**Advice For Overheating GPU**

[mello_newf]

Have a issue here that I haven't encountered before. After hours of fiddling I'm gonna throw up my hand and ask for help.

NOTES
CPU and GPU's are on water
One card is running super hot idle & stress, then it was fine, then it ran even hotter...

GPU = EVGA 680 X2 (1 running at 45-90C when main) (1 running at 25-52C)
CPU = 2500K (OC 4.6)
MOBO = Asrock Z77 Fatality

EK GPU Blocks + Backplates + SLI Bridge (dual parallel)
Rainstorm CPU Block
D5 + Tube Res Combo
1X 120 + 1X 240 rad (60mm thick)
Noctua Industrial fans 3000RPM


FULL STORY
Alright so I upgraded my SLI GTX 480's with SLI GTX 680's. After finishing the loop I did some bleeding and GPU temps looked fine at Idle 25-30C Load 42-47C CPU was low 50's (10min 3dmark loop) (NO O/C on CPU or GPU's)

Few hours later I go to run some 3dmark benchmarks still at stock and I notice things are shuddering, cut it short and look at HWM that was running in the BG. My card hooked up to the monitor was reading 85C the other 35C. I immediately check the card in question with my hand and it's hot as all hell. I check the loop for air pockets by loosening my bleed port, get some bubble action and notice the flow back into the res had increased and water level had dropped quite a bit. Shut down the system and topped up the water supply.

On the restart temps were back to the original Idle 25-30C Load 42-47C ran through the benchmarks without issue at stock, O/C'd my CPU and more benchmarks all good (cpu temps increased a few degrees to 55-60C)

Again a few hours later I decide to play some COD4 and see what the quality difference might be. After a few minutes I get a little shuddering (thought it was the server) 10 min other players give me a heads up that im the only one having issues. close and look at HWM and the main had gotten up to 90C the other 35C

Now I figured I had more air in the loop again so I start working on it..... Nope after 3 hours and fiddling I concluded the following

- No air in the loop
- No kinks
- Pump is running
- SLI Bridge is correctly hooked up
- Switched monitor to other card and idle is 28C for the main 45ish for troublesome card doing nothing. Load was 38C (95%) for the main and 52C for the other with (10-15% usage)
- Enabling disabling SLI made negligible difference in the temp issues.


Thoughts? Suggestions? My next move was to take the loop apart, pull the hot card, remove the block and see if I can see something. Obviously would like to avoid that if im just missing something stupid....

UPDATE
GPUs are used, came mounted to ek blocks. Fedex left them outside and they were ice cold when I got them, condensation was forming on tops of blocks. (I let them sit for over 24hrs and blasted them with a blow-dryer for 20min before putting in) This is something that has gotten me a little worried.

Im looking to see if anyone has had a card run hot without it being a cooling issue. Would condensation damage to the PCB for example ever result in abnormal temps? I have a few days till I will get replacement thermal pads.

 

[natr0n]

reapply thermal paste perhaps ?

 

[mello_newf]

reapply thermal paste perhaps ?

Im pretty sure it will come to that just seeing if I can avoid it haha

One thing I didn't mention

GPUs are used, came mounted to ek blocks. Fedex left them outside and they were ice cold when I got them, condensation was forming on tops of blocks. (I let them sit for over 24hrs and blasted them with a blow-dryer for 20min before putting in) This is something that has gotten me a little worried.

 

[manofthem]

I'd be looking at that first (I think first top card, the hot card). I'd unplug the power from the second card and just run the hot card alone, just so you know the other card isn't influencing anything.

Since the card is that hot, it sounds like it's not making proper contact and thus overheating so I'd be removing the block and refitting it, along with repasting it; thermal pads on vram and vrm might even be salvageable.

Sucks bro, hope you get it figured out

 

[Henkenator68NL]

A hot GPU, well that actually doesn't say anything about the exact cause.

