8800GT voltage mods

[jason4207]

my advice would be to everybody who owns or is buying a 8800gt. do not run fur bench as i havent and i have no stablilty issues or lockups at 876/20xx/1902 at 1.3 idle and 1.38 load. running great knock on wood.

+1. If you can pass ATI Tool, 3DMark0X loops, and games run fine there is no reason to tempt fate. 3 dead cards already...the stats don't look good for FUR!

Boycott FUR if you voltmod!

*Has anyone volt-modded, ran FUR, and can live to tell the story?

[largon]

Before the bugger died @ 1.68v I had had ran FUR heavily overvolted for a looong time. So I wouldn't jump to a conclusion that "FUR kill G92s" so lightly, it doesn't make much sense anyways. For example, when my card died it was OCP modded (the stupid way) and a mosfet burned consequently shorting the GPU.
I'm quite certain FUR didn't kill my card.
I killed it.

[scottc19]

just for giggles how high did you get clocks at 1.68v before it died?

[jason4207]

Before the bugger died @ 1.68v I had had ran FUR heavily overvolted for a looong time. So I wouldn't jump to a conclusion that "FUR kill G92s" so lightly, it doesn't make much sense anyways. For example, when my card died it was OCP modded (the stupid way) and a mosfet burned consequently shorting the GPU.
I'm quite certain FUR didn't kill my card.
I killed it.

Good point. I think my bad mount killed mine. This may be more coincidence than anything else. Still it probably isn't a bad idea to be extra careful if you decide to run FUR while volt-modded. I was running it full screen, and wish I had run it windowed so I could monitor temps more carefully.

I remember you posting that you replaced the MOSFET and the card didn't come back to life. Do you have any ideas on what other component died w/ the MOSFET?

Edit: re-read your post. So, you think the core shorted out? Can you see any evidence of this? Is it even possible to see evidence of a fried core? I know my Southbridge fried on a P4P800E-Deluxe Mobo I had from a static discharge via USB. I could actually see a small blister on the plastic casing of the chip.

[largon]

scottc19,
I don't remember exactly but it was at something like: 900 / 2500 / 1100 when it died.

jason4207,
No visible evidence except that DMM reads GPU-2-GND resistance to be pretty much non-existant.

[jason4207]

No visible evidence except that DMM reads GPU-2-GND resistance to be pretty much non-existant.

So, does that mean you can still boot? zbomb and I only get a 1/2 second power flash.

[Deluge]

Hey guys, guess I'm in the club now too.. I attempted this mod last weekend since my OC is severely limited by voltage, even with the stock cooler (core @ 715, with the 1.1v BIOS mod loaded around 45c, would jump to 46 after an hour of crysis), and my XFX alpha dog (black PCB, so sexy ) is now dead, may she rest in peace

Anyway, I'm trying to figure out what went wrong, as it never booted up after the mod. The only thing my friends or I could come up with is that the main soldering point couldn't touch the three surrounding optional points, but from what I've read, that's fine. The computer still 'boots', but it hangs before POSTing, so I've checked the voltage with my dad's old school analog multimeter (after desoldering everything) and it shows zero. I'm going to check with a friend's DMM this weekend and see if that also shows zip, then I'll decide what to do from there. Any ideas on what went wrong / any way to fix it are greatly appreciated.

[STEvil]

hard to tell anything without pics. The "3 points" are all connected in the PCB so it doesnt matter if you do a crappy job.. like me

edit - have you measured the resistance from the mod point to ground? Maybe your VR was set incorrectly.

[jason4207]

Hey guys, guess I'm in the club now too.. I attempted this mod last weekend since my OC is severely limited by voltage, even with the stock cooler (core @ 715, with the 1.1v BIOS mod loaded around 45c, would jump to 46 after an hour of crysis), and my XFX alpha dog (black PCB, so sexy ) is now dead, may she rest in peace

Anyway, I'm trying to figure out what went wrong, as it never booted up after the mod. The only thing my friends or I could come up with is that the main soldering point couldn't touch the three surrounding optional points, but from what I've read, that's fine. The computer still 'boots', but it hangs before POSTing, so I've checked the voltage with my dad's old school analog multimeter (after desoldering everything) and it shows zero. I'm going to check with a friend's DMM this weekend and see if that also shows zip, then I'll decide what to do from there. Any ideas on what went wrong / any way to fix it are greatly appreciated.

I like to use a magnifying glass and strong light to triple-check everything. Also, taking the 'beep' feature of the DMM and checking to see that you have conductivity where you want it, and don't have conductivity where you don't want it are good steps to take. You can also measure resistance at a few key locations. Typical Ohm ratings are spread throughout this thread and the GTS thread.

Make sure you tuned the VR to max resistance, and not the other way around! Since you have desoldered everything, is there still a lot of solder mess on the board? It's good to have a solder wick or suction to clean up the mess.

