7900GT Voltmod-2

[madgravity34]

@onion, what are the two chips that get extremely hot? Could you post a pic please?

Also, did you do the 1.7v mod AND the 100k mod at the same time? I bet if you did that it was the 1.7v through the core that did more damage than the 100k mod. If not, then I think its important to figure out if you did anything wrong or if the mod isnt needed on some cards.

[onion]

Here is my setup.
At first I did the 1.5v mod by changing the resistor to 50k. I used two 100k in parallel to equal 50k. I also used the massive VF1 cooler with fan Picture so temps were very low. Everything was GREAT. I was running over 650mhz core rock stable and the core was at a cool 52c at full load. I was very happy with the results.

But I wanted MORE. I installed the 100k resistor in parrallel with the 80k "k" resistor as mentioned in this thread to boost freqency to the voltage regulator. Then I did the 1.7v mod. It was a BAD move! everything was Very unstable. I first ran 700mhz and the screen started to flash on and off and on and off. I dropped down to 650mhz and the same thing happen. Finally stock speeds were fine and stable. I removed the 100k and 1.7v and now the card does not even run stable at 650mhz and Im even getting tons artifacts at 600mhz. Stock speeds everything is fine though. So now Im back at 1.5v and the card does not clock even 600mhz . Below is what I did when the problems began. Note that the 100k resistor leads are NOT touching any part of the PCB even though it looks like it is in the pic below.

[onion]

These are the two chips that get BOILING HOT and MUST be cooled!! As you can see 2 members have already fried thier chips! Just put on a stick-on heat sink and you'll be fine. My heatsinks get really warm but thats good. Far from hot enough to damage anything though so as long as you sink them chips you will be 100% fine. Sorry for the blurry pic. Just sink them 2 chips!!!!! The smaller one gets much hotter than the larger by the way.

[quijonsith]

ah, that 100k resistor mod. jumping straight to 1.7v was a bold move onion. Good thing you had those sinks on the back of the card. Otherwise your card might be totally dead. Word of advice: if you change anything in a computer system only change one thing at a time. That way if you have problems you know exactly what caused them. Also, when overclocking anything and you need more voltage, only increase the voltage in small increments as needed so you can monitor performance, temps, and artifacting.

I'm running 1.5v right now but did the pencil mod for the switching frequency and wound up getting higher stable clocks after said pencil mod. I'd say it was the 1.7v that caused your problems. Sry you lost you clocks. Hopefully after the card has settled down for a while you'll be able to push it back to 650 again.

[snakeskinner]

Hmm, I have a wierd problem. I got a new 7900GT to play with, and so far it doesn't seem to be as good as my first one. The core goes high, but the memory won't go passed 830 or so.

Ok, the problem is, my resister labeled "K" in the first post of this thread, seems to not be working. I checked it with my meter, and there is nothing there. I made sure I cleaned both solder pads at each end of the resistor incase there was some film on there keeping me from getting good contact. Still nothing at all.

So what kind of problems could this cause?

Also, I checked the voltage to my memory and it seems to be at .800v. I know that can't be right, but i'm pretty sure I checked it at the right place. I shaded the resistor to up the vmem like the vrzone site says, and checked again at still .800 and the memory didn't seem to overclock anymore than before.

The card will run 730/830 with a 40mhz delta pretty well, but the scores are low compared to my other card. I only get around 11,200 or so in 3dmark05.

If you guys have any ideas I would appreciate them.


Thanks,

Marlon

Anyone have any ideas on this yet?


Marlon

[quijonsith]

Marlon, you sure your meter is working correctly? Are your leads good and sharp? Are you using a good ground with the black lead when measuring mem voltage? 830 is a pretty good o/c for the memory without any volt modding. Esp if it is, in fact, only getting .8v; which doesn't sound like the case given you got 830.

Also, who made your card? Double check that you are using the correct measuring point for the vmem:

[C-BuZz]

After a fair bit of searching I found this locally:

http://www.jaycar.com.au/products_uploaded/BUZ71.pdf

14A, 50V, 0.100 Ohm, N-Channel Power
MOSFET
This is an N-Channel enhancement mode silicon gate power
field effect transistor designed for applications such as
switching regulators, switching converters, motor drivers,
relay drivers, and drivers for high power bipolar switching
transistors requiring high speed and low gate drive power.
This type can be operated directly from integrated circuits.
Formerly developmental type TA9770.
Features
• 14A, 50V
• rDS(ON) = 0.100W
• SOA is Power Dissipation Limited
• Nanosecond Switching Speeds
• Linear Transfer Characteristics
• High Input Impedance
• Majority Carrier Device
• Related Literature
- TB334 “Guidelines for Soldering Surface Mount

Will this work as a replacement? Sorry not sure on how to read the electrial characteristics

Thanks

C-BuZz

[xit]

Hey if it does, and if it runs significantly cooler than the stock one, I may voluntarily replace that chip!

