My Experiences Reflowing Videocards (so far 100% success rate)

[iamdarkyoshi]

So for christmas, instead of asking for parts for my gaming PC and stuff, I asked for a good soldering iron/reflow station. So I ended up with a unit of crummier quality to this one http://www.ebay.com/itm/853D-Rework-Soldering-Station-SMD-Hot-Air-Iron-Gun-DC-Power-Supply-6-Gifts-/252027806219?hash=item3aae07220b:g:uRQAAOSwu4BVphKX

But I mean it works great, much better than guess and check with the 20 year old two temp heat gun and my old weller 40w plug-in iron.

So first, I determine if the GPU needs a reflow in the first place. I run it, if it artifacts, almost certainly a reflow is needed. If it simply shows no video or crashes under load, it might be a bad BIOS image, faulty voltage regulators, it could actually need a reflow, or is might just be broken beyond economical repair.

The PROPER way to fix stuff like this is a reball, but reballs are a HUGE pain.

So if it needs a reflow or I simply cannot figure out any other fix to try, I usually just dissasemble the card, remove anything that can melt (like plastic parts near the GPU die) clean it well, prop it up on my table with little tinfoil balls and  slowly warm it up over the course of 3-10 minutes, depending on the board. I set my hot air gun to 270 degrees. I know this is a little high, but I think the calibration is a bit off on my unit.

Removable mobile videocards (imac/gaming laptops) are usually a 3 minute preheat. Desktop cards I usually give longer, so maybe 5-10 minutes, and large laptop boards might even need more. The goal here is even heating to minimize thermal expansion strain across the board.

After preheating, I "go in for the kill." This is where I slowly move the nozzle closer to the GPU chip, trying to keep it evenly heated and the surrounding area warm as well. Once I work my way about 3/4in away from the chip, I keep applying heat for a couple minutes, again, more time for the larger boards as they contain lots of copper and will sink the heat, making it harder for the solder to melt and reflow.

After this, I slowly back out in the same procedure as before. After slowly cooling it down by moving the hot air gun farther away, once I get about 2 feet away and the board is under 100 degrees, away I usually put a cardboard box or something over it and let it cool down on its own for another 10 minutes. The key is NOT to rush.

After this, it is time to put it back together and test it! Apply new thermal paste, thermal pads, and make sure everything is fastened properly. Put it in the machine and boot it up!

So far I have fixed:
R9 270x
Three iMac video cards
HD4850x2

And nothing has yet to fail after a reflow so far. I know there are plenty of things I should do differently, but I have learned all of this through experimentation. Feel free to drop a comment or suggestion below, I look forward to hearing your opinions!

[timb]

Hehe...

[Ian.M]

The problem is:  Any failed joints on any balls have cracked and the surface oxidized.  When you reflow without adding flux, all you are doing is melting the solder balls, it does nothing to clear the oxides off the pad on the chip or the board (whichever side had the crack in the joint).  It then cools to the solidus temperature and as it cools further the solder ball is squished against the oxidised pad it hasn't wetted by the contraction of the other balls and the resin underfill.  This makes a pressure contact, and all appears OK, but solder cold-flows with pressure and time,  so the contact pressure will decrease to zero.  Add some thermal cycling and vibration and it will soon become bad again.

If the chip isn't underfilled, and you can ultrasonic clean the board then run some diluted no-clean rated liquid RMA flux under the chip, your chances of long-term success improve.  Otherwise you have just thermally stressed the chip and the board to no good purpose.

[ElektroQuark]

Storm is coming...

[wraper]

Storm is coming...

The question is how long they'll live after this procedure. Did you check if the solder has actually melted (by trying to move the chip)? It is very hard to heat a big BGA chip with such hot air station. Also, in most cases this is chip fail (bump ball fail inside the chip) rather than soldering fail. In such cases heating the chips usually revives them for some short time. So you can judge how successful those repairs were only after a year or two.
In cases when there was actual soldering fail, pads get oxidized and just melting the solder again, usually won't make them attracting the solder again.

[iamdarkyoshi]

The problem is:  Any failed joints on any balls have cracked and the surface oxidized.  When you reflow without adding flux, all you are doing is melting the solder balls, it does nothing to clear the oxides off the pad on the chip or the board (whichever side had the crack in the joint).  It then cools to the solidus temperature and as it cools further the solder ball is squished against the oxidised pad it hasn't wetted by the contraction of the other balls and the resin underfill.  This makes a pressure contact, and all appears OK, but solder cold-flows with pressure and time,  so the contact pressure will decrease to zero.  Add some thermal cycling and vibration and it will soon become bad again.

If the chip isn't underfilled, and you can ultrasonic clean the board then run some diluted no-clean rated liquid RMA flux under the chip, your chances of long-term success improve.  Otherwise you have just thermally stressed the chip and the board to no good purpose.

How would one ultrasonically clean the board? I have thought about running flux underneath the chip, but I have never heardnof someone doing it.

