MSI Lightning 290X w/ Corsair H75 liquid AIO cooler - high temps

[destrekor]

BLUF/TDLR: 94 degree C !!! HAAALP!

I've had two of these Lightning 290X cards for a few years now, and used to run in Crossfire, which necessitated having the top card modded with a Corsair H75 AIO cooler. Everything ran great for a long time, even overclocked. Lately, I've been having some issues. I removed the bottom card (which has a stock cooler). Now, games like Battlefield 4, Battlefield 1, and now BF V, have just never played nice with Crossfire on my system, and a buddy had similar issues and gave up on multi-GPU and AMD entirely.

But that aside, even running with a single card, what I've noticed now is that the temperature is just not right at all. I had went so long without playing games that I completely forgot just about everything about my original setup! lol I've been trying to get back into it and have just become incredibly frustrated.

What I've come to realize is that these temps were never this insane. They were getting toasty and obnoxiously loud when there were two cards in play and the top card was trying to have a literal meltdown, but the top card never remotely approached the 90s.

What's been happening is essentially the card is throttling, using Afterburner I can watch the frequencies stay at the max setting for awhile and then start randomly fluttering lower while temperature is pegged at 94C.


What I've looked into:
The pump RPM has been maintaining the usual ~1500 RPM and both fans on the radiator are running, controlled by the card/afterburner (pump/block is plugged into a PWR FAN header on mobo.
Wires are all connected as they should be, and have visibly confirmed everything is running, and using touch confirmed the pump seems to be running. And as stated earlier I have confirmed that pump RPMs seem to be reporting correctly -- I at first thought maybe something went wrong with an update or BIOS reset or something and the pump was running at a lower speed than it should be, but I can't find evidence that is the case, as both the BIOS and software readings all suggest the pump is running at max.

If it matters: these cards were utilized fairly heavily in some GPU crypto mining, mainly XMR and ETH. I ran them typically 24/7 for the better part of a year, starting to taper down when the house got way too hot in the summer (ancient central AC needs upgraded, badly).

I would say nothing was improperly installed/mounted at the time, or for years. I maintained temp control while mining, and had plenty of time gaming, pushing everything I could out of the cards, without any issue. But now throttling seems to be the norm and that's killing the gaming experience.

Could it be as simple as needing new thermal paste? Idle temp right now is at ~40C, which based on recollection and some quick search for reviews seems to be normal... I would think idle would be much higher if thermal paste was the issue but I could be wrong.

Could the mining have caused some kind of chip destruction hell that keeps it running rather normally just really hot? That seems the least likely of possible causes.

Does it sound like perhaps the AIO kit has lost water through evaporation? I never saw evidence of prior puddling or other signs of leaks, and nothing else behaves weirdly to suggest anything was shorted. Granted the liquid coolant *should* be non-conductive but I don't know what they use.

 

[tarmc]

Likely need to replace the paste. Has a simular situation with a 970 ramping up to 90c while gaming. Just replaced the tim and it went back to normal

 

[Stuka87]

I agree with the above. Thermal compounds all have a life expectancy, and some are shorter than others. Pull it off, clean it well with alcohol, and reapply the thermal paste. Should fix your issue.

 

[destrekor]

So I replaced the thermal compound (I used the Gelid GC Extreme I had on hand), tested, and initially the idle temps looked lower (I last looked after having the system on for a long time), but I did testing in game, and sure enough, the temp slowly but surely rose right up to the 94C limit!

I tightened down the screws holding the pump on, just a little bit more, until the tension was enough for my perhaps ill-fitted screwdriver to start stripping the screws. It felt tight. It felt tight the first time but I thought maybe it still had some more room to push down... I was quite worried about cracking the exposed die, but alas I didn't.

I went to both dissect the whole kit again, and likely just throw in the other GPU that still has the stock Lightning cooler on (never removed). I wondered if perhaps I had too little or too much paste the first go around, but when I removed the pump I found the coverage looked just about right - not much spread off the sides and what looked to be a thin but complete coverage across the die... it was rather lacy but between what was left on the pump and what was left on the die it looked to be complete. Perhaps it could use just a *touch* more if anything, but I don't think so, it looked good. I already wiped the pump so no I don't have any photos to share. Oops.

With the other card in the temps are ridiculously lower. Now I remember the AIO liquid kit being able to match the other card, so that right away tells me something isn't quite right still.

I'm suspecting there may be a nearly invisible leak or something - nothing that pools on components, but perhaps enough to let the liquid evaporate? I dunno. But looking into the warranty, it looks like these come with 5 year warranty, and I purchased it in the spring of 2015. I had tested and the pump is running at the correct speed so it doesn't appear to be a pump issue. Not sure what else it could be.

