7900GT Voltmod-2

[t024484]

I started a new thread to have a clean start for those who have not followed the previous thread, and to bring things to a final state.
This was possible because I received my 7900GT yesterday, and could do quite some testing to solve some problems that where still there.

The heaviest test I could use was "Comanche 4 Demo Benchmark Test", because with this Benchmark I got black screens long before other benchmarks like 3DMark05 were doing.

Black screens are the prove that the Voltage Regulator, the ISL6568, doesn't "like"something.
This something can be either some sort of voltage protectection, either undervoltage, overvoltage or too much ripple.
It can also be overcurrent in some form.

I could go to 710Mhz on the core with AtiTool without artefacts, but Comanche did not let me get passed the 545Mhz, and gave me black screens above. This was with 1.55Volt Vgpu.

To make a long story short, it was not the overvoltage protection giving me problems, neither was it the overcurrent protection. It turned out to be that with the preset switching frequency, the convertor could not supply enough energy.
This type of "bucket"convertor supplies a packet of energy with each bucket that is tranferred at a clock period. To get more energy, more buckets have to be transferred, and hence the swiching frequency has to go up.

That is exactly what I did, and every benchmark can now run at 710 Mhz with a Vgpu of 1.55Volt.

The picture below shows AtiTool, where you see that it is running for quite some time without artefacts at 710Mhz Core and 1800Mhz Mem.
Temp does not exceed 62C with my ZalmanVF700.



All the necessary mods are combined in the next mail, because I did not manage to get everything in this mail!

[t024484]

I have tried to bundle all the wisdom of previous mails in the pictures plus the explaination below.



Resistor "J" is a 5Kohm resistor, and causes the Core Voltage, Vgpu, to be 1.2Volt in the unmodified situation.
Replacing this resistor with a 50Kohm resistor, changes the Vgpu to 1.5Volt.
In the picture hereafter, you can add to this voltage, thus being 1.2 or 1.5Volt, another .05, .1 and .2, either as a single step or as a combined set of steps.
So your options are: 1.2, 1.25, 1.3, 1.35, 1.4, 1.45, 1.5 or 1.55Volt
and with the 50Kohm mod: 1.5, 1.55, 1,6, 1.65 and 1.7Volt as a maximum.

Resistor "H" measures 3.5 Kohm, and controls the Over Current Protection.
If you suspect the OCP giving you trouble, you can double this resistor to 6K8 Ohm. I had no problems, so you can probably leave it as it is.

Resistors "I" are said in the previous thread to disable the Over Current protection . THIS IS NOT THE CASE.
These resistors. measuring 1,8Kohm, are meant to balance the current between the two pair of switching Mosfets to carry an equal amount of load.
Leave them, or replace them with values of 3,9 kohm which should only be done if you are having mysterious problems.

And than comes a very important Mod, being resistor K.
This resistor measures ca 80Kohm, and sets the switching frequency.
Connecting a 100Kohm resistor in parallel, almost doubles the switching frequency, and solved my problems completely.
You could eventually do this mod by pencil, try to get the value down to approx. 45Kohm.

The next picture shows the Mods to change the 1.2 or 1.5 Volt to a higher value.
C gives 0.05 Volt increase when shorted to ground
B gives 0.1 Volt increase to ground and
A gives 0.2 volt increase to ground,
Plus any combination of A, B and C.



I have given two possible routes to achieve your Vgpu increase.
Use either the upper or the lower set of traces.
When you only use the red trace, you will have 0.05Volt increase. (1.25V or 1.55Volt)
When you use Red plus Yellow, you will have a 0.15Volt increase (1.35Volt or 1.65Volt)
When you use Red, Yellow plus Purple, you will have a 0.35Volt increase ( 1.55Volt, do not use it in combination with the 50Kohm resistor mod)
Of course you can also connect a single point to get what you want, f.i. ground point C (sorry for the earlier typo) to get 1.4Volt.
Although a bit more difficult to achieve, it is still doable.
Go with conductive ink with a large U around all the other points towards ground.

You can either do the modding with a thin wire and solder, or with conductive ink.
Do not worry about all the points between C and B, they are not connected to anything, and are used to keep the lines as simple as possible.

Look at the bottom of Page 22 of this thread for an even easier way to do your voltmodding.

Also have a look at the bottom of page 20 for the possible addition a 47Kohm resistor, giving the system a higher stability

I have republished the expired Pictures on Page 26, posting nr 640

[DesertShooter]

Very nice m8 that's usefull information, will try the 100k mod this afternoon.

[SF3D]

I will also try 100K mod this afternoon.

This new thread is very good idea.