It seems to run at a rediculous temperature even after removing all mods. nvidia. That little chip is the single hottest in my PC.

[eva2000]

Some good (i suppose) & bad news.

Bad news, card is dead.. for now

Good news is dont use 1.85v on the core Now ya know.

The chip circled bottom left (attatched pic) caught fire... literally, & went up in smoke when I turned pc on. I need some info on what type of chip it is. I cant tell what type of chip it is as it's all charred & burnt. If someone could take a close up & provide some information on it be much appreciated

Chip has been removed & i'll try source a replacement to see if I bring her back from the dead..

Thanks

C-BuZz

ouch sorry to hear mate, hope you can fix it

[t024484]

After a fair bit of searching I found this locally:

http://www.jaycar.com.au/products_uploaded/BUZ71.pdf

14A, 50V, 0.100 Ohm, N-Channel Power
MOSFET
This is an N-Channel enhancement mode silicon gate power
field effect transistor designed for applications such as
switching regulators, switching converters, motor drivers,
relay drivers, and drivers for high power bipolar switching
transistors requiring high speed and low gate drive power.
This type can be operated directly from integrated circuits.
Formerly developmental type TA9770.
Features
• 14A, 50V
• rDS(ON) = 0.100W
• SOA is Power Dissipation Limited
• Nanosecond Switching Speeds
• Linear Transfer Characteristics
• High Input Impedance
• Majority Carrier Device
• Related Literature
- TB334 “Guidelines for Soldering Surface Mount

Will this work as a replacement? Sorry not sure on how to read the electrial characteristics

Thanks

C-BuZz

The Mosfet that you found is not good enough. It should have a channel resistance of less than 10mOhm. This one has 100mOhm, meaning that the power dissipation will be 10 times higher ( I*I*R).
Also the current is too low with 14A. Max Current should be between 50 and 100Amp.
The IRF7821 can be bought from: www.farnell.com.
I also found a supplier for the original SI4420: www.elfa.se.
Both Mosfet types are made by International Rectifier.
Both Mosfets on your board should be replaced. You can use the same Mosfet.
Good Luck

[xit]

The Mosfet that you found is not good enough. It should have a channel resistance of less than 10mOhm. This one has 100mOhm, meaning that the power dissipation will be 10 times higher ( I*I*R).
Also the current is too low with 14A. Max Current should be between 50 and 100Amp.
The IRF7821 can be bought from: www.farnell.com.
I also found a supplier for the original SI4420: www.elfa.se.
Both Mosfet types are made by International Rectifier.
Both Mosfets on your board should be replaced. You can use the same Mosfet.
Good Luck

to24484, what is your opinion on that chip running so hot?

Is it likely that a replacement (non-identical but adequate) could run cooler?

[t024484]

Thx all for your input.

I did not touch Vmem @ all & I had no vmem mods other than pencil mod in place. Only thing I increased was the Vcore. Please point out the second mosfet that needs replacing.

The mofset has a small metal pad underneath it, whith 6 contact points (contact points where underneath the metal pad connected to the pcb) the metal pad came off with the chip. Which was connected the the unburnt pins. What purpose do they serve?

In the first pic the contact point on the first pin seems to have completely burnt out? Anything I can do about that? Or just pray it still makes contact?

I did not notice any other burnt out chips, only the 1.



C-BuZz

Do not worry about the pin that is burnt away.
It is one of the three "Source" pins from the Mosfet.
You still can connect the other two pins to the board, that will do.
The Pin most to the right is the "Gate" and the four pins to the other side are the "Drain" of the Mosfet.
The second Mosfet to be replaced is directly under the damaged one, and is called "4410". This one is a Mosfet with slightly less specs than the 4420, better to use the same Mosfet for both.

The point that you probably refer to as "contact points" are just solder points in case of using a different type of mosfet with a solid underplate "Drain" instead of the four discrete pins, like the ones used for the Vgpu regulation (Q506 and Q510).
So take no notice of these "points", you won't need them.

[t024484]

to24484, what is your opinion on that chip running so hot?

Is it likely that a replacement (non-identical but adequate) could run cooler?

The reason the Mosfet gets so hot is because of a) its Channel resistance R and b) because of the switching on and off that takes some time.
The heat generated when switched on is I*I*R, with I being the current that is flowing.
The heat generated in the switching on and of, depends on several aspects like the Mosfet's caracteristics and the Voltage regulator.

There are two ways to reduce the heat, the easiest one being the use of heatsinks.