I haven't looked at an analog MM in years. How well does the one you used scale at these low voltages? I'd definitely recommend using the DMM.

Do you have any pics?

[jason4207]

Before the bugger died @ 1.68v I had had ran FUR heavily overvolted for a looong time. So I wouldn't jump to a conclusion that "FUR kill G92s" so lightly, it doesn't make much sense anyways. For example, when my card died it was OCP modded (the stupid way) and a mosfet burned consequently shorting the GPU.
I'm quite certain FUR didn't kill my card.
I killed it.

Here's some food for thought. I was doing some testing on my GTS last night. Idle voltages were 1.35v. ATI Tool load was 1.4v. 3DMark06 load was 1.42v. I remember having different load voltages depending on the app w/ my GT, but this could turn into a problem if FUR stresses the card so bad that the load voltage jumps up even higher. I wasn't measuring voltage when I ran FUR before.

Does anyone have any readings they can share that show how different apps increase load voltages by different amounts? I'm very curious if FUR might of increased my load voltage astronomically to unsafe levels.

[largon]

My 1.68v was during a FUR run. 1.63v on desktop.

[Deluge]

edit - have you measured the resistance from the mod point to ground? Maybe your VR was set incorrectly.

Yeah, the first time I booted it was at ~700 ohms (I used a 1k), I checked the VR legs and then the solder joints to the board. I tuned it down to ~400 ohms, then back up to 1k to make sure that wasn't the problem. I got a voltage reading one of the times (pretty sure it was at 400) of .18 to .21, but it varied too much so I figured it was wrong.

I like to use a magnifying glass and strong light to triple-check everything. Also, taking the 'beep' feature of the DMM and checking to see that you have conductivity where you want it, and don't have conductivity where you don't want it are good steps to take. You can also measure resistance at a few key locations. Typical Ohm ratings are spread throughout this thread and the GTS thread.

I got my friend's DMM at school today (security guards thought it was a weapon ) so I used the conductivity option to check pretty much everything I could think of. Everything seems right except the main solder point isn't connected to any of the 3 surrounding points. After a closer look, it seems like all the solder's gone (see pics) so I'll be adding some later.

Make sure you tuned the VR to max resistance, and not the other way around! Since you have desoldered everything, is there still a lot of solder mess on the board? It's good to have a solder wick or suction to clean up the mess.

That was the first thing I checked, booted up at ~700 ohms the first time and then tuned it up and down (while the computer was off, of course) to see if it would work with more or less resistance. I was extremely careful to clean everything of excess solder, took me a good 2 hours.

I haven't looked at an analog MM in years. How well does the one you used scale at these low voltages? I'd definitely recommend using the DMM.

I checked with the DMM a few minutes ago and it shows zero too, so I guess the analog one isn't all bad


(click for full size)

[jason4207]

It looks like you got some solder on the resistor to the right of the main solder point. Did you short that connection?

Also, the board looks dirty. Did you clean it well w/ some alcohol before you began the work? Was your soldering iron clean? Did you hold the iron on there too long? I like to get a little solder on the tip of the wire (tinned), and then just get it positioned and do a quick solder, so as to not heat up the components too much. If you try and hold the iron w/ 1 hand, and the solder coil w/ the other you end up using way too much solder, get things too hot for too long, and you have a mess on your hands.

You booted up at 700Ohms? Did you measure that while the mod was in place? It should have a reading closer to 13-15Ohms iirc when measured w/ the mod in place. With the mod in place you can't just measure the VR you are measuring the whole circuit. What I did was measure the VR before I installed it, and figured out that I wanted to turn it clockwise to turn up the voltage (like turning up the volume), so I used the legs that allowed that to happen, and cut off the other leg. Then I turned the VR counter-clockwise while I watched the resistance increase until the VR 'clicked'. That was max resistance and it was around 470Ohms. Then I soldered it on there, and cranked it up as needed while measuring voltage.

[Deluge]

The solder on the resistor to the right was an accident after I cleaned everything up, it never touched anything else, though.

I never cleaned the board, but the soldering iron was clean. I'm sure the soldering iron wasn't on too long, I added a small drop of solder to the tip of the wire, then held it in place and heated it until the solder melted.

Yeah, the 700 ohms was when it was all in place. What do you mean by the whole circuit? A vGPU point to ground, or..?

I did the opposite you did, clockwise = more resistance, and since I used a 1kohm VR, I decided to start at ~700 because of the previous BIOS vmod. Never really got a chance to play with the voltage, though

[scottc19]

Its hard to tell from the pics as only 1 of them is not blurry. It looks to me like the board might have gotten too hot, it sort of looks like the points around it melted a bit... I dont know for sure, because I'm just looking at the picture. I think this is what jason is refering to as looking dirty. Maybe you should use the beep feature or check resistance on the dmm and see if the point where you soldered is connected to the point where you spilled the solder. (the point to the right). Maybe its touching a trace or something.

