Cooling two 1080TIs (FE).

balnazzar

Weaksauce
Joined
Mar 13, 2013
Messages
121
Hi mates. I got two 1080TIs Founder's, one of which hits regularly 90C (sometimes 91) with a room temperature of 20/21C, and its fan at 88-90%. The other runs marginally cooler, but always above 85C.
Needless to say, they are loud as hell, and I get no boost (sometimes they even go below 1400 MHz).

My case has a good overall airflow but it has to be said that I use the cards for training neural networks, so it's not like gaming.. They always run at full load.

Now, I got two possible solutions:

a. Accelero Extreme IV. But this will exhaust a lot of hot air inside the case. Moreover, the cards are very tightly spaced.

b. A couple of H55s with nzxt G12. This should be a good solution, but I'm a bit worried about vram/VRMs. Look at what that guy on reddit did:
This leaves the original fan/heatsink for cooling vram and vrms, and should be finer than just a custom fan that blows over them, as you would have with the G12. It looks a bit too much DIY for my tastes, though.

So, in the end I'm asking:

  • Do you have experience with a solution based upon H55 and g12? Pros and cons to report?
  • Will the stock fan included with the g12 be sufficient for cooling the parts? Is it sufficiently quiet? If not, which 92mm fan would you suggest?
  • The H55 seems to be one of the few compatible 120/140mm AIOs (e.g. the h75 2018 is no longer compatible), but if you have any other suggestion, it would be welcome.

Thanks!
 
Hi mates. I got two 1080TIs Founder's, one of which hits regularly 90C (sometimes 91) with a room temperature of 20/21C, and its fan at 88-90%. The other runs marginally cooler, but always above 85C.
Needless to say, they are loud as hell, and I get no boost (sometimes they even go below 1400 MHz).

My case has a good overall airflow but it has to be said that I use the cards for training neural networks, so it's not like gaming.