Second GPU overheating on a 5970

[xanadu]

I picked up a 5970 today and after dropping it into a rig it runs great!  Except for the fact that the second GPU runs about +23C hotter than number one.

Running BAMT, 820/300, getting 379M/hash from both GPUs.  GPU 1 is running at 77.5C and GPU 2 is at 101.5C

Obviously I didn't leave it running at those temps  

The card is clean and the fan is running properly, even if I run it up to 90% the temps are reported as very high.  I'm thinking maybe they did a bad job applying the thermal compound on the second GPU? Are there any major tricks to removing and re-applying the heatsink compound?  I have both generic "white paste' and silver based CPU paste to use, which would be better in this application?

Anyone ever had a bad thermal sensor on a GPU?  The back of the card measures the same temperature as a properly behaving 5970 with my IR thermometer.

Or, is it just time to return the card and call it a day?

Thanks for any advice,
-X

[1714949723]

1714949723

[1714949723]

1714949723

[1714949723]

1714949723

[1714949723]

1714949723

[1714949723]

1714949723

[coretechs]

I had the exact same thing happen the first time I replaced the padding/paste on one of my cards.  I'm sure your card is fine.  It can easily run that hot if the thermal paste isn't done correctly.  Do you know if the thermal pads were replaced?  If so, make sure they used the correct thickness, otherwise you may not have enough contact to the GPUs for effective cooling.  Did you buy the card brand new or used?  Who is "they"?

For the GPUs, I'd recommend using a high quality thermal paste like AS5 (or some newer fancier one) and applying it in an X pattern.  There are some other threads documenting 5970 maintenance so you may want to search for those as well.  

[DeathAndTaxes]

yeah the 2nd GPU tends to run a "little bit" hotter (5C to maybe 10C) but 25C+ and 100C at a decent overclock is indicating poor contact with heatsink.

Don't use silver stuff.  You got exposed traces.  If you put too much on and short a trace well the card does make a good $400 doorstop.

If all you have is white paste that is better than the risk of conductive silver.  In future buy a non-conductive TIM for GPU like MX2 or IC Diamond.

You need a lot less than you think.  It should be a tiny layer.  TIM is more conductive than air but it is about 25x less conductive than copper.  Too much acts as an insulation.

coretech is spot on about the thermal pads.  If you replaced them with pads that are too thick you can literrally keep the copper base from contacting the GPU.

As far as cleaning alcohol works fine.  Don't use too much make a lot of "passes". If you did "glop" some silver on the traces be sure to clean it all up w/ qtips and alcohol.  It will take forever but eventually you will get it all off.  You know the die is clean when you can drop alchohol on it, use a qtip and the qtip doesn't pick anything up.

[The-Real-Link]

yeah the 2nd GPU tends to run a "little bit" hotter (5C to maybe 10C) but 25C+ and 100C at a decent overclock is indicating poor contact with heatsink.

Don't use silver stuff.  You got exposed traces.  If you put too much on and short a trace well the card does make a good $400 doorstop.

If all you have is white paste that is better than the risk of conductive silver.  In future buy a non-conductive TIM for GPU like MX2 or IC Diamond.

You need a lot less than you think.  It should be a tiny layer.  TIM is more conductive than air but it is about 25x less conductive than copper.  Too much acts as an insulation.

coretech is spot on about the thermal pads.  If you replaced them with pads that are too thick you can literrally keep the copper base from contacting the GPU.

As far as cleaning alcohol works fine.  Don't use too much make a lot of "passes". If you did "glop" some silver on the traces be sure to clean it all up w/ qtips and alcohol.  It will take forever but eventually you will get it all off.  You know the die is clean when you can drop alchohol on it, use a qtip and the qtip doesn't pick anything up.

That's good advice D&T .  I've been used to swabbing my fans / GPU dies with alcohol lately as well and it works very well (just need to be patient for the tiny areas). 

@OP - that seems too hot to be just a fluke.  The first GPUs heat will be blown over the second one so it's normal for it to be hotter by a bit.  It also depends on how your cooling is as you need to be moving a good amount of air for those cards if in a case.  If it's just the card in a case with average cooling, you may eve need to downclock the GPUs by the exhaust if you can figure out which they are.  I'd guess either it's bad or as D&T said, maybe bad contact.

[johnyj]

Time to re-apply thermal compond and thermal pads (1mm thick pads will work, best with those brown colored soft pads), my 2 GPU on 5970 never differ more than 2C and they are all on AS5 (very thin layer)

[ssateneth]

If you want to save about 10 watts worth of heat, instead of running 300mem and 256 worksize, try ~205 (or less, higher may be unstable/slower) and 128 worksize. I was able to save 13 watts at the wall 1 5870 and 2 5830's and lost maybe 1 mhash between the three (used to run 330 mem)

[neo_rage]

for open rig, i think so cooling system of 5970 must be complete modified.

in the standart cooling system you're always bake second gpu.

[xanadu]

Thanks for all of the advice!

I opened the card up and it was a mess, clearly the previous owner had been inside at some point and didn't really know what they were doing.  There was way too much thermal paste on the GPUs and the pads were far too thick, never mind that some were out of place and that they left old adhesive behind.

I redid the thermal paste but since the pads were wrong it still had the same overheating issue with the second GPU, I returned the card and got my money back, not worth dealing with all that hassle, maybe I can find another 5970 somewhere else.

Thanks,
-X