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

[rx8pilot]

I use temperature indicating crayons to help me heat a pcb with a reasonable profile.

That's a good idea. What temperature(s) do you use? I just checked Amazon... those damn things are expensive! Some of them are listed for over $125?! For a crayon??

McMaster Carr $13.64/ea

575F (a little over 300C) is what I try not to exceed at the PCB.
300F (150C) is a good guess where flux activates. This is about the 60-second countdown where you should be flowing.
437F (225C) this is just before the flow begins with Pb solder

I generally just cut off some tiny pieces and put them on the board. They were helpful to understand my batch oven profiles as well as for re-work and 1-off hand placed prototypes.

[Russ314]

My 212*C Markal Thermomelt stick arrived in the mail today. It is, without a doubt, the worst item I have ever purchased on Amazon. I've spent the past hour trying to make it work. I've literally tried everything I can think of to make it melt, but all it does is out-gas (spew white vapor), or char and turn black, or crumble into sawdust-like pieces and blow away. Out of frustration, I tried melting it with a torch, but it instantly caught fire and burned. The worst part of the ordeal is that the vapor it emits is a skin irritant, so my arms currently feel like I was grinding pink fiberglass insulation into them.

In a last-ditch effort to make it melt, I cranked up my heat gun to 1,000*F, and ran it for 9 minutes. The thermocouple reached 247*C, but the Thermo-"melt" just sat there as if I was trying to melt drywall.

 

[rx8pilot]

Wow! That is rotten.

[Ian.M]

I'd complain about that.  It sounds like it could be fake.

Sn-Ag-Cu ternary eutectic lead free solder melts at 217 deg C, other Sn-Ag-Cu near-eutectic compositions melt within a few degrees of that.

Cut or scrape the Thermomelt stick to expose a fresh surface and try melting a piece of Sn-Ag-Cu solder against it with a freshly cleaned and tinned bit (using the same solder), you should see the surface start to fuse just before the solder melts.

N.B. Thermomelt sticks are intended to be 'chalked' on a lightly roughened surface before that surface is heated, and may not give a useful indication if used in bulk or if directly heated. See http://www.ksdistribution.com.sg/ksd/wp-content/uploads/pdfs/atoes_products/E2.%20Laco%20Markal/Thermomelt.pdf

[Russ314]

Thermomelt sticks ... may not give a useful indication if used in bulk or if directly heated.

*sigh* That's exactly the conclusion I arrived at, after dicking with it for 6 straight hours yesterday. I assumed it was like wax, and would melt/solidify around a set temperature. In reality, it's a binary material, consisting of flux and powdered chalk. The only time it actually melts is if the mixture is thermally shocked, and changes temperature instantly. If you try to slowly heat it to melting temperature, the flux component evaporates, leaving behind the unaltered chalk. You can see in one of the pics where I broke a piece open, and you can clearly see how the "spent" outer shell protected the inside material.

Also note the white powdery condensate. That is the flux residue. Even when the stick is used "properly," the flux instantaneously evaporates into a giant cloud of white vapor, which engulfs your face and burns your eyes and sinuses similarly to hydrogen chloride (a.k.a. acid vapor). It also condenses on your skin, covering your body with a thin layer of plastic-y material. If you've ever spray-painted, you'll know what it feels like when paint mist sticks to your arm hairs. It creates a sticky, uncomfortable feeling. That's what this flux does. I had to shower twice yesterday (and wash my face 5 times) because I kept getting coated every time I messed with it, and I refused to stop messing with it. I wouldn't attempt to use this stuff without a bionic face shield (although that didn't protect my face entirely).

Also note the pic that shows a thousand crumbles - that's what happens if you try to use the stick like a crayon. If the metal isn't quite hot enough, it simply turns the stick to... it's like when your laundry or dish detergent gets damp, and it forms a solid clump - but the clump easily explodes back into granules? That might be a bit esoteric, but that's what this stick does. Most of the smaller granules blew away, so the ones in the pic are just the larger pieces that remained in the area.