The other one is to use Mosfets with a lower Channel resistance and/or a faster switching characteristics, which is not too easy to achieve since the Voltage regulator is also involved.
The fact that the Voltage regulator is also getting hot, can only mean that the switching process is done with significant losses, but you will either have to find better Mosfets (IRF7821 ?) or use another heatsink.

[C-BuZz]

Ok i've got a couple of these IRLR7821 N Mofsets underway for some testing. Will update when they arrive..

Farnell do not have IRF7821 listed on there AU catalogue.

http://au.farnell.com/jsp/home/homepage.jsp

I'll do some basic testing with these until I can source IRF7821.

Thanks for your help, much appreciated.

C-BuZz

[xit]

The reason the Mosfet gets so hot is because of a) its Channel resistance R and b) because of the switching on and off that takes some time.
The heat generated when switched on is I*I*R, with I being the current that is flowing.
The heat generated in the switching on and of, depends on several aspects like the Mosfet's caracteristics and the Voltage regulator.

There are two ways to reduce the heat, the easiest one being the use of heatsinks.

The other one is to use Mosfets with a lower Channel resistance and/or a faster switching characteristics, which is not too easy to achieve since the Voltage regulator is also involved.
The fact that the Voltage regulator is also getting hot, can only mean that the switching process is done with significant losses, but you will either have to find better Mosfets (IRF7821 ?) or use another heatsink.

Thanks, that is very informative indeed.

Do you think that nVidia should have specced a better part that wouldn't get so hot even at default volts? It seems odd to me that these 2 tiny parts would end up being the hottest components in my PC. Perhaps nVidia should have specified them as parts requiring a small thin heatsink if a better (cooler) component was simply too expensive in comparison?

I gather it was essentially heat and not volts that caused Cbuzz's to explode, and therefore a heat sink at say 1.7 volts would make overvolting virtually 'safe'?

My other question is about why the GPU volt mod seems to increase RAM overclocking without doing a RAM mod specifically. I read previous that this 'hot' component in question was based around RAM and not GPU?

[albertwesker]

i have read the whole thread, but i didn't found something about this mod on a 7900 GTX, because i think i have the same problem like t024484, that ati tool works on high frequencies, but at comanche or 3dmark 2005 i get a blackscreen.




"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."


on my 7900 GTX the U502 is near the pcie connector.

[t024484]

Thanks, that is very informative indeed.

Do you think that nVidia should have specced a better part that wouldn't get so hot even at default volts? It seems odd to me that these 2 tiny parts would end up being the hottest components in my PC. Perhaps nVidia should have specified them as parts requiring a small thin heatsink if a better (cooler) component was simply too expensive in comparison?

It may be that is has to do with the memory chips. I have no idea what the spread in current demand is, or is it the voltage regulator chip that is making the difference? The two Mosfets are supposed to switch in alternation. If one mosfet is turned on while the other is still conducting for a short period, this causes a short circuit with huge currents to flow from 12V to ground.
Funny enough, the two components on my board do not get hotter than 50C.

I gather it was essentially heat and not volts that caused Cbuzz's to explode, and therefore a heat sink at say 1.7 volts would make overvolting virtually 'safe'?

The 1.7 Volt is for the GPU. The Mosfet that burned out is for Vmem.

My other question is about why the GPU volt mod seems to increase RAM overclocking without doing a RAM mod specifically.

Speed of information coming and going from RAM to GPU depends on speeds on both sides. If one of both gets faster ( set up time, hold time, etc )by applying more voltage, the frequency between the two can be increased.

[Nosfer@tu]

you don't have to use conductive ink, and you don't have to redo your mod everytime you want new voltage. you can use switches between the mod points and ground to toggle the +.05v, +.1v, and +.2v aswell as a switch to toggle between the 5k resistor and 50k resistor to toggle the base voltage between 1.2v and 1.5v. Doing so lets you adjust from 1.2v to 1.85v in .05v increments (though going higher than 1.7v is risky with any cooler and you'd need to adjust overcurrent protection to go over 1.7v)

Here's how you'd use a switch to toggle 5k and 50k:


Here's a breakdown of the different groundable points if you want to use a dip
switch.

thz alot mate. made me a lot more confident

[sluggo]

The mofset has a small metal pad underneath it, whith 6 contact points (contact points where underneath the metal pad connected to the pcb) the metal pad came off with the chip. Which was connected the the unburnt pins. What purpose do they serve?

In the first pic the contact point on the first pin seems to have completely burnt out? Anything I can do about that? Or just pray it still makes contact?

