8800GT voltage mods

[largon]

gOJDO,
If your Asus G92GTS has the same PCB as in Demo's pic then it works your card too.
My Asus G92GTS 512MB is reference board so it wouldn't work on mine.

[Heinzkopf]

Heinzkopf,
That's perfectly normal, those components are mosfets that convert vGPU and vMEM voltages so it's normal they get hot especially when voltmodded and overclocked. All you can do is sink 'n' fan 'em.

The buzzing sound gets the stronger the higher your card OCs. Most likely it's caused by the encased vGPU inductors (big block-like components) on the frontside of the card. It's not dangerous at all.

Hmm okay that sounds appeasing to me.

But i'm still searching for a safe VGPU for permanent operation.

What max. IDLE Voltage would you (and of course also other members) personally choose without extra heatsink and cooling of the card's backside?

[gOJDO]

gOJDO,
If your Asus G92GTS has the same PCB as in Demo's pic then it works your card too.
My Asus G92GTS 512MB is reference board so it wouldn't work on mine.

Mine is reference board too. So, no vGPU pencil mod for us?

[jason4207]

Hmm okay that sounds appeasing to me.

But i'm still searching for a safe VGPU for permanent operation.

What max. IDLE Voltage would you (and of course also other members) personally choose without extra heatsink and cooling of the card's backside?

RAM sinks are ~$1/each. To me a few dollars spent to maintain some level of safety while improving clocks 15+% is money well spent.

I wouldn't go any higher than where you are w/o some sinks on there. The equivalent components on the reference design are cooled by the stock cooler. If nVidia thinks they should be cooled at stock voltage then don't you think they especially need to be cooled when volt-modding?

[Maduk]

I have find it for U

[Heinzkopf]

RAM sinks are ~$1/each. To me a few dollars spent to maintain some level of safety while improving clocks 15+% is money well spent.

I wouldn't go any higher than where you are w/o some sinks on there. The equivalent components on the reference design are cooled by the stock cooler. If nVidia thinks they should be cooled at stock voltage then don't you think they especially need to be cooled when volt-modding?

I have RAM heatsinks installed on my card. The RAM doesn't get hot anyway because I didn't increase the vDimm.
The stock cooler had just contact to GPU and RAM with pads. Other parts of the frontside weren't cooled by it. I just have a 120mm fan below the frontside at the bottom of my chase to blow cool air from outside onto the graphics card.
The back of the card gets really hot because here is no nearly no airflow. But these parts there had already no sinks installed with stock.
But you are right, maybe ill get some small sinks on the mosfets

[jason4207]

I have RAM heatsinks installed on my card. The RAM doesn't get hot anyway because I didn't increase the vDimm.
The stock cooler had just contact to GPU and RAM with pads. Other parts of the frontside weren't cooled by it. I just have a 120mm fan below the frontside at the bottom of my chase to blow cool air from outside onto the graphics card.
The back of the card gets really hot because here is no nearly no airflow. But these parts there had already no sinks installed with stock.
But you are right, maybe ill get some small sinks on the mosfets

When I said 'RAM sinks' I meant it generically. I used 'RAM' sinks on my MOSFETS, but I guess it would be better to just call them sinks.

You're right that the RAM doesn't really require sinks, b/c vdimm is low. However, the PCB will get hot, and through conduction heat up the RAM.

On the reference design the MOSFETS are on the other side of the board and are cooled by the stock HSF. Since you have a non-reference design you are going to have to adjust a little bit to compensate.

The entire PCB gets pretty hot when you start getting the volts/clocks up there so it is good to get sinks on anything you can...especially components that are hot enough to burn you. Also, rig up some fans or get a PCI fan card if at all possible.

[Heinzkopf]

The entire PCB gets pretty hot when you start getting the volts/clocks up there so it is good to get sinks on anything you can...especially components that are hot enough to burn you. Also, rig up some fans or get a PCI fan card if at all possible.

Yes that's true, the whole PCB gets really hot especially around the area of the mosfets. Well best solution was installing a fan that blows air to the backside of the card. But that is a bit dump to put the fan on it- and since I have a silent system another fan would increase the noise of the computer.
Well i'll try to improve the temperatures passivly by heatsinks first. that means sinks on mosfets and maybe the copper mounting holes.
Think that should be enough then.
With stock cooler the GPU reached 90°C as temperature by standard and so I think the card got even a bit hotter as now. additionally I think these mosfets are designed for high temperatures. But yes the ramchips may be the most sensitive parts on the card

[jason4207]

Yes that's true, the whole PCB gets really hot especially around the area of the mosfets. Well best solution was installing a fan that blows air to the backside of the card. But that is a bit dump to put the fan on it- and since I have a silent system another fan would increase the noise of the computer.
Well i'll try to improve the temperatures passivly by heatsinks first. that means sinks on mosfets and maybe the copper mounting holes.
Think that should be enough then.
With stock cooler the GPU reached 90°C as temperature by standard and so I think the card got even a bit hotter as now. additionally I think these mosfets are designed for high temperatures. But yes the ramchips may be the most sensitive parts on the card

Silence and volt-modding don't really mix well, but you may be able to find a good balance.

