Graphics Card Voltage Regulator Circuit Replacement? (Update: IT LIVES!!)

[iamdarkyoshi]

Quite a while ago I had an HD6950 Directcu ii graphics card die on me. It would work about one in 10 bootups, and when it does, it makes a very loud pulsating whining noise. The pitch changes based on the screen resolution and the time between pulses changes based on many frames the card is outputting on a graphics benchmark (kinda neat actually). The card is very unstable, and it often crashes at random times. Does anyone know what could be wrong? It happened randomly. I know of coil whine, but this seems to be more than just standard coil whine. My new cards work fine, and it does this in different systems.

[tyguy2]

I had this happen to me once, but I just replaced the power supply with a higher wattage one. What is the wattage of yours?

[obiwanjacobi]

If the power supply is of the switching kind I would look into any bad caps that may be on there. 9 out of 10 times that is the problem...

[2c]

[wraper]

Likely something wrong with VRM. You should check conductive polymer capacitors.

[iamdarkyoshi]

It has to be something on the card. The power supply is a brand new 750 evga supernova and it runs two hd7950s and it never needs to run its fan. So the caps might be bad? I couldnt find any shorted fets or open inductors. I measured every capacitor in each vrm and they measured 1.1 for the gpu, 1.25 for the vram, and one single phase vrm that drives...something measured 1.65, which seems strange. Here is some info from one of my posts in a different fourum:

I already checked each mosfet for shorts and they are all fine. I checked each choke while the card was running (it works occasionally if you mash the cmos reset) and the vram chokes measured the same AC and DC voltage, and the GPU chokes also measured the same. However, the unknown choke measured a bit higher and that VRM was outputting the highest voltage.
 
Measuring the  6 phase GPU VRM set, I got 1.0v idle and 1.1v load
Measuring the VRAM 2 phase VRM set, I (think) I got 1.25v
Measuring the unidentified single phase regulator, I got about 1.65V
 
Attached is an image of the basic layout (sorry, didnt feel like finding a high res image and launching photoshop)

 
 Does anyone know what this unknown section doeS?
 
Before anyone asks the cards serial number is missing and asus wont even touch it without a serial.

[tyguy2]

You might want to replace the odd ball cap. Sounds promising.

[iamdarkyoshi]

I have a cheap ESR meter but it only goes down to 0.01 ohms. Should I just order enough caps to recap the card? Nichicon 5MOhm ESR  7.2A ripple current sound good for the vrm output caps?

[kripton2035]

you dont need 0.01? to detect bad caps ... I use to have an esr meter with a 10 leds bargraph on it
and was able to detect bad caps for years before I got a deree 5000 lcr meter ...

[iamdarkyoshi]

you dont need 0.01? to detect bad caps ... I use to have an esr meter with a 10 leds bargraph on it
and was able to detect bad caps for years before I got a deree 5000 lcr meter ...

Its actually a full everything tester. capacitors, resistors, diodes, mosfets, inductors, ect. And it was about 25 usd. But you guys think the oddball vrm cap might be poop?

[tyguy2]

That's mostly likely the case. Most components should be below 1.5 volts

[iamdarkyoshi]

That's mostly likely the case. Most components should be below 1.5 volts

I traced out a small portion of the board a while back and there is a chip (8-pin surface mount) that controls this single-phase vrm. This chip is near some corrosion that formed on the card when someone decided to put a cold drink on the pc chassis. One drop of water must have gone through a fan and landed on the pcb, but this was a week before the failure. I am almost certain that the fault is with the oddball regulator and possibly some sense line on the chip. I am at work right now but when I get home I can post the part number of this chip. Its a 5-12v in single phase buck regulator.

[Monkeh]

I think you'll find the 1.65V is VRAM.

[iamdarkyoshi]

I think you'll find the 1.65V is VRAM.

would the vram be a single phase supply? What is the 2 phase vrm doing next to the vram?

Also, the 1.25v on the duel phase was from memory. I dont remember for sure.

[Monkeh]

I think you'll find the 1.65V is VRAM.

would the vram be a single phase supply? What is the 2 phase vrm doing next to the vram?

Also, the 1.25v on the duel phase was from memory. I dont remember for sure.

Well, the 1.65V is next to the VRAM, too.. so is the core regulator.

Try running GDDR5 at 1.25V and see how far you get. They simply don't specify it that low.

[iamdarkyoshi]

I think you'll find the 1.65V is VRAM.

would the vram be a single phase supply? What is the 2 phase vrm doing next to the vram?

Also, the 1.25v on the duel phase was from memory. I dont remember for sure.

Well, the 1.65V is next to the VRAM, too.. so is the core regulator.

Try running GDDR5 at 1.25V and see how far you get. They simply don't specify it that low.

Okay, well then what could the duel phase section be doing?

[iamdarkyoshi]

Update!! It lives!!!! The water damage is what caused the failure. There were a couple VIAs near the corroded area, and the vias were actually GONE! I put a pice of tiny copper wire in each one, soaked it in flux, soldered it for a good 10 seconds and the card actually works!!!! I think the major via that caused the issue was the sense wire for the buck regulator. It goes througn a couple passives and over to the fets. I ran unengine heaven for a while and the card is stable. I did try to desolder the capacitors in the vram section before this and they just popped off the leads. How in the holy cheezeballs do you solder such a massive slab of copper? My 6mm chisel tip couldnt take them out so I bodge soldered some super low ESR sanyo caps from a retired intel server onto the old leads and it seems fine. They run a bit warm though...

[tyguy2]

  awesome!

[iamdarkyoshi]

Now I should see if I can fix my old ecs elietegroup motherboard. The cmos chip seems to have gone into some sort of read only mode. Hitting "save and exit" does not change anything so you have to deal with the date being lost every time you reboot. The battery is fine. It doesnt even need to be power cycled to loose settings. The settings never get applied in the first place.