VERY EASY GF6800/le Vmem mod

[SikaRippa]

I found an original (anybody seen before ???) way to make a very easily a Vmem modification for GF6800-cards. Tested with Gainward 6800le, but should be applicable also many (all?) cards using the same PCB (so not ultra/GT..)

Original thread ("Memory voltage modification" -chapter):

http://www.xtremesystems.org/forums/...t=59391&page=1

.. but this figure is all you should need




Comments, please !

[$a1Ty]

very nice, i'll take a look at my sparkle 6800nu and see if it matches up, wouldnt mind having a bit extra voltage to my ram

[ZeroHero]

Very nice
Shal perhaps try someday that.

Do you think you could figure out how to do a pencil vgpu mod to teh 6800LE?
Would be happy if you could do that.

[SikaRippa]

Very nice
Shal perhaps try someday that.

Do you think you could figure out how to do a pencil vgpu mod to teh 6800LE?
Would be happy if you could do that.

In fact that mod has already posted in the 'all vmod for all 6800' post #1.
The figure is not so informative, since under that green dot is TP12 (=hole).
And using this hole you (somebody) can make almost-like-a-pencil-mod by connecting potentiometer between TP12 and ground. I have measured a resistance of ~545 ohms between that node and GND, which might be a good starting point when calculating value for variable resistor. The voltage in that node is allways 0.6 V (=node tracks internal voltage reference, see the datasheet for details). And if having bios for 1.1 V the resistance between Vcore and TP12 should be then 455 ohms.

BUT !! There is a special circuitry, which changes these resistance values using transistors. So, actual resistances seen in this node might be different when card is running ((DO NOT EVEN THINK TO MEASURE RESISTANCES WHILE CARD IS RUNNING)).

=> I cannot say what value should be connected to get certain Vcore-voltage. So, if you want to test use similar connection (resistor + potentiometer) I posted above using HIGH resistance values at first..

[ZeroHero]

In fact that mod has already posted in the 'all vmod for all 6800' post #1.
The figure is not so informative, since under that green dot is TP12 (=hole).
And using this hole you (somebody) can make almost-like-a-pencil-mod by connecting potentiometer between TP12 and ground. I have measured a resistance of ~545 ohms between that node and GND, which might be a good starting point when calculating value for variable resistor. The voltage in that node is allways 0.6 V (=node tracks internal voltage reference, see the datasheet for details). And if having bios for 1.1 V the resistance between Vcore and TP12 should be then 455 ohms.

BUT !! There is a special circuitry, which changes these resistance values using transistors. So, actual resistances seen in this node might be different when card is running ((DO NOT EVEN THINK TO MEASURE RESISTANCES WHILE CARD IS RUNNING)).

=> I cannot say what value should be connected to get certain Vcore-voltage. So, if you want to test use similar connection (resistor + potentiometer) I posted above using HIGH resistance values at first..

Thanks, first I must say that electrical isn't my best area, but this semms posibel to do.
What voltage shal i set from bios?
SO can I buy a high value resistor, and then with the potentiometer, lower the reistance for a higher volt?
So how many ohms resistor do you recomend me to get?
Will be doing this when i get the items and have time.

[SikaRippa]

Thanks, first I must say that electrical isn't my best area, but this semms posibel to do.
What voltage shal i set from bios?
SO can I buy a high value resistor, and then with the potentiometer, lower the reistance for a higher volt?
So how many ohms resistor do you recomend me to get?
Will be doing this when i get the items and have time.

You can use what bios you want: because correct resistor value needed is not known you should use potentiometer in any case.

You can use idea you presented, HIGH value potentiometer between those points (it is again recommended to put a fixed resistance in series) and lower the value VERY carefully AND MEASURE vcore frequently. I do not want to give any values because of i do not exact resistance required ..

By the way, what is your target value for vcore ? I wonder that having vcore much higher than obtained with a modified bios is a HIGH risk for GPU (even if cooled with e.g. NV5)

[ZETT FX]

i do this mod
always 960 on 2.8
i think to get 2.95///

[SikaRippa]

i do this mod
always 960 on 2.8
i think to get 2.95///

2.95 V !! ... with your own risk

I hope, you have decent cooling (NV5 etc.) (?)

[ZETT FX]

why?
samsung memory may work on 3.0+

[ZETT FX]

find fast vgpu mode=)!

[S_A_V]

Hi there,

Sorry for my bad english btw, i'm not use translators.

I'm succesfully did Vmem-mod for my Leadtek 6800LE OEM

What I use:
1. 20k variable resistor, set it to 2.21 kOm for first time and connect. Sure that 10k (or may be 1k) is more suitable, but I use 20k may be used too if you will turn it very slowly (better do this not "on the flу" and in unconnected state).
2. Molex connector from some old case-fan.
No soldering, no pencil. Just set wires of molex connector to TP13 and TP14 and connect VR to molex.