[Ian.M]

The whole board gets dunked in a tank of an appropriate cleaning fluid in an ultrasonic cleaner.  What happens after that depends on the cleaning fluid used, but may involve rinsing with deionised water and/or IPA then low temperature hot air drying.

The hacks who think they can fix boards by reflowing them with a DIY store hot air gun or in their domestic oven don't have the knowledge and experience to know that flux helps, or if they do don't want to buy it by the litre, or how to select a suitable flux.  The pros, either replace the chip, or if the chip is unobtanium, want the best possible chance of success so will remove it and reball it. Either way they get to clean the pads on the board and use their favourite SMD rework flux. Stuff that dies three months down the line will kill your reputation so they don't want to risk it - the job should be done right or rejected.

However as Wraper has pointed out, the odds aren't in your favour, because especially for high wattage chips, the failure is more probably the die attach bumps inside the package, rather than the substrate to PCB balls, as they are smaller and under a lot more thermal stress.  Reflowing works if there is a board level mechanical design fault e.g too flexible a PCB that is distorted  by the heatsink mounting or other pressure, otherwise the odds are poor.

[iamdarkyoshi]

The whole board gets dunked in a tank of an appropriate cleaning fluid in an ultrasonic cleaner.  What happens after that depends on the cleaning fluid used, but may involve rinsing with deionised water and/or IPA then low temperature hot air drying.

The hacks who think they can fix boards by reflowing them with a DIY store hot air gun or in their domestic oven don't have the knowledge and experience to know that flux helps, or if they do don't want to buy it by the litre, or how to select a suitable flux.  The pros, either replace the chip, or if the chip is unobtanium, want the best possible chance of success so will remove it and reball it. Either way they get to clean the pads on the board and use their favourite SMD rework flux. Stuff that dies three months down the line will kill your reputation so they don't want to risk it - the job should be done right or rejected.

However as Wraper has pointed out, the odds aren't in your favour, because especially for high wattage chips, the failure is more probably the die attach bumps inside the package, rather than the substrate to PCB balls, as they are smaller and under a lot more thermal stress.  Reflowing works if there is a board level mechanical design fault e.g too flexible a PCB that is distorted  by the heatsink mounting or other pressure, otherwise the odds are poor.

I totally agree, this is why I sell my stuff with a one+ year warranty on reflow failures and I do include details in the ad that I had done a reflow on it and the expected lifetime before another repair can vary. I do put them under rigorous tests for at least a week, turning them off and on every once in a while just to really thermally torture it

[crispy_tofu]

Just putting this here:

[wraper]

This is not even a reflow. Without any flux presence nothing can even possibly reflow. You can call it 'reheating"at best.

[AlxDroidDev]

I, personally, prefer to buy GPUs with lifetime warranty (like the 2 EVGA GTX480 Superclocked I have) than have them undergo such dubious procedure.

I've had an old Asus Ti4600 (quite fancy for the time, but I bought on the grey market, without any warranty, and it died on my after 8 months) GPU reflown, but it only lasted for 6 months or so.

[oldway]

The stupidity of some is such that a friend of mine said he was reballing GPU with a hair dryer and it worked. 

Laughing, I gave him some solder and asked him to melt it with his hair dryer.

[theelectronicegg.com]

i understand all the different problems...

but for some home user / gamer... it might give them another 3 month before needing to buy the next card...

just don't sell it as a repair service.

[oldway]

...just don't sell it as a repair service.

And don't say you fixed it !!!!!!!!

[samnmax]

I'd also like to contribute my experiences with BGA "repairs".

I started with a Compaq Presario with a bad Radeon 9000. After attacking it with a 1000W heat gun from a distance, it booted, but with artifacts and crashed after a while. I tried to repeat it, but I got too enthusiastic and blew the chip up with too much heat. Luckily the GPU was replaceable so I got a new one.

Later on, I got a regulated heat gun that I used successfully to bring many laptops to (temporary?) life.
For most of them, I don't know how long they lasted, as they were from friends of friends. However, my uncles are using a Core 2 laptop with an ATI X1300 that has been reflown twice, first time lasted 2 years, let's see how long does the second pass work. I also fixed a BGA in a Seagate 2.5" hard drive, I use it regularly for non-important stuff.

Afterwards, I got a cheapass handheld hot air gun for SMD and some flux paste from Dealextreme, and I have to say that for the moment I only had success with them.

Recently I reflowed a GeForce 8600GT on a 17" Vaio. In this case I'm pretty sure the problem is in the chip bumps but hey, it worked, and allowed me to quickly recover the data off the laptop's RAID0 array. And a friend has a huge spare laptop to watch movies or so until it fails again (gaming forbidden  ).

[Skimask]

So far I have fixed:
R9 270x
Three iMac video cards
HD4850x2

A sample size of 5.

Going by that logic, I fixed a Donkey Kong 3 arcade cabinet by reattaching a wire to the V-hold pot and re-cap'ing the monitor.
That means I also have a 100% success rate.