 

[MrTeal]

It could still be the cooler. Those kind of high temps are way too hot for water cooling parts 24/7, and at elevated temperatures plasticizer can slough off the tubing and gum up the micro channels on the copper plate. A good cleaning might solve it.

 

[destrekor]

It could still be the cooler. Those kind of high temps are way too hot for water cooling parts 24/7, and at elevated temperatures plasticizer can slough off the tubing and gum up the micro channels on the copper plate. A good cleaning might solve it.

These aren't meant to be serviced by the end user - it's an AIO cooler assembly, pump is essentially integrated into the copper assembly that sits on the processor.

 

[Stuka87]

What does corsairs software say? It should list the pump RPM and such. Does this seem to scale with temps?

 

[railven]

Are you using a NZXT bracket? If yes, make sure the pump/plate are making correct contact with the GPU. I had this happen on one experiment build I was doing for kicks.

Chances are the cooler's pump is failing. Unless he has a iLink version of the cooler, you can't use software to control it, and I don't recall the H75 having PWM on the power connect, pretty sure its just straight 3 pin.

You can also check whatever header is feeding the pump, make sure its set to full power.

 

[MrTeal]

These aren't meant to be serviced by the end user - it's an AIO cooler assembly, pump is essentially integrated into the copper assembly that sits on the processor.

They aren't meant to be, but you usually still can. I've never owned an H75, but it looks like a pretty standard Asetek design. If you toss the head in a vise upside down or something else to keep the pump level and flat (with the res below the head) you can pop off the copper plate without doing disassembly on the unit itself. If you do lose a little, just top up with a bit of DI water before reattaching the block.

Are you using a NZXT bracket? If yes, make sure the pump/plate are making correct contact with the GPU. I had this happen on one experiment build I was doing for kicks.

Chances are the cooler's pump is failing. Unless he has a iLink version of the cooler, you can't use software to control it, and I don't recall the H75 having PWM on the power connect, pretty sure its just straight 3 pin.

You can also check whatever header is feeding the pump, make sure its set to full power.

Yeah, the pumps on these aren't the best. It can be hard to tell as they'll still feel like they're spinning, even if the pump's not moving much water.

 

[destrekor]

What does corsairs software say? It should list the pump RPM and such. Does this seem to scale with temps?

This is just the straight H75, not the H75i with the iLink software tie-in.

Are you using a NZXT bracket? If yes, make sure the pump/plate are making correct contact with the GPU. I had this happen on one experiment build I was doing for kicks.

Chances are the cooler's pump is failing. Unless he has a iLink version of the cooler, you can't use software to control it, and I don't recall the H75 having PWM on the power connect, pretty sure its just straight 3 pin.

You can also check whatever header is feeding the pump, make sure its set to full power.

Yes that's the bracket I'm using. This worked for a couple years with no problems. I thought maybe that was the issue when I put fresh paste on and installed it again, thinking the paste must be good this time but I didn't have it on all the way - I even tightened the screws further (this card has a backplate as well, so it'd get tight before they stopped screwing in by reaching the end of the threads), and I basically got it down to where I had zero movement left using hand-torque only. Frankly I worried I might crack the die. But when I looked the other day after disassembling one more time, the paste had a spread pattern that looked like there was a perfect mating between the two surfaces. I think I checked that concern off the list.

As for the pump speed, yeah the header (a PWR_FAN header) was not power-limiting in any way. And using SpeedFan I found the "fan speed" reporting at a little over 1500RPM, which according to what I've read that's the max pump speed of this model. I also installed the Asus Fan Xpert software and checked it (it's an ASUS motherboard) and the numbers aligned.

They aren't meant to be, but you usually still can. I've never owned an H75, but it looks like a pretty standard Asetek design. If you toss the head in a vise upside down or something else to keep the pump level and flat (with the res below the head) you can pop off the copper plate without doing disassembly on the unit itself. If you do lose a little, just top up with a bit of DI water before reattaching the block.


Yeah, the pumps on these aren't the best. It can be hard to tell as they'll still feel like they're spinning, even if the pump's not moving much water.

While it may be able to be cleaned up, I'd think that'd be a bad idea considering it should still be under warranty (purchased Spring of 2015).


All of this may be moot - as I'd very much like to, sometime soon, pick up a new card. It seems many games these days just don't take advantage of mGPU all that well, or if they do they suffer various problems. And if I only keep one card installed, I don't really have a need for the AIO liquid route -- the MSI Lightning cooler does wonders with only one card -- it was the fact that when I had two, and they take 2.5 slots, there was very little breathing room for the top card with the stock cooler and needed to have fans spooled up to maximum obnoxiousness just to keep it playable. And the games I mostly play as of late, BF1 and now BFV, I get nothing from having a second card. Some games I've been meaning to get back to definitely offer significant improvement but who knows if I'll get back into them by time I finally decide the time is right to buy a new GPU.
If it's in warranty I'll still go down that route, never hurts.