One of the few times I got it to melt is shown in the pic with the zinc-plated bolt. I heated the bolt with a blowtorch for a while, then melted the stick on it. It successfully melted, released a cloud of vapor, then turned into a chalky layer that very closely resembles dried toothpaste. It's worth noting that this method only seems to work on steel. Heating copper or aluminum with a torch, and then rubbing the stick on it, only caused the stick to crumble. The only other time I got the stick to melt properly was when I rubbed it along the stainless steel thermocouple probe (seen in the pics). It wouldn't, however, melt on the adjacent aluminum plate (which was much hotter than the steel probe).


Btw, I don't own (nor have I ever used) lead-free solder. A few years ago, after exhaustive research, I decided to use lead-based solder in all my products, because I concluded that it's better than lead-free in every way (including environmental and health safety concerns). If anyone has a few hours to spare, this article is a good read: https://nepp.nasa.gov/whisker/reference/tech_papers/2011-kostic-pb-free.pdf. I tend to think of NASA as a reliable source.


So here's the million-dollar question: is Tempilstik any better?

[debininja]

I originally thought this trick would work, but unfortunately, it does not, at least, not permanently. The fix is only temporary. I bought a used GTX580 off eBay for $60 (for Vegas CUDA rendering. Apparently the Fermi cards were the last to support it). 6 months afterwards, it stopped working.

I set my hot air gun to 240C and directed the airflow on and around the GPU for close to 20 minutes, popped it back in and...it worked! I was thrilled. Unfortunately, it stopped working again after 2 months. So I did the same thing, but this time I set it to 250C and ran it over 5 minutes, alternating between cooling and re-heating, for 20 minutes. Popped it back and...and it worked! But it stopped working again after 2 weeks, and that's when I got fed up and chopped it off the PCB.

Now I plan on using the original, sexy looking beefy heatsink on the assembly, strap on a bunch of power mosfets, and build an Arduino powered constant current, constant resistance, and constant power load assembly. (Basically, it'll look like a real GTX580, but with the guts missing). Add some CCFL lighting all around it and it'll look really slick.

[Rasz]

I originally thought this trick would work, but unfortunately, it does not, at least, not permanently.

of course it doesnt, whole thread is pretty stupid, even nVidia admitted the chips are internally defective 

[Armadillo]

So here's the million-dollar question: is Tempilstik any better?

In the photo, the tube thing, is that the thermo-sensor? I think the thermo response will be so laggy in such a heavy thermal mass assembly over a very doubtful air/heat distribution system in terms of even temperature distribution by the look of it.
Forget about Nyquist stability criterion with such laggy system, I supposed its hand controlled.

It's just my view base from my first impression, but I am sure you have your design theory and reasons in doing those.

[Oogway]

2. This is critical - BEND THE SPRINGS on top of the thermal cover/pad that rests on top of the GPU. Bend the springs upward, so that whatever you bolt on top will press downward with maximum force. On my laptop, there is a large cast metal heatsink (for the GPU RAM) that bolts on top. By bending the springs upward, I made the cast metal part push the GPU pad aggressively against the GPU. This is a big part of the permanency of the fix. I don't know whether pushing on the GPU forces out more thermal paste, which closes the gap, which increases heat transfer from the GPU, or whether the force squeezes the GPU package and closes any internal gaps/cracks. Those are both ideas I hadn't thought of until reading this thread. To be honest, my reason for doing it was that I wanted the GPU to be held in place for whenever the solder melted again. After all, the reason for the GPU failure was gravity. When the GPU overheated, it would DE-solder if the laptop was right-side up, and RE-solder if the laptop was upside-down. So I figured the added spring pressure would counter the force of gravity and hold the chip against the board.

Wondering if this reflow with added pressure method would work with a GTX 1070. Bending the springs back to add pressure and keep the underfill in place a bit better. Could you expand on this a bit more?