[jason4207]

The solder on the resistor to the right was an accident after I cleaned everything up, it never touched anything else, though.

I never cleaned the board, but the soldering iron was clean. I'm sure the soldering iron wasn't on too long, I added a small drop of solder to the tip of the wire, then held it in place and heated it until the solder melted.

Yeah, the 700 ohms was when it was all in place. What do you mean by the whole circuit? A vGPU point to ground, or..?

I did the opposite you did, clockwise = more resistance, and since I used a 1kohm VR, I decided to start at ~700 because of the previous BIOS vmod. Never really got a chance to play with the voltage, though

When the mod is in place it is impossible to measure the resistance of just the VR. There are other paths for the electrons to take w/n the circuitry on the board, and you will be measuring the accumulated resistance of all those possible paths to ground. When you measure the main solder point to ground (or just try to measure the VR while it is attached to the board) you should get somewhere b/n 10-15Ohms. I'm getting 11.1Ohms right now at 1.35v idle. If you're reading 700Ohms something is definitely wrong unless you have removed the VR or it isn't connected properly.

I don't think it matters that you did the BIOS mod as I did the same thing. I do know that the 500Ohm is a little better to use, but you should be able to get by just fine w/ the 1k. It always best to tune to max resistance before you attach the VR b/c the more resistance there is the more it is like there is no mod there at all. Infinite resistance already exists b/n any 2 given points on the board through the air, so adding a high value resistor has the least amount of effect on the overall circuit.

I prefer to get it setup where turning clockwise increases voltage just b/c it is easier for me to remember. I also did the vmem buck mod to reduce memory voltage, and since decreasing resistance decreases voltage w/ that mod it is wired the opposite way, but the end result is that I turn counter-clockwise to reduce the vmem voltage. It doesn't matter which way you do it as long as you remember it correctly. Of course if your watching the voltage w/ a DMM you'd know pretty quickly which is the correct way to turn.

Here's a pic I borrowed from earlier in the post...thanks Dinos22!



You can see the trace in question. Make sure you have full connectivity from the IC leg all the way to the main solder point or at least to the last resistor. I'm not sure where the circuit or if the circuit continues below the surface to a lower level, or if it just travels through the SMC/R's from this point, but if you don't have full connectivity along the red trace I drew the you could at least try and make a solder bridge to reconnect and possibly save the card. All points along the red line are the same point in the circuit.

[jason4207]

Here are some pics of my adventure last weekend guys. I tried to take lots of pics of the GT along the way, but I kind of rushed the GTS, so i don't have as many pics.


---Brothers-------------GT+HR03GT-----------Kitchen work bench-------Operating table-----GT trembling w/ fear!


--Primarion--------------Easy Access------------vmem buck mod---------OCP mod [3.92k]--------all 3 mods----


---pre-testing-----------remove stock HS-------------Dirty GPU-------Clean GPU!-----


------Gt in action-----------same-----------------Close-up-----That's vmem, not vGPU!

I meant to take some pics of the card out of the case w/ all the mods, and HR03GT attached, but apparently I was too eager to get the card in the case!


If you read my above posts you know the GT didn't survive the weekend. I put it back to stock, and reused my mods on the GTS. No need to do OCP again, though.


------GTS in action!----------Underside--------Ain't she a beaut?------better angle-----


---More GTS------------------Night shot------------same--------

If you look at my avatar you might also be able to see the 25cm fan placed below the gfx card to help keep the RAMside cool while testing.

[os008]

I also did the vmem buck mod to reduce memory voltage, and since decreasing resistance decreases voltage w/ that mod it is wired the opposite way, but the end result is that I turn counter-clockwise to reduce the vmem voltage.

Hey Jason, could you please tell me why you reduced the V-Mem? seems odd, don't you want to increase it to overclock better?! or do those chips respond better to lower voltage because of heat or other reasons? and what kind are your chips? thanks.


@Deluge
I feel what you're in right now, i messed up my 3850 the same way you did exactly. I used way too much heat on the soldering point, to the extent that the traces got burnt, and the resistor nearby got out of place. I replaced that resistor and it worked again. Got greedy and made the volt-mod again, but if you looked at the board before that last volt-mod you'd pitty it, it was a MESS.

Now you'd guess everything was fine, it was, i got the card from 667 to 950, stable and great ... BAM, shorted the PCI-e cable poles, imagine, losing the card NOT to the volt-mod, but to n00000000bness :P.

