4
\$\begingroup\$

I've had my ENGTX480/2DI/1536MD5 GPU for 3 and half years now.

Over the past 2 months it started causing trouble. At first, it started seldom displaying big zones with

weird color, but simply turning my TV off and on again and it got back ok. Then it got worse. Under windows 7 x64, 340.52 driver, I got wierd pixalated screens, then screen would go black and the pop up, "driver stopped responding and was restarted" or something like that. Soon followed by 116 BSOD. It got more and more often. And completely random. Usually starts on windows, but once it happened, also occurs on boot screen. I can still use it, though I can never rely on it. It got unstable.

I have double checked temperatures, they were all under 60 C, under 69 worst cases. Auto cooling was working perfectly fine. I switched PC&C 910 W for a Corsair 750W, and nothing changed.

I've cleaned the gold fingers at first, checked for poor contact. Then dismounted the video card and got it all clean - no warranty anymore. The same goes on.

I have changed the video card for an old one and it works great, so I am sure motherboard is fine.

It seems to me a classical case of soldering issues over time. A reflow seems to be the last resource. I have checked over the net and the vast majority of people who somehow "baked" their GPUs and MBs got them working or living again, but for just some weeks/months. It seems clear to me that baking helped by changing the state of the solder, but done under the wrong profile!

Like many of you guys, I am a licensed electronic engineer and I have means to use a fully programmable oven. I mean an oven that supports programmable temperatures ramps and times. However, I can't find any temps and time profiles that are safe, indicated by any manufacturer or sustained by engineering professionals at all. Just found "i tried this", "do that", but no real references.

Does anybody know the correct complete profile for reflowing a GPU??

\$\endgroup\$
1
  • \$\begingroup\$ Well, it's not unheard of GPUs themselves to suffer damage after repeated thermal cycles, a famous case being Nvidia's G84 and G86--amply documented in the press if you google them. So you can't be certain it's just the board. By the way, since you asked this several months ago, did you solve your problem this way (reflow)? \$\endgroup\$ Commented Jan 2, 2015 at 16:25

4 Answers 4

4
\$\begingroup\$

A typical soldering profile for a semiconductor is as follows (this one is from an ST datasheet).

enter image description here

In addition to this, you must bake the part for 24 hours under nitrogen atmosphere.

That's for one particular part. To guarantee safety you must look up every part on the board and make sure that the profile does not violate any limits for each component (perhaps the part was originally attached manually after reflow). Since you also want this to work, you have to ensure the profile is adequate for whatever solder was originally used. A conveyorized oven with multiple zones is usually used to get the required short residence time at the peak temperature.

In practice this is non-trivial, and for an attempted repair of something relatively cheap and where reliability is of little importance it is probably best to have at it with a nominal profile, perhaps shielding parts like through hole connectors and electrolytic capacitors with aluminum foil if they are far from the suspect areas. Batteries should be removed.

\$\endgroup\$
6
  • \$\begingroup\$ Spehro, tk you very much. You are right, I don't have a conveyorized oven, much less a reflow machine like the Euro ones. I also believe a programmable oven will do. What I don't know for sure is the optimal or best numbers. For ex., avarage ramp-up rate is 3ºC/sec max, but is the recommended? Considering this curve is valid for all components, should I look for the average of preheat temps min and max (175) or look for getting close to 200? Also for time, considering I can input it and also considering the thermal inertia fo the system, there is a great difference between 60-150 secs. \$\endgroup\$
    – tfm
    Commented Sep 6, 2014 at 15:53
  • 1
    \$\begingroup\$ For rework I've just used my oven's default lead-free profile. It's a seven minute cycle with a peak of 250°C. It spends less than 2 minutes above 200°C. \$\endgroup\$ Commented Sep 6, 2014 at 16:17
  • \$\begingroup\$ I see. After reading this link, I would like to try the nitrogen atmosphere, I may dig access to it in University. What is the proper procedure? I mean, I have cleaned the whole PCB and components with isopropyl alcohol. What comes next? applying no clean flux before baking it under nitrogen? and how is that, how do I "bake" it, what is the profile? then just proceed to "normal" oven and bake again under the standard profile? I would really like to make a good reflow on this gpu with what I have! \$\endgroup\$
    – tfm
    Commented Sep 6, 2014 at 17:40
  • \$\begingroup\$ Now there is one more thing, please take a look at the pic! The 4945N and the 4935N are 35A and 93A mosfets, respectively.121F seems to me power resistors, although I could finf no reference or datasheet for them. Then there are the 8 RAMDAC chips and the CPU itself. There are the high power parts I consider important to reflow. Agree? I can't see how I can shield the electrolytic capacitors close to the components shown! See here! There is no way to wrap alum. foil around the whole pcb! Ideias for shielding? \$\endgroup\$
    – tfm
    Commented Sep 6, 2014 at 19:25
  • \$\begingroup\$ I would mostly be worried about the BGA(s) getting enough heat, that's 99.9% where the problem lies. \$\endgroup\$ Commented Sep 6, 2014 at 20:28
4
\$\begingroup\$

The main problem in using an oven is that if you have components on the bottom side, these may fall off.

The last time I had a problem like this, I used a hot air gun, first heating up the IC and PCB from a distance, and then going nearer for a short time.

Worked fine.

\$\endgroup\$
3
\$\begingroup\$

Aside from the answers concerning the true reflow cycle and problems associated with that process step, you should be aware that you aren't dealing with new components from a sealed bag.

All semiconductor housings except the ceramic ones or some special polymer types with certification for space flight will accumulate humidity inside the polymer matrix. If you just put it in your reflow oven programmed with the exact profile the outcome might be as catastrophic as doing it on a camp fire.

Sudden expansion of embedded humidity might burst open the package of your GPU and other components of your device. Chances are that you won't notice a crack in the housing and functionality isn't even affected but I recommend to install counter measures.

The problem is well known in primary manufacturing as well as in professional rework suppliers. The common method is again to bake the assembly at 80°C for 24 to 48 hours. If you ever bought a reel of components in original packaging you may have noticed a card with humidity indicators. It usually says exactly the same for even brand new parts. If the humidity indicator shows excess humidity in the bag, you should bake it for a recommended time and at a recommended temperature before using it in assembly.

Such a baking process will greatly reduce failures for a complete assembly, as humidity even can build up in the FR4 base material. I have seen PCBs with complete delamination of prepregs due to improper baking and humidity control in assembly processes.

\$\endgroup\$
0
\$\begingroup\$

If anyone is still interested in this topic, this is a curve from PowerColor factory (from video by der8auer) :) enter image description here

\$\endgroup\$
2
  • 1
    \$\begingroup\$ Can you explain the flow profile? \$\endgroup\$
    – Voltage Spike
    Commented Sep 2 at 14:07
  • \$\begingroup\$ I took a closer look and it looks like, that the graph is measuring of a temperature on the board during reflow process. So it's maybe not exactly temp curve for the oven, but it should give at least some info about the temperature. Otherwise I'm not familiar with such technology, sorry \$\endgroup\$
    – Daniel
    Commented Sep 3 at 7:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.