When you have chips that you know will get hot, you can manufacture them with the die mounted directly onto a metal pad, then encase the die, pad, etc in epoxy as usual. In the finished part, the bottom of the metal pad is exposed to the outside, on the bottom of the part. Now you can solder that metal pad to the board, and this allows heat out of the die more quickly than if the part had no exposed pad. In essence, you're using the PCB as a heat sink. It's not a great heat sink, but it definitely helps.

If you've burnt a pad on the board then you're at a bit of a disadvantage, to say the least. The sort of currents that were being switched on this part apparently required multiple leads to guarantee low R, and losing one pad means the other three now have to pick up the slack. If you're ambitious, you could hard-wire the pin to it's source, bypassing the traces on the board.

I wouldn't count on the integrity of a burnt pad - even if you can solder to it, there's no guarantee that the trace supplying it is in good shape.

Edit: Looking at the datasheet, I see you've lost one of the three source pins. I don't know if the part has those three pins each bonded to the die or they are joined in the leadframe and then bonded as one. If it's the former then I'd say there's a problem. If it's the latter then maybe not.

[C-BuZz]

Thanks for your input.

The metal pad is still intact & in good shape. The pad was sitting on top of the 6 contact points which would mean that they are all connected somehow & transmitting current (if there's a current flowing through it). The chip was sitting on top of the metal pad like a sandwitch.

The metal pad (very thin metal sheet) WAS actually connected to the 4 x unburnt pins & sitting under the chip. However when I removed the chip the pad snapped off with the chip. Which would also mean that the 4 x unburnt pins & the 6 x contact points were all connected to each other somehow

Cheers

C-BuZz

[snakeskinner]

Marlon, you sure your meter is working correctly? Are your leads good and sharp? Are you using a good ground with the black lead when measuring mem voltage? 830 is a pretty good o/c for the memory without any volt modding. Esp if it is, in fact, only getting .8v; which doesn't sound like the case given you got 830.

Also, who made your card? Double check that you are using the correct measuring point for the vmem:

Ok, i'm an idiot, I either looked at the wrong picture or read it wrong, cause I was measuring the wrong place for the memory voltage. I just checked again and its at 2.00.

Still nothing on the resistor marked "K" in the first post. I double checked it, and I'm on the right resistor for sure. My meter is working fine because it was reading the resistor for the mem voltage right. One side of that resistor is ground, and I get a good reading on that side. The other side I can't get anything.

I don't know much about how this stuff works, or what it does. But will the card even run if that resistor isn't working? And if it will, what will it do? I been playing oblivion on the card with no problems. Maybe there is just some hard coating on one solder pad that won't let me get a good connection to it, but I scraped it off pretty well.

BTW, card it EVGA, stock clocks 450/660.


Thanks,


Marlon

[quijonsith]

make sure your lead is decently sharp and, in the words of my shop chief, "stab the f**cker". sometimes you gotta stab into the solder pretty hard to get the reading.

[xit]

The 1.7 Volt is for the GPU. The Mosfet that burned out is for Vmem.

Yes so I thought, but Cbuzz says that he only overvolted the GPU and his MEM was at default volts. Why would that cause this mosfet to burn? Could it be burnt running MEM at 800 or 900 MHz even at default volts?

I'm amazed that your mosfets are only at 50 degrees....have you heatsinked them? What GPU / MEM volts have you settled on?

I am looking to get a few of these to do the job of cooling the mosfets. They come in either copper or aluminium, but I imagine that copper is better?
http://www.scorptec.com.au/index.php?prdid=10331

[sluggo]

The metal pad (very thin metal sheet) WAS actually connected to the 4 x unburnt pins & sitting under the chip. However when I removed the chip the pad snapped off with the chip. Which would also mean that the 4 x unburnt pins & the 6 x contact points were all connected to each other somehow

On that part the metal pad on the bottom of the chip is connected to the drain. The drain is also connected to all four leads on the side of the chip that didn't burn, so yes, you actually have (had) a lot of contact area for the drain.

The contact pad coming off the board is not great, but even if it had stayed on the board you wouldn't have been able to solder the replacement part to it anyway. With the loss of the pad and the source lead, this board's clockability may suffer.

[C-BuZz]

On that part the metal pad on the bottom of the chip is connected to the drain. The drain is also connected to all four leads on the side of the chip that didn't burn, so yes, you actually have (had) a lot of contact area for the drain.

The contact pad coming off the board is not great, but even if it had stayed on the board you wouldn't have been able to solder the replacement part to it anyway. With the loss of the pad and the source lead, this board's clockability may suffer.

I should be able to fabricate somthing so the points are all contacting the drain area. Anyway, replacement chips due in tomorrow so i'll see how it goes **fingers crossed**

Thx for your help, I understand how it all works now

C-BuZz