A GPU has several "hot zones", did you check the specific hardware sensor read outs? You should look for sensor info from for example gpu core, gpu memory, gpu shader, gpu vrm's.

I agree with other replies: only way to go is to disassemble the gpu (I would do this always and for all gpu's especially if they have been used).

Gpu's need both thermal paste and thermal pads (my advice buy some replacement pads immediately since your really want to replace those).

It sounds like a bad mounting job... So check that first.

 

[fullinfusion]

I'd say a mounting issue with the problem card. There's a good chance the block still had water inside and froze. If it froze something moved and till you remove the block you may as well do what most ppl reply as...

Check the memory
The PSU is bad
CPU pin bent
Pull the memory and run 1 stick
Use DDU and install the driver again

Am I far off? Lol

Seriously save some time and pull the block. The answer will be seen straight away. Good luck!

 

[micropage7]

first, are you sure that its pretty hot, and not miss temp read?
i agree to check your paste and recheck how good the contact area between your GPU and the water block
and check the block too

 

[Iceni]

How have you got the loop setup.

CPU>GPU>GPU>rads>pump.

Have you got a splitter on the GPU's so water flow gets divided between them or is it directly 1 to the other? If your splitting flow you may need to restrict the cold card's flow so the hot card gets more flow (parallel).

 

[mello_newf]

I'd be looking at that first (I think first top card, the hot card). I'd unplug the power from the second card and just run the hot card alone, just so you know the other card isn't influencing anything.

Since the card is that hot, it sounds like it's not making proper contact and thus overheating so I'd be removing the block and refitting it, along with repasting it; thermal pads on vram and vrm might even be salvageable.

Sucks bro, hope you get it figured out

Didn't think about pulling the power from the other card all together Ill give it a shot. I only set this loop up to test the cards as I have a new case mod in progress but still need to get some stuff including new pads (was going to remount before putting into new mod)

Cheers

A hot GPU, well that actually doesn't say anything about the exact cause.

A GPU has several "hot zones", did you check the specific hardware sensor read outs? You should look for sensor info from for example gpu core, gpu memory, gpu shader, gpu vrm's.

I agree with other replies: only way to go is to disassemble the gpu (I would do this always and for all gpu's especially if they have been used).

Gpu's need both thermal paste and thermal pads (my advice buy some replacement pads immediately since your really want to replace those).

It sounds like a bad mounting job... So check that first.

Yeah I got a few days before ill get new thermal pads to remount the block so im just gonna play with stuff till then and make sure thats the issue before ripping apart the test loop.

What are you using to read individual readouts? I have 5 programs that do temp readouts and none of them show those that I can see.

Cheers

I'd say a mounting issue with the problem card. There's a good chance the block still had water inside and froze. If it froze something moved and till you remove the block you may as well do what most ppl reply as...

Check the memory
The PSU is bad
CPU pin bent
Pull the memory and run 1 stick
Use DDU and install the driver again

Am I far off? Lol

Seriously save some time and pull the block. The answer will be seen straight away. Good luck!

Maybe ill try a new driver install. thanks

first, are you sure that its pretty hot, and not miss temp read?
i agree to check your paste and recheck how good the contact area between your GPU and the water block
and check the block too

I did do a physical check on the cards. The problem card is warm at idle showing 45C and to hot to keep your hand on when at load and showing 85-90C. The other card is cool to the touch when idle and only slightly warmer when at load.

How have you got the loop setup.

CPU>GPU>GPU>rads>pump.

Have you got a splitter on the GPU's so water flow gets divided between them or is it directly 1 to the other? If your splitting flow you may need to restrict the cold card's flow so the hot card gets more flow (parallel).

Loop order is RES-PUMP-RAD-CPU-RAD2-PARALLEL BRIDGE-RES

Have fill/bleed ports at the top and bottom of loop

Using the dual parallel ek bridge.

---------------------------------------------

Im looking to see if anyone has had a card run hot without it being a cooling issue. Would condensation damage to the PCB for example ever result in abnormal temps?