I'm sure the MOSFETs are designed for higher temps, but I'm also sure they'll work better if you can keep them cooler.

[Nosfer@tu]

1.2k isn't too low, but there is no reason to go that low unless you are trying for some serious clocks w/ major voltage under DI/LN2. ~4k is better suited to air/water.

The guy that lost his card had removed an SMR. Thus removing the circuit that detects OCP, and also making it difficult for the card to balance the load b/n the MOSFETS. One MOSFET received too much current, and blew up!

And here's why. I am set for ~1.45v load. Every bench runs around that. Lightsmark will hit about 1.475v. But FUR on the other hand spikes the card over 1.52v. This is probably why people's cards are dieing. Mine is water cooled and never breaks 38°C loaded. I have sinks on everything the stock cooler touched, plus a couple other components I found to be hot.



I used 2.2k ohm 1/4 watt resistors from Radioshack. Soldered wire leads on them and soldered the leads to the surface mount caps on the board.

Thx guys

I will try out with 1.2 K so the card can get some Dice

[scottc19]

What should be the resistance between these two points (the blue ones)? I think maybe I messed this up, I have close to 16mil between these two points.

[largon]

scottc19,
Resistance between the SMR you marked should be around 22Ω.

[Heinzkopf]

Silence and volt-modding don't really mix well, but you may be able to find a good balance.

I'm sure the MOSFETs are designed for higher temps, but I'm also sure they'll work better if you can keep them cooler.

Yes u're right.
Now I installed some small heat sinks on the card, hope they help keeping it cooler.


With them I think I can set the idle voltage to 1.2V and see what's max clock with it.

[XSS]

1100MHZ no V-Mod stable my card rocks !

[largon]

XSS,
Samsung GDDR3, I presume?

[Kabauterman]

Hey guys,

at first the pic:



this is my second try with the OCP mod all soldering points have contact but I think the mod didn't works.

First problem: Card clocks 770/1900/1030 without VMods but for a "little" more GPU Clock I need much more Voltage ( 1,45 V for 850/2200) I think that can not be normal

Second problem: more than 850 MHz isn't possible, not important what voltage I give

Third problem: 850/2200/1000 @ 1,45 V is stable in all Benches but AM3 fully runs and before the score is displayed the screen goes black and don't get back.


This sounds to me like a non working OCP mod :-( can someone help me?
I know that the memory clock is important to the Shader clock, so I test each Clock once.

[Mod_Aap]

Vmods for the new XFX8800GT alpha dog edition and the 8800GS:



VGPU: 5K var. resistor.

VMEM: 15K var. resistor.

PWM used for GPU is BN-9G
PWM used for DDR is SC2621A

[largon]

Kabauterman,
What is your vGPU after a crash/black screen?

[Kabauterman]

Kabauterman,
What is your vGPU after a crash/black screen?

vGPU is the same as before 1.45V
I can not realize why only AM3 shows the Black screen and why it runs fine during the test.

[largon]

vGPU is the same as before 1.45V

Then your problem has nothing to do with OCP. That leaves just the traditional "draw of luck" OC'ability of each chip.

[dinos22]

yaaaaaaaaaaaaaaaaaay

i'm no longer a modding virgin

[aussie-revhead]

If I can do it anyone can

Good work Dino , now how well does it clock ?

Have you done the second card yet ??

[dinos22]

gained 100MHz on core

still on stock cooling

haven't done the second card
my soldering iron isn't that great cause i had trouble soldering to ground points
just wouldn't stick :S

but i got it in the end

the read point was reading 1.988v LOL and then i checked it manually and found it was reading 1.111 ....that was a relief heh

[aussie-revhead]

gained 100MHz on core

So what does that add up to ?

That vgpu point is damn tiny ey ? (in your third last pic)

[dinos22]

to be honest that was the easiest one to solder

i am just quickly benching

did a 92K 3DMARK01
single card :o pretty good for card clocks and CPU at 5.1GHz