Here is the table with resistance/voltage and higher stable memory clock reached in Mother Nature test.
------------------------------
|Resistance| Vmem | MemClk |
------------------------------
| -stock- | 2.73V | 1002MHz |
| 2.21 kOm | 2.89V | 1040MHz |
| 1.50 kOm | 2.94V | 1040MHz |
| 1.10 kOm | 2.99V | 1050MHz |
| 0.80 kOm | 3.04V | 1056MHz |
| 0.50 kOm | 3.11V | 1056MHz |
| 0.40 kOm | 3.13V | 1063MHz |
| 0.30 kOm | 3.18V | 1071MHz |
| 0.24 kOm | 3.21V | 1071MHz |
| 0.15 kOm | 3.26V | 1080MHz |
------------------------------

My heatsinks on memory chips stay cold with 3.26V as always with default voltage. But requires active aircooling with dedicated case fan.
Got max stable memory clock for all 3DMarks: 1002@2.73V (before) -> 1063@3.26V (after). Then I test performance difference.

PC configuration:
Venice 3200+ @ 2731MHz
BH-5 @ 227MHz/2.0-2-2-5/2T
6800LE @ 438/1002-1063 12/6
Windows 2k SP4, DirectX 9.0c, ForceWare v76.45

Results:
3DMark2001: 28599 -> 28737 (+138)
3DMark2003: 12847 -> 12981 (+134)
3DMark2005: 5379 -> 5426 (+47)

Got from +0.5 fps to +2.0 fps performance boost in all tests of all 3DMarks. It's very small for games/etc, but not for benches (anyway IMHO +1 fps in 05 is a not bad for half-a-hour of easy work with this mod). I think it may help to get #4 place with 6800LE in 03 for me soon

See pics:


2 ZETT:
thx to you and @lf for helping with Vmem mod - I'm not expect that will do it so easy and simple!

[SikaRippa]

why?
samsung memory may work on 3.0+

You have GW6800le with Samsung memories ??? I have assumed that all those 6800le-cards have Hynix-memories (HY5DU283222AF-xx), which have maximum absolute ratings of 2.95 V (3.6 V VERY absolute limit at at time).

--

Easy Vgpu-mod ? Read #4 in this thread carefully..

[SikaRippa]

Hi there,

Sorry for my bad english btw, i'm not use translators.

I'm succesfully did Vmem-mod for my Leadtek 6800LE OEM

<clip little bit..>

Very nice report !!

Quite nice performance increase, indeed ! (I hope your memories will not BURN with that voltage)

[@LF]

SikaRippa
On my Leadtek 6800LE the resistance between GND (TP9) and TP12 is 543 Ohms.
The resistance between FB1 and GND is 543 Ohms too.
The resistance near FB1 (Marked as R519) is 543 Ohms too!
I placed 4.7 kOhms between TP9 (GND) and TP12. The resistance between FB1 and GND decreased to 480 Ohms. On TP11 (GPU monitoring??) was 1.55V if in NiBiTor is 1.5V.
Sorry for my bad english, I'm russian overclocker

[ZETT FX]

and i have with 6.68 om 1.51
SikaRippa
Thank you

Альф ну тут без разговоров :beer: ! !

[@LF]

ZETT FX

and i have with 6.68 om 1.51

6.68 kOm...

|3dmark05-5787-barton 2950Mhz-# 2 on ORB

aaa... I'm #2 in 3dMark05...

SikaRippa
Thank you. Your fast Vmods - super!!!!!

[ZeroHero]

You can use what bios you want: because correct resistor value needed is not known you should use potentiometer in any case.

You can use idea you presented, HIGH value potentiometer between those points (it is again recommended to put a fixed resistance in series) and lower the value VERY carefully AND MEASURE vcore frequently. I do not want to give any values because of i do not exact resistance required ..

By the way, what is your target value for vcore ? I wonder that having vcore much higher than obtained with a modified bios is a HIGH risk for GPU (even if cooled with e.g. NV5)

Okay,even if I now have take holiday from benching now I think.
My target is about 1,5vc for benching and perhaps a bit lower for 24/7. If it not puts out too much heat and if voltage help a lot I wil perhaps test 1.55 or at least with cold air(to autum/winter) But not more. have a NV5 for cooling.

S_A_V: You have realy good scores for only 12/6 Congrats on the 30k too.
Someday shall i try for 30k too.

[S_A_V]

I'm succesfully did Vgpu-mod too!
It was a bit harder for me than Vmem-mod, but still without soldering and pencil!

Before doing this mod I flashed back Leadtek BIOS with default voltages (~1.1V).
Then I set 10k VR to 0.60 kOm and connect it to TP12 and GND (TP4).
Swith power on, see 2.15V at my multimeter and instantly switch power off.
IMHO 2V+ is too big voltage for NV40, so I increase resistance to 2.00 kOm and get voltage exactly that I get with "1.5V BIOS-mod" eg. 1.44V.

