTT isn't 0.5*VDDR when we have a standard VDDR mod) but gets it from another point

[tictac]

ok now there are 4 voltage to check:

V ref = half od Vdimm
V dd = Vdimm
V tt = half of Vdimm
V ih = half of vdimm

Vref is correct it shown half of vdimm voltage (after Vdimm=V I/O mod)

Have u measure the others volt.... Vtt and Vih?

Vref is not Vtt

Vtt is not Vref

as shown on the pdf file u give there

[tictac]

here... i just check my dead board voltage reading all over again

1st row is Vdimm mod Dirty Punk Ways(Take note that my winbond ic leg no 11-20 is gone)

2nd row Vdimm mod Dirty_Punk Ways + Vbt=Vdd mod

The Vref track the vdimm correctly after dirty punk vdimm mod
But i cant check my vtt as my winbond ic messed up

after i add vbt=vdd mod
the pin 2 of dimm show 1.7V same reading as my Vbt mosfet reading and Vdd reading
I measure pin 2 of winbond ic (Vbt sense line it show 1.6V)
Even after my vdimm @ 4V the vtt didnt track the half vdimm voltage correctly...

cheers

[uwackme]

Well, there are 4 voltages on a DDR, Vdd-Vddq-Vref-Vtt

On the DDR dimms on most motherboards... Vdd and Vddq are simply run off the same source and are equal. So all Vdd and Vddq pins are soldered to the same PCBoard power plane.

On my NF7-S's the Vtt and Vref are very definitely connected together and are the same. And the Vtt mosfet gets hotter with Vdimm mod's due to the fact that it runs the terminating resistors, NOT because of the DDR chips themselves and Vref supply.

Think about it, the purpose of Vtt is to "pull-middle" the data lines so that the data bus sits IDLE in a Vtt state...is not allowed to drift to 0volts or 3.3Volts. Reason being, so that which ever direction the data pin needs to go...IDLE to a "1" or IDLE to a "0" the line only has to move from 1.65V to 0V or 1.65V to 3.3V.

The instantaneous current draw of the data bus going from 0V to 3.3V at DDR speeds would be astronomical. As it is, going from 0V-1.3V-2.6V leads to a certain amount of heat generation from the stock level of current draw through the mosfet. Increasing Vdimm you now go 0V-1.65V-3.3V increasing the difference from 1.3V to 1.65V which yeilds a current increase of ~18%... so the mosfet gets ALOT warmer than stock settings.

The mention on AMDMB about one GLOWING RED, comes from a fight between voltage regulator outputs.... where they both try and supply to Vxxx and if ONE regulator is particularly WEAK, the little mosfet has to not only supply Vtt/ref to the rams but ALSO attempt to DRAG the Vdd regulator's output UP in voltage, all the while the Vdd regulator is trying to "REGULATE" and keep Vdd lower, so they fight and one of them burns out.

Remember, 50% of the mod's you find on the boards are developed by PSEUDO engineering, not real engineering. So beware just slapping wires and interconnecting circuits. Things are usually design a certain way for good reason. Sometimes the reason is cost, and thats why wedont have a seperate Vtt control in the bios, etc; or why Vdd and Vddq of the DDR chips are both tied together in the design as Vdimm.

Remember, DDR is DDR... the same volt mod's and setup on the 9700/9800pro video cards holds true on the motherboards. But the video cards are going WAY WAY out of the way to get maximum performance. The motherboard designers figure, hell the rams are slow anyway, why bust our humps and add cost for negligible increases in speed/timing.

I'd personally LOVE to see someone slap 16 2.8ns Samsung videoDDR's onto some DIMMs and let us get REAL BUSY, FK 5ns BH5's

[hyem]

Tictac,.... it's me TrueEnd, I already make 3 in one mod for my board, and it's Rock Solid at FSB 240 at 2-2-2-11,... Great Job man,.. I use VBT mod,... and there's no problem yet,... it's the 3rd day since I use that mod,... I use VDD = VBT = 1.7V , VIO = VDIMM = 3.4V
I use the 3rd methode,.... I think it's no problem for my BH-6

[tictac]