Anyway, i learnt my lesson, and took the pictures of mods posted by experts here as reference ... presoldered the wire, fixed it in place with glue BEFORE soldering it, meaning ... look at Jason's mod, you'll notice he used a stickie substance below the wire on the board. THEN soldered it in place with just touching the iron to the tip of the wire for 1 or 2 seconds, and i was done. Put the card in place and voila, it worked, thanks God.

So it paid off losing the first card; became more aware and a bit more tidy , though a very expensive lesson .


EDIT: Thought i would add the picture of my mod, though not very clear.

[scottc19]

Hey Jason, could you please tell me why you reduced the V-Mem? seems odd, don't you want to increase it to overclock better?! or do those chips respond better to lower voltage because of heat or other reasons? and what kind are your chips? thanks.


@Deluge
I feel what you're in right now, i messed up my 3850 the same way you did exactly. I used way too much heat on the soldering point, to the extent that the traces got burnt, and the resistor nearby got out of place. I replaced that resistor and it worked again. Got greedy and made the volt-mod again, but if you looked at the board before that last volt-mod you'd pitty it, it was a MESS.

Now you'd guess everything was fine, it was, i got the card from 667 to 950, stable and great ... BAM, shorted the PCI-e cable poles, imagine, losing the card NOT to the volt-mod, but to n00000000bness :P.

Anyway, i learnt my lesson, and took the pictures of mods posted by experts here as reference ... presoldered the wire, fixed it in place with glue BEFORE soldering it, meaning ... look at Jason's mod, you'll notice he used a stickie substance below the wire on the board. THEN soldered it in place with just touching the iron to the tip of the wire for 1 or 2 seconds, and i was done. Put the card in place and voila, it worked, thanks God.

So it paid off losing the first card; became more aware and a bit more tidy , though a very expensive lesson .


EDIT: Thought i would add the picture of my mod, though not very clear.

The vbuck mem mod will work better for quimoda mem chips, like those on the 8800gt, from what I understand the 8800gts has samsung mem chips, and as far as I know these work better with more voltage. I have done the buck mem mod on my 8800gt and went from 930mem max stable to 1000 mem stable. I dont know exactly why this is, but I seem to remember hearing from someone that the quimoda chips were already overvolted to begin with... Maybe someone can help me out and better explain it... Possibly Viper John, as he is the founder of this mod.

[os008]

The thing is, up to 1000 is already stable on this card, and on my previous 3850. But i indeed would like if someone could explain the reason please. And thanks for replying .

[jason4207]

Both my GT and GTS have Qimonda RAM chips. I wish I had some Samsung's! The way I understand it is that both the vRAM, and the memory controller are fed from the same voltage line. The vRAM by itself should respond better to more voltage, and I think the specs have them listed up to 2.5v. The problem arises b/c the memory controller doesn't respond well to voltage. So, if you start raising the voltage you are just making the controller crash, and that is seen as the memory avalanche.

There is a definite 'sweet spot' that I was able to find on both cards. The GT liked ~1.84v, and the GTS liked ~1.96v iirc. Dialing in that sweet spot actually had me making the smallest turns possible on the VR to find that spot b/n memory avalanche (red or green crap all over the screen) w/ lock-up if the voltage was too high, and just plain lock-up if the voltage was too low. Those last few changes I made to the VR in finding the sweet spot didn't even register on my DMM, but they did allow me to pass 8 loops of 'Battle of Proxycon' form 3DMark03 at 1 higher notch on the slider. On my GTS I started at 1062 stock to 1101 w/ the mod. That was pretty easy, but it was tedious work to get 1107. On my GT I was able to go from 1026 stock to 1056 w/ the mod, but had to go back down to 1044 after I increased the core/shader clocks and voltage.

I forgot to mention this above, but a big thanks goes out to Viper John for all his highly informative help he gave me via email. And also to all you guys here like Largon who helped to find these mods. I'm not sure who all found what, but kudos to you gentlemen. My knowledge has increased 10 fold in the past couple months as I've researched, prepared, and then accomplished these mods.

Thank you!

[Deluge]

Alright, well I found the problem, and it's not looking too good. The two resistors directly below the solder point (red) aren't there, and the resistor next to them (blue) may or may not be shot. It's reading 10ohms, is that normal? Anyway, what do you guys think - try to save it or eat the $283 and move on? I've got no money for a replacement, being 15 blows

[BrokeDown]

Anyway, what do you guys think - try to save it or eat the $283 and move on? I've got no money for a replacement, being 15 blows [/IMG]

Could always sell a kidney on the black market.

[STEvil]

iirc I measured 18 ohms on my XFX on the blue resistor, but I could be wrong.

If the traces are still there it is fixable, but you would probably have to get a professional to do it.

[os008]

Both my GT and GTS have Qimonda RAM chips. I wish I had some Samsung's! The way I understand it is that both the vRAM, and the memory controller are fed from the same voltage line. ...

Thank you for the explanation Jason.