Table with resistance/voltage and higher stable memory clock reached in Mother Nature test:
-----------------------------------------------------------------
|Resistance| Vgpu | GpuClk | Cooling target
-----------------------------------------------------------------
| 2.00 kOm | 1.44V | 438MHz | aircooling
| 1.80 kOm | 1.48V |untested|
| 1.50 kOm | 1.55V | 455MHz | very good aircooling (zalman, etc)
| 1.35 kOm | 1.59V | 462MHz | stock/unmodded watercooling
| 1.20 kOm | 1.65V | 466MHz | good/custom watercooling
| 0.90 kOm | 1.82V |untested| chilled water or phase change
| 0.60 kOm | 2.15V | unsafe | better don't try w/o LN2 on GPU
-----------------------------------------------------------------

I set VR for 1.65V and get +60ºc temp on GPU under load with my modified Aucma CoolRiver (in normal mode, without cold water).
Got max stable GPU clock for all 3DMarks: 438@1.44V (before) -> 462@1.65V (after).

In "cold water mode" I already can bench with 450-455MHz at only 1.44V and I test 1.65V with cold water soon (may be on next weekend). I hope with about +25ºc..+30ºc under load GPU can reach freq. like 470-480.

Results of fast testing performance difference:
GpuClk: 438->462 (+24MHz)
MemClk: 1050MHz
3DMark2001: +697
3DMark2003: +490
3DMark2005: +207

Very good prformance increase! Vgpu-mod helps 4-5 times more than Vmem-mod.
I got +20 fps (250->270) for +24MHz@GPU in Nature test / 3D Mark 2001

See pic:


SikaRippa: Yours "fast Vmods for 6800LE/NU" helps very much. Thank you. I not planned running my card at 1.65/3.26 voltage for 24/7. It will be used for a short time when benching, so I hope my card stay alive

ZeroHero: Thanks. I think you have big chance to get 30k too if you do this Vgpu/Vmem mods. Good luck. But if my GPU will runs at 470-480 MHz under cold water then I may get about 30.5k btw beating 6800NU/DDR1 WR on 12pipes LE is my next target

[SikaRippa]

I'm succesfully did Vgpu-mod too!
It was a bit harder for me than Vmem-mod, but still without soldering and pencil!

Before doing this mod I flashed back Leadtek BIOS with default voltages (~1.1V).
Then I set 10k VR to 0.60 kOm and connect it to TP12 and GND (TP4).
Swith power on, see 2.15V at my multimeter and instantly switch power off.
IMHO 2V+ is too big voltage for NV40, so I increase resistance to 2.00 kOm and get voltage exactly that I get with "1.5V BIOS-mod" eg. 1.44V.

Table with resistance/voltage and higher stable memory clock reached in Mother Nature test:
-----------------------------------------------------------------
|Resistance| Vgpu | GpuClk | Cooling target
-----------------------------------------------------------------
| 2.00 kOm | 1.44V | 438MHz | aircooling
| 1.80 kOm | 1.48V |untested|
| 1.50 kOm | 1.55V | 455MHz | very good aircooling (zalman, etc)
| 1.35 kOm | 1.59V | 462MHz | stock/unmodded watercooling
| 1.20 kOm | 1.65V | 466MHz | good/custom watercooling
| 0.90 kOm | 1.82V |untested| chilled water or phase change
| 0.60 kOm | 2.15V | unsafe | better don't try w/o LN2 on GPU
-----------------------------------------------------------------

(clip)

Nice that you and others find that info usefull

--

Yep, 2.15 V would be kill the chip quite soon, that why I said HIGH value at first

[zombie]

How works the mod of the 6800Ultra to drop down the overvolt protection ?

[SikaRippa]

How works the mod of the 6800Ultra to drop down the overvolt protection ?

You mean how to make the mod OR how it actually works ?

Only mod for 6800 ultra (in this thread was originally mods for NON ultra) I know can be found here:

http://www.vr-zone.com/?i=926&s=1

I suppose that the so called overvolt protection of those cards is in fact a overcurrent protection of the switching regulator driving GPU. At least I have seen (I do not remember page) a mod to alter the current sensing resistors to shift up the maximum acceptable current level.

[total_assault]

will this one http://www.radioshack.com/product.as...Fid=271%2D1124 be good, and does it matter if its 1/4 watt and 1/2 watt, I don't understand what wattage has to do with resistence.

[zombie]

use 1/4 W, because this is the right size. the 1/2W are to great for soldering.

[total_assault]

I also Found this http://www.radioshack.com/product.as...5Fid=271%2D343 I would use this with a 2.2kohm resistor (as safety) for varible

[SikaRippa]

I also Found this http://www.radioshack.com/product.as...5Fid=271%2D343 I would use this with a 2.2kohm resistor (as safety) for varible

Hi,

Power rating (=watts) is not important here and you can use either resistor of 1/4 or 1/8 W, since current throught the resistor is low. More important is NOT to use too low value, I hardly suggest to put a fixed resistor series with potentiometer or trimmer (as explained in the figure).