Originally posted by hyem
Tictac,.... it's me TrueEnd, I already make 3 in one mod for my board, and it's Rock Solid at FSB 240 at 2-2-2-11,... Great Job man,.. I use VBT mod,... and there's no problem yet,... it's the 3rd day since I use that mod,... I use VDD = VBT = 1.7V , VIO = VDIMM = 3.4V
I use the 3rd methode,.... I think it's no problem for my BH-6
kalo menang Special Price lah,..

wah.... nice :thumbsup:


p/s : ntar kalau menang nf7s yg baru kasi special price dong.... kikiki

[tictac]

I dont recomend new user to try this new ways of Vtt/Vbt mod

but if u have done Vbt=Vdd mod .... and it is SAFE
here is better way to track back the vtt/vbt half of Vdimm voltage automatic without have to set vdd half of vdimm voltage anymore

The Vref track half of Vdimm voltage correctly even after Vdimm mod...
The only problem is the Vbt/Vtt Voltage didnt follow it ...

Vref = Always Half of Vdimm Voltage even after a vdimm mod

By this: 4th Method

Vbt/Vtt Voltage = Vref Voltage
( Always Half Of Vdimm Voltage & No need to adjust it anymore... )

For those who have tested my 3rd method Vdd=Vbt/Vtt mod SAFE.... i can recomend u this new way...


Follow the White Line

[hyem]

Yup...that's great Idea tictac... but I don't know if it's work... for your Vref... (I'll try it for You...)


p/s : I'll use u'r 4th way... for my next NF7-S

[tictac]

dont worry hyem.... i will help u as much as i can...

250MHz here we come... hehehe

[uwackme]

Weird, when I was prepping for the mod, I checked Vtt and Vref (pin1 ddr sockets) and they BOTH were the same and DID NOT track Vdimm correctly after the mod. I used the Fluke "beeper" and they were shorted together, no question of it.

Wonder if ABit made a design change in the board at some point that we dont readily see on examination?

I'm testing now with the heat spredder OFF my 512M stick....which BTW is winbond CH-5 0338WI date code....all 16 chips. The SPD on these is 8,3,3,2.0 @ 200 and 9,4,4,2.5 @217

What do others have for SPD code on HyperX3500 sticks?
You find it by using CPUZ v1.21 and go to "ABOUT" and hit "memory SPD" it will put a spd.txt on your desktop. Open and read the report. My 2x 256 have same SPD but they do 11,2,2,2 at 250, so are they BH5 or not, dunno. Guess I gotta look. It's a getting the heatspredders off.

[Bugs]

My 2x256 KHX 3200 (with BH-5) SPD:

Frequency (MHz) 200
CAS# 2.0
RAS# to CAS# delay 2
RAS# Precharge 2
TRAS# 6

Stable at 243MHz - 11-2-2-2 - Vdimm: 3,2V - Vtt: 1,49V

[enduracell]

Vbt when follow the Vdimm (at half of it) certainly gives better ocability in ram. This was the case when I moved from NF7 to AN7 (AN7 gives 3.34Vdimm/1.67Vbt both from the HW monitor form BIOS) my ram performed better about 7FSB more at high Volts comparing to NF7, but I wasn't able to test it with CPUinter enable because of the bug to do a direct comparison. Unfortunately never measured the Vbt (or any other value) on directly on the dimms so I don't know if indeed Vbt follows on the dimm but not on the fet on my mobo, but I will check it out.
uwackme

So either 2x256 or 1x512 are both happiest with 1.65Vtt 3.3Vdimm but together...they want 3.3Vdimm and 1.545Vtt... weird....not 1.55Vtt...but 1.545Vtt

This is very strange!! I will check it out on both mobos to see if there is a difference.
Cheers, really great work guys, I can't wait go home and do my measurements and stuff too

[tictac]

measure the vref from the back of mobo it move zig zag way

pin 1 on right pin 2 on left
1,3,5,7,9,11..... 1 line
2,4,6,8,10..... 1 line

[uwackme]

Just as a datapoint... I was doing a freinds ASUS A7N8Xdeluxe. It's Vtt tracks the Vdimm regardless of mod.

[hyem]

I've try da 4th way,... It's safe,.... and it's work fine,... (hope it's better than other methode).