 

[GhostRyder]

From my perspective it sounds like one of the cards is getting bad contact after heating up which is causing the abnormal temps. If the card itself is what is getting hot more than the block what is probably happening is a bit of warping caused by the temps that separates the block enough to lose contact and thus you get great early temps but they get worse over time.

Try taking the block off, cleaning off the paste and make sure the pads are properly aligned on the VRMs and RAM then put some new paste on and reattach the block making sure it goes on properly and (While not to tight) tight enough so it cannot move. After that I would test it and see what happens then post back.

 

[Henkenator68NL]

Best program IMHO, is HWINFO64

it comes in different versions, I always use the latetst beta version = portable version = no need to install, can put it on a usb stick if you want.
can be downloaded here:
http://www.hwinfo.com/beta/hw64_449_2365.zip

run the sensors list and you will see tons of info!

 

[mello_newf]

From my perspective it sounds like one of the cards is getting bad contact after heating up which is causing the abnormal temps. If the card itself is what is getting hot more than the block what is probably happening is a bit of warping caused by the temps that separates the block enough to lose contact and thus you get great early temps but they get worse over time.

Try taking the block off, cleaning off the paste and make sure the pads are properly aligned on the VRMs and RAM then put some new paste on and reattach the block making sure it goes on properly and (While not to tight) tight enough so it cannot move. After that I would test it and see what happens then post back.

Hmm yeah the order of it going back and forth with the temps is whats got me thinking. I wonder if the air in the line was the cause for the first temp increase but getting that hot then being benchmarked also loosened up the contact to the block.

Would also explain why the second time the issue came up it was even worse (hotter faster) because it wasn't just a air bubble it was block separation. More of a relieving thought than anything haha

 

[xvi]

If it's a GPU to heatsink contact issue, your water block should be cold when your GPU is hot. If it's a flow or bubble issue, your water block should be hot. That should help diagnose where your temp difference is coming from.

I'd have to +1 the general consensus of checking to make sure it's getting proper contact with the block though. It's no fun, but it should probably be done.

 

[peche]

UPDATE
GPUs are used, came mounted to ek blocks. Fedex left them outside and they were ice cold when I got them, condensation was forming on tops of blocks. (I let them sit for over 24hrs and blasted them with a blow-dryer for 20min before putting in) This is something that has gotten me a little worried.

well, I would take block off, check PCB and Thermal pads and also re-paste them with a trusted TIM, before putting them on my rig…

it wasn't just a air bubble it was block separation. ...

kinda odd, just one card with this issue?


Regards,

 

[fullinfusion]

Didn't think about pulling the power from the other card all together Ill give it a shot. I only set this loop up to test the cards as I have a new case mod in progress but still need to get some stuff including new pads (was going to remount before putting into new mod)

Cheers



Yeah I got a few days before ill get new thermal pads to remount the block so im just gonna play with stuff till then and make sure thats the issue before ripping apart the test loop.

What are you using to read individual readouts? I have 5 programs that do temp readouts and none of them show those that I can see.

Cheers



Maybe ill try a new driver install. thanks



I did do a physical check on the cards. The problem card is warm at idle showing 45C and to hot to keep your hand on when at load and showing 85-90C. The other card is cool to the touch when idle and only slightly warmer when at load.



Loop order is RES-PUMP-RAD-CPU-RAD2-PARALLEL BRIDGE-RES

Have fill/bleed ports at the top and bottom of loop

Using the dual parallel ek bridge.

---------------------------------------------

Im looking to see if anyone has had a card run hot without it being a cooling issue. Would condensation damage to the PCB for example ever result in abnormal temps?

I was being a smart ass by adding those comments

Really you need to yank the block to see where the problem is.

 

[mello_newf]

Remounted both cards and it resolved the issue. 28C on idle 60C with a hour under full load

I found 2 loose screws under the problem cards backplate holding the block to the card.

Thanks for the reply's everyone