[paulhm81]

I usualy only heat up a gpu in laptops to confirm that the chip is the problem and if it works I outsource the gpu replacemet to a guy who does that for a living. A few years ago the reballing and reflowing business only gave me head headaches as it didn't work. Of the top of my head I would say that 90% of the laptops that had a reball came back after a few months. I refuse to reball or reflow them and the ones that get their gpu replaced are above 90% working 1 year later and I'm talking about 40-50 laptops!

[gnasirator]

My experience:
2008 Macbook pro with nVidia GPU issue
- First try in IR-oven @ 220° - 7 min cycle: worked for about 1 week
- Second try in IR-oven @ 250° - 7 min cycle: worked for maybe two weeks
- third try in IR-oven @ 260° - 7 min cycle: same
- Last try in IR-oven @ 290° (max)° - 2x7 min cycle: still works (2 months ago)

Note: I shielded the whole PCB with aluminium foil and built a small square shield for the die as well during the last try.
My approach was, either it works or I get a new laptop which I would've needed then anyway. I'm glad it worked as I'd rather save the money until a good eGPU solution using Thunderbolt 3 is available.

[iamdarkyoshi]

This is not even a reflow. Without any flux presence nothing can even possibly reflow. You can call it 'reheating"at best.

Fair enough, whatever it is, it seems to be working so far. I am aware of the issues though and do NOT sell them as repaired, but I do sell them as "I will either attempt to repair or replace it within a year of sale if it fails"

[Bud]

I have a GTX460 that failed after 4 years. It still has signal output on all connectors but only in VGA mode with the generic driver. As soon as i install the GPU driver (windows 7), i get a blue screen on reboot. It gets as far as "Loading Windows" message, then blue screen with "failed to reset the video card" sort of message. In VGA mode it shows in Device Manager as GTX460, so perhaps that means the system recognizes the GPU. I can also read the BIOS using nvflash utility. So i am not sure where to look or what to try to reheat/reflow, the GPU or memory ( do not want to torture the entire board in a IR oven, the board has plastic connectors). And this is not a power supply, computer or OS issue, tried it on a different computer with a monster power supply, also on a brand new Win 7 installation. Did anyone have experience with this sort of symptoms?

[iamdarkyoshi]

I have a GTX460 that failed after 4 years. It still has signal output on all connectors but only in VGA mode with the generic driver. As soon as i install the GPU driver (windows 7), i get a blue screen on reboot. It gets as far as "Loading Windows" message, then blue screen with "failed to reset the video card" sort of message. In VGA mode it shows in Device Manager as GTX460, so perhaps that means the system recognizes the GPU. I can also read the BIOS using nvflash utility. So i am not sure where to look or what to try to reheat/reflow, the GPU or memory ( do not want to torture the entire board in a IR oven, the board has plastic connectors). And this is not a power supply, computer or OS issue, tried it on a different computer with a monster power supply, also on a brand new Win 7 installation. Did anyone have experience with this sort of symptoms?

What brand is it? It might be a reflow issue, but again, I cannot be sure. I look at it like "If its already fucked, there is no harm fucking it up further"

When you do a reflow, remove the little socket shell on the fan header, and the rest of the board should be able to handle reflow temps. The plastic used in them is supposed to handle the temps without melting.

Those fermi GPUs did run quite warm, so there is a chance that the thermal cycling is what caused the issue... Worth a shot I guess to attempt a "reflow"

[Bud]

Oh yes, the card has taken a lot of abuse from my teen son, several hours a day for 4 years in a row, Steam games and stuff, you can imagine. I think it did pretty well. It is an EVGA brand, Nvidia GPU.
I think i will try measuring the current consumption for any abnormal behaviour before going for a reflow. When i get the blue screen i can hear the HDD clicking, not sure if that is because the card may be short circuiting  the power bus when the driver kicks in, in which case may be a problem with one of the onboard power regulators.

[iamdarkyoshi]

Oh yes, the card has taken a lot of abuse from my teen son, several hours a day for 4 years in a row, Steam games and stuff, you can imagine. I think it did pretty well. It is an EVGA brand, Nvidia GPU.
I think i will try measuring the current consumption for any abnormal behaviour before going for a reflow. When i get the blue screen i can hear the HDD clicking, not sure if that is because the card may be short circuiting  the power bus when the driver kicks in, in which case may be a problem with one of the onboard power regulators.

What happens if you use onboard gfx for the monitor and just leave the GPU in the system? Can you get drivers installed when the card is not being used to output video?

[Ian.M]

Possible bad caps?

[Bud]

I can leave it safely in the computer and use the computer's Intel HD graphics, no problems.
And I can install the driver while outputting generic VGA video through the card , not a problem, (still on VGA output after install before reboot), it is only when i reboot after installing the driver i get the blue screen. I then can boot in safe mode (again the card being in VGA generic mode) and disable or uninstall the driver.
Did not try installing the driver while on Intel HD graphics but i recall i cant do that because in the computer BIOS i have to set either onboard or PCI graphics to be used.