Fixing an overheating Radeon chip possible?

[beeblebrox]

Folks,
since yesterday I am having a problem with my T40p. The graphics chip!

First, to give some facts:
It was never ever lifted with one hand. In 4 years it was 99.9% sitting on my desk without any movement. Now I added a large screen connected to a docking station and did a lot of Google Earthing and iTunes Cover Flow. It was about the first time ever I have stressed the chip. Usually it does only powerpointing and wording.

After about 1h I got a lot of artefacts, which I know from this forum are typical for a so called flexed board. However the notebook was never ever flexed. Temperature of the Radeon never exceeded 60°C.
After reboot, Windows was fine. To check, I moved the notebook by hand (even with one hand!). No problem.
Started PC-Doctor and I only had half the graphics memory left. I assumed the on-board RAM was damaged in the docking by the heat (conclusion of reading some forum articles).
After lunch I started the notebook again and the whole memory was here again! 64MB.
But PC-Doctor brought errors in the second 32MB part and then a lot of artefacts in the burn-in phase.
I then have let the notebook cool down and used NHC to downclock the GPU and graphics memory by 10%. Now, PC-Doctor went through without a problem!
Half the memory does always work, the other half only when cold or downclocked by 10%.

Do I have a heat problem here? Maybe resetting the heat pipe? Putting on some new pad? Has anyone had a similar problem and found a solution? Or is the motherboard fried by 60°C?

[sjthinkpader]

I personally don't believe flexing apply enough force to break a solder joint. We had guys around the office who holds their T40 by the front with the LCD cover open and this machine is still working five years later.

In contrast, my T41 was never held by the front and its GPU crapped out 2 months after the warranty expired.

I think temperature cycling is the main culprit in breaking the solder ball joints. In the 3 years I had that T41, it was turned on and off twice daily. I venture to guess that if a T40 was never turned off, it probably would be working for a long time.

Temperature cycling is a common solder joint reliability test method.

http://www.sciencedirect.com/science?_o ... 7c6510aee7

[beeblebrox]

That's an interesting question about flexing.
I mean, there are about 100.000 people clicking on that topic in various thinkpad forums around the globe.

What I find interesting is that nearly all problems occurred around year 2002-2004. Regardless of brand or type: I think about the usual suspects: T4x graphics, T3x ram slot, Sony GRX ram slot, HP graphics.

These are parts, that are under constant mechanical stress, even negeligible like the spring ram slot.

It seems to me it is a common manufacturing problem in the soldering line, maybe the shift from common, but well-known, soldering tin to lead-free tin might be the reason.

BTW: I successfully recovered my Thinkpad. I let it cool down, restarted and turned off FanControl. Now the chip is passing all burn-in tests again without problems.

I am sure, that the constant downloading on wifi was the reason, because the wifi card is tremendously hot and just next to the graphics chip. However, TPFanControl still monitored only 60°C, while ATI's on-board ram might have been much higher. In addition running a closed notebook in a docking station under full program load is the easiest way to kill it.
I read in another article, that someone was playing doom with a docked T42p. After 2h the graphics chip was totaled and burned to death. Luckily the warranty was still in place.

[sjthinkpader]

Yes, you are right. 2002-2004 was roughly the time when RoHS soldering process came into play. Lead containing solder and processes had been in use for a long time and there were a lot of experience with that. BGAs were introduced in mid-nineties.

Before multi-chip modules (GPU+two RAM), GPUs had separate RAM chips as with the 7500 in A31 or on chip RAM as with the single chip NeoMagic in 600 series. ATI elected to use Multi-chip modules for flexibility in RAM sizes and it is clear they had not done a proper engineering job.

By the way, RoHS solder reflow temperature are about 245 deg C and non-RoHS reflow temperature were lower, around 220 deg C.

[beeblebrox]

Just to let you know. My FireGL 9000 developed the problem with the missing half of video ram after a while, after it had scrambled video even during booting.
I thought that was it, the board was flexed without reason.

I read the other article with the T43 and the copper plate. So I took off the fan assembly, actually for the very first time.
It immediately appeared to me that the pink sheet impossibly could cover the whole Radeon chip, because there was no pressure on it and it could easily be lifted.

I creamed the CPU and GPU with white silicon paste and put back the fan assembly.

Problem solved: temperatures never go beyond 60°C under full load.
DURING IDLING THE FAN EVEN SHUTS OFF AFTER A SHORT TIME!
Even in BIOS mode. I never had this in the past 3 years!!!

So it seems that the pink heat conductor loses contact after a few years. The subsequent garbled screen is then the effect of overheated on-board RAM and not a flexing phenomenon.
So in case anyone has similar problems and figures out that a little pressure on the GPU is all that is needed, it just might be a lose heat contact and not a damaged solder point.

Just my 3 cts (inflation adjusted)..

[aaa]

I've also done the mentioned paste trick, it indeed shut the fan up most of the time.