[pcfreak]

The next picture shows the Mods to change the 1.2 or 1.5 Volt to a higher value.
C gives 0.05 Volt increase when shorted to ground
B gives 0.1 Volt increase to ground and
A gives 0.2 volt increase to ground,
Plus any combination of A, B and C.



I have given two possible routes to achieve your Vgpu increase.
Use either the upper or the lower set of traces.
When you only use the red trace, you will have 0.05Volt increase. (1.25V or 1.55Volt)
When you use Red plus Yellow, you will have a 0.15Volt increase (1.35Volt or 1.65Volt)
When you use Red, Yellow plus Purple, you will have a 0.35Volt increase ( 1.55Volt, do not use it in combination with the 50Kohm resistor mod)
Of course you can also connect a single point to get what you want, f.i. ground point B to get 1.4Volt.

I am a little bit confused here. On the upper route, in order to get 0.05v increase do we have to connect the C point to ground or all the red traces? And for a 0.1v increase the B point to ground or all yellow traces as well?
Thanks for the great info!

[SF3D]

It work great. Now I'm able to use 1.6V
All benchmarks run fine 740+20MHz delta now.

[t024484]

I am a little bit confused here. On the upper route, in order to get 0.05v increase do we have to connect the C point to ground or all the red traces? And for a 0.1v increase the B point to ground or all yellow traces as well?
Thanks for the great info!

The text says red for 0.05 and red plus yellow for 0.15. Red means ALL the red and yellow means ALL the yellow.

[illuminati-hwt]

Sweet guide t024484!

Regarding "resistor K", I measured it at 85-86kohm. Did the pencil-mod (bare leads 0,5 HB) and hit 44,6kohm. So the lower the resistance, the higher the switching frequency?

85k+100kohm = 45,9kohm.

Will high switching frequency have any negative impact?

[t024484]

Sweet guide t024484!

Regarding "resistor K", I measured it at 85-86kohm. Did the pencil-mod (bare leads 0,5 HB) and hit 44,6kohm. So the lower the resistance, the higher the switching frequency?

85k+100kohm = 45,9kohm.

Will high switching frequency have any negative impact?

Nothing that I can imagine.

[Hell-Fire]

Very nice thread bro. I love consolidation threads like this. Sometimes people need just get the info at hand versus reading page after page of conjecture and opinion.

Many times people have screwed their hardware by doing mods that are on one page of a thread, just to learn that 10 pages later they realized the mod wouldn't work. You will get people saying you should have read the whole thread, and this is true, but even great overclockers get impatient and just want the bare facts.

[fhpchris]

When you use Red, Yellow plus Purple, you will have a 0.35Volt increase ( 1.55Volt, do not use it in combination with the 50Kohm resistor mod)

Does :

1)50K ohm resistor mod + Yellow + Purple = 1.7V?
2)50K ohm resistor mod + Red = 1.7V?

I think #2 is the case.

[***Deimos***]

Does :

1)50K ohm resistor mod + Yellow + Purple = 1.7V?
2)50K ohm resistor mod + Red = 1.7V?

I think #2 is the case.

50Kom resistor mod 1.2V->1.5V

red = +0.05V
yellow = +0.1V
purple = +0.2V

so what do you think?

all of your above options are WRONG.

EDIT: on GTX:
"50K ohm resistor mod + Yellow + Purple = 1.7V?"
that would correspond to invalid vid selection of 1.8V. (same for GT)
"50K ohm resistor mod + Red = 1.7V?"
that would correspond to invalid vid selection of 1.75V. (1.55V for GT)

EDIT2:
For 1.7V you only need to connect purple (pin21 to gnd), which should already be the case for the GTX.

[DesertShooter]

A bit messy, but here are the vmods i did :

[Ebola]

Very good job, clear and explicit, at least this time, the voltmod is more exiting than a boring trimer soldering

[t024484]

A bit messy, but here are the vmods i did :

The look of it is almost frightening.
I can see the Vmem mod, the 50Komm Vgpu mod, the 100K Frequency mod, but I'm not sure what the red wire is with the white something attached to it.
I am surprised how well the board accepts all those long wires, they must be acting as antenna's.
But one thing is for sure, you had the guts to do it. A great achievement.

[t024484]

I have removed all the wires and resistors from the various experiments I did over the last two days, cleaned up the card and made all wires as short as possible.
Much to my surprise, I was unable to get the 710 Mhz I had before with so much ease after having increased the frequency.
The reason for that took me quite some time to find out.
I had from earlier experiments a 47Kohm resistor still attached from pin 3 to the 5volt.
Without this resistor I am falling back to 649Mhz as a maximum with Comanche 4 Demo.
I have no explaination for the why, but my board definitely needs this extra resistor. Reading the mail from SF3D, he obviously does not need it.
This proves that there is quite some spread between boards.

[fhpchris]

50Kom resistor mod 1.2V->1.5V

red = +0.05V
yellow = +0.1V
purple = +0.2V

so what do you think?

all of your above options are WRONG.

I was only asking because it was incorrectly stated in some of the pictures in the other thread. I also notice that you cannot go above 1.7V this way.

50k ohm + red + purple = 1.75, but that is not in the table:

-------1.500-1.525-1.550-1.575-1.600-1.625-1.650-1.675-1.700Volt

Pin22----0-----0----0------0------0-----0------0----0-----0
pin21----1-----1----1------1------1-----1------1----0-----0
pin30----1-----1----1------0------0-----0------0----1-----1
pin31----1-----0----0------1------1-----0------0----1-----1
pin32----0-----1----0------1------0-----1------0----1-----0

Notice how 0-0-1-0-0 is NOT in that table? Look @ 1.45V below

-------1.200-1.250-1.300-1.350-1.400-1.450-1.500-1.550

Pin22----0-----0----0------0------0-----0------0----0
pin21----1-----1----1------1------0-----0------0----0
pin30----1-----1----0------0------1-----1------0----0
pin31----1-----0----1------0------1-----0------1----0
pin32----0-----0----0------0------0-----0------0----0

Saying what you said is true most of the time, but not all of the time. I only want to make sure that everyone knows that the 50k ohm mods are not all the same.

so what do you think?

What do I think? Some of these facts were misstated in the other thread, and I just want to make sure the compilation here was extensive and included things NOT to do. Go take your attitude elsewhere, I know how to add.

[Haltech]

What has been everyones experience with the BFG 7900GT OC ? This is the card i have. Still waiting for my conductive ink pen to arrive. I did the pencil mod to 91C resistor to bring it down to 45k.

[Hugo van Dijk]

I had from earlier experiments a 47Kohm resistor still attached from pin 3 to the 5volt.
Without this resistor I am falling back to 649Mhz as a maximum with Comanche 4 Demo.
I have no explaination for the why, but my board definitely needs this extra resistor. Reading the mail from SF3D, he obviously does not need it.
This proves that there is quite some spread between boards.

Interesting, This is defanlty something to look into

Maybe it might even give an extra boost to other cards.

[***Deimos***]

I was only asking because it was incorrectly stated in some of the pictures in the other thread. I also notice that you cannot go above 1.7V this way.

50k ohm + red + purple = 1.75, but that is not in the table:



Notice how 0-0-1-0-0 is NOT in that table? Look @ 1.45V below




Saying what you said is true most of the time, but not all of the time. I only want to make sure that everyone knows that the 50k ohm mods are not all the same.



What do I think? Some of these facts were misstated in the other thread, and I just want to make sure the compilation here was extensive and included things NOT to do. Go take your attitude elsewhere, I know how to add.

To the best of my knowledge nothing was mis-stated. Anything that SF3D, TimeOut, DeserShooter, Cronic, illuminati-hwt, acjof, and in particular t024484 have posted is as good as gold. Now, just because somebody doesnt read the diagram etc properly and asks "is this right for 1.55V?" or something to that effect, does not make their assertion valid. Just follow the original guides. I grouped them all together in one post on page 8 in the 7900GT volt mod thread.

[t024484]

Does anybody have a clue why the pictures in the first two mails of this thread are invisible.
When I edit those mails and preview them, the pictures are there, but when I have saved the message, the pictures are no longer visible.

Forget my question, it is working again.

And now it is gone again. Quite mysterious.

[illuminati-hwt]

t024484, no problems with your pics all day..

btw, if some ppl are wondering, "preset switching frequency"-mod did nothing for oc at 1,2v. As t024494 says, it'll only add "stability" when vmod is applied.

Cant wait to get my fingers on a VF-900.. anyone seen 'em yet?

[Garrett]

t024484, no problems with your pics all day..

btw, if some ppl are wondering, "preset switching frequency"-mod did nothing for oc at 1,2v. As t024494 says, it'll only add "stability" when vmod is applied.

Cant wait to get my fingers on a VF-900.. anyone seen 'em yet?

I assume that depends on where you live, here in The Netherlands they're easily available

[illuminati-hwt]

They dont even have 'em in germany mate.. nor sweden or denmark. I mean "in stock" vf-900. Everybody says 5th april for stock status.

[cronic]

great work on consolidating the info.