[wraper]

I originally thought this trick would work, but unfortunately, it does not, at least, not permanently.

of course it doesnt, whole thread is pretty stupid, even nVidia admitted the chips are internally defective 

Especially considering it never reached solder melting temperature.

[Rasz]

Wondering if this reflow with added pressure method would work with a GTX 1070. Bending the springs back to add pressure and keep the underfill in place a bit better. Could you expand on this a bit more?

to completely destroy your card? yes
this thread is a scam

[Oogway]

To be fair though...it's already destroyed. I've seen your posts and I know you know what you're talking about but the only alternative is throwing it in the garbage. Unless you know of another way to fix the underfill issue?

[Rasz]

Pascal (GTX 1070) doesnt have underfill issues (that I or anyone else in the industry heard of)
make new topic about your card and describe problem in full, or return it for warranty replacement

[Oogway]

Pascal (GTX 1070) doesnt have underfill issues (that I or anyone else in the industry heard of)
make new topic about your card and describe problem in full, or return it for warranty replacement

Okay will do thanks Rasz

Sent from my XT1635-02 using Tapatalk

[Oogway]

New topic on my issue posted here if anyone is interested in following.

https://www.eevblog.com/forum/repair/gtx-1070-no-video-underfill-issue/

[Russ314]

So here's the million-dollar question: is Tempilstik any better?

In the photo, the tube thing, is that the thermo-sensor? I think the thermo response will be so laggy in such a heavy thermal mass assembly over a very doubtful air/heat distribution system in terms of even temperature distribution by the look of it.
Forget about Nyquist stability criterion with such laggy system, I supposed its hand controlled.

It's just my view base from my first impression, but I am sure you have your design theory and reasons in doing those.

You don't SOUND sure... At any rate, the thermal mass of the probe is matched to the thermal mass of the populated boards. A fast-response probe (e.g. thermistor) would simply tell me the temperature of the air, which I already know (from the heat-gun readout).

Convective heat transfer is a bitch. I was actually hoping to fine-tune the temperature probe distance/placement so that it showed exactly 183*C when my boards began to reflow, but my "only melts at one temperature" thermal stick decided to "melt" across a huge, unpredictable range of temps.

[Russ314]

even nVidia admitted the chips are internally defective 

No one is denying that (despite your implication).

Wondering if this reflow with added pressure method would work with a GTX 1070. Bending the springs back to add pressure and keep the underfill in place a bit better. Could you expand on this a bit more?

I'll do you one better - I made you a picture. See attached.

Especially considering it never reached solder melting temperature.

If you use the nVidia System Monitor to track GPU temps, you'll find that a faulty GPU will start to fail when the temperature is only ~75*C. If you use the "wrap your laptop in a blanket and flip it upside down" method to repair it, you'll find that the GPU heals itself at around the same temp. We can therefore conclude that the underfill epoxy begins to soften at a surprisingly low temperature. This method of fix will only be temporary though, because the solder balls beneath the die are still cracked, which means the circuit will fail again as soon as the GPU temperature exceeds ~75*C.

to completely destroy your card? yes
this thread is a scam

You seem like a very, very smart person; I had better not argue with you.

[Russ314]

See attached. I just screen-capped that from my system overview page. I've been using this laptop every day for the past 10 years.

I also checked my old maintenance notes, and can tell you definitively that the date of my last/final GPU fix was 7/16/2016. So its been exactly 11 months. Not as long as I remembered, but still a long time.

My GPU is currently 143*F, which is 5*F hotter than usual, but I'm doing a lot right now. I frequently watch Youtube vids and have no trouble with the GPU overheating. Compare that to before the "spring" fix, when a 5-minute Youtube vid would cause the GPU to overheat and fail. Even after a fresh heat-gun reflow, Youtube vids would always cause the GPU to overheat and fail.

I can't remember the last time the GPU/CPU fan spun up higher than its lowest state. Even doing frequent CAD work doesn't cause the fan to wind up.

[Armadillo]

You don't SOUND sure...

No dude!, is just from my culture to be polite and allow you a face saving way out.

If anyone wants to figure out the exact heat-gun conversion ratio, just pour a small amount of distilled water into a test tube, and try to boil it. Common sense will have you start with temps near ~220*F, but you'll quickly realize that 220*F is not even hot enough to burn your hand. In order to make the water boil, you'll have to get the gun to around 400*F. Since water boils at 207*F in my area (Los Angeles), that means the conversion factor is about 2:1. In other words, to create a 420*F reflow profile, you'll have to set the gun to 420*2 = 840*F.

Your theory is so much flawed.  Even a low grade student will know that the water specific density and specific heat capacity are so much different from solder. You are at least 192 times OUT! You ought to start from the basic.

Convective heat transfer is a bitch.

The surface area of your probe is so huge and here you are bitching on convective heat transfer.

At any rate, No, you can't convince anyone with those...

[WastelandTek]

Storm is coming...

gawd...you called it

I just don't get all the vitriol.  OK, sure, if someone is selling baking as a service, or baking something and immediately selling it that is messed up, granted.  But I have had lots of success with my own equipment over the years, hell, there is an old 8800 over there /glances across room/ been faithfully flogging SETI at home for the last two years on a bake.

What the people who get their panties in such a bunch need to remember is that these cards were BROKEN, UTTERLY WORTHLESS E WASTE.  Whatever the mechanism, I fail to see the problem with squeezing weeks, months, or indeed YEARS of service out of an otherwise useless device.  Does the solder actually reflow?  Is it the chip bumps? I could give a rip as long as the dammed thing boots and does what I want, which they frequently do, and if not, hey, IT WAS BROKEN ANYWAY.

[Lolucoca]

Had to reball some RAM chips last week, wasn't too bad since it was only a ~30 ball BGA...

[flexy123]

I just tested the heat gun in a formal experiment. I'm too tired to post a full write-up, so here's a summary of my lab notes:

Heat Gun: Steinel HL2010E (not recommended)
Temperature Setting: 800*F
Airflow setting: LOW
Nozzle: 75mm "Spreader Nozzle"
Distance from nozzle to work surface: 1.0"
Method of temperature measurement: thermocouple (i.e. a kitchen thermometer, with the probe laid flat on the work surface directly below the nozzle output)

TIME     TEMP (*C)
T=0         24.9
T+30       86.6   <-- The gun reached 800*F at exactly 30 secs
T+60      142.8
T+90      169.1
T+2min   178.6
T+2.5     187.3
T+3min   193.2
T+3.5      205.9
T+4min   210.8
T+4.5     212.6
T+5min   213.9
T+5.5      215.1
T+6min   216.7
T+6.5     218.3
T+7min   218.5
T+7.5     222.1
T+8min   220.3
T+8.5     219.6
T+9min   220.4
T+10min 219.1

As you can see, it's a pretty good reflow profile. Starts to reflow after 2 to 2.5 minutes, and peaks at ~210*C after 4 minutes. Leaving it on for a full 10 mins only adds a few more degrees beyond that. At the 10 min mark, I switched the gun to "cool down" mode, and it took 2 mins 10 secs to cool all the way back to 120*F.

I would like to amend my previous post btw; 850*F is a bit too high. I tried it and it melted the white plastic SMD connectors on the board before it melted the solder. Dropping the temp to 800*F worked perfectly, and the plastic never melted/singed/discolored again.

Hopefully this clears up once for all the myth about cooking/scorching/incinerating/vaporizing IC's with 450*C heat guns.

I apologize that I am quoting this in length, but it's needed for context.

So, I am experimenting with various "ghetto reflow" techniques recently, including semi reflow with a heat gun, and this post confuses me.

(My heatgun is similar to yours, it's a cheap 2000W one with two settings, low = 350C, and high is 550C)

I can't see why you need more than 3 minutes (!!) to even reach 200C. Further above you say even that you need to set your heat gun  to 800F (425 deg C.) to be even able to melt anything. (Suggesting that anything below 800F wouldn't work). This contradicts all my own observations.

But then again, you may well be right, as you say somewhere "convection heat transfer is a bitch".

I used a thermocouple right above the chip, and it showed 210C, but of course this shows only the temperature of the hot air coming from the gun and is not an indication whether the PCB itself or whatever part actually has this temperature for "reflowing". Saying: It is *DARN DIFFICULT* to even assess the actual temperatures you're getting when you try a "reflow" with a heat gun. So even with a thermocouple to measure, I am still there where I was before.

If what you say is true that you need to set your H.G. to 450 deg C., then basically this means I can try "reflowing" at low (350 deg) until I am blue in the face, it won't ever reach the temp to actually do anything. (Now not considering that most of these "reflows" are in reality only fixing things by softening the underfill, which seems to be the actual problem in many cases)

Furthermore, I am confused why you use this odd spreader nozzle which basically spreads out into a thin line, so no surprise that moving the H.G. just a millimeter messes things up. Likewise, I'd consider this nozzle extremely impracticable if not impossible to "reflow", say a GPU chip. (How would you do this with this type of nozzle?).

Add: I see you're using a capillary thermo probe there, like those which are used in ovens. I am not a pro by a long-shot, but I wonder how accurately this would reflect the actual temps. I mean it's possible this is also not accurate (not better than the thermocouple) and/or that the setup is very "laggy" (as someone mentioned). It's just that the temps in your list and the times there to me make not much sense.

If I take what you say there as true, it would mean reflowing with a heat gun, if at all, would need MUCH higher temps, and a MUCH longer time, like 7-8 minutes at least until something even melts?

Regards

[janoc]

If what you say is true that you need to set your H.G. to 450 deg C., then basically this means I can try "reflowing" at low (350 deg) until I am blue in the face, it won't ever reach the temp to actually do anything. (Now not considering that most of these "reflows" are in reality only fixing things by softening the underfill, which seems to be the actual problem in many cases)

...

If I take what you say there as true, it would mean reflowing with a heat gun, if at all, would need MUCH higher temps, and a MUCH longer time, like 7-8 minutes at least until something even melts?

I wonder that you even bother with these guys ... This nonsense is like religion - or audiophoolery.

At the temperatures in his "reflow profile" he wouldn't even have melted solder at the peak temperature of 210 C. Common lead-free solders don't even start to melt below 217 C or so and some mixtures often don't completely melt below 240 C - that is with perfect heating (like in an oven), not when you have tons of losses all over the place like with hot air. Even with a soldering iron lead-free is difficult to solder with the tip temperature set below 330-360 C.

Standard oven reflow profiles go up to 260 C for the actual soldering phase.

https://www.microsemi.com/document-portal/doc_view/131105-solder-reflow-leadfree


You have nailed it that all he is doing is melting the underfill (and cooking the board for minutes). Which, sometimes, "fixes" the problem for a while because it holds the cracked internal solder balls on the flip-chip carrier together, restoring electrical contact. Of course, it will work only until the next few thermal cycles cause it to move out of position again ... So much for such "reflow" repair.

[WastelandTek]

Getting cold here in the Northern hemisphere, recently went around and fired up all the "heaters"

Yep, that old 8800 mentioned above is still going strong, 3.5 years now.  Good thing I listened to the experts and binned it back in...oh wait...

[jacan]

Getting cold here in the Northern hemisphere, recently went around and fired up all the "heaters"

Yep, that old 8800 mentioned above is still going strong, 3.5 years now.  Good thing I listened to the experts and binned it back in...oh wait...

nice, can you tell me your average temps on the gpu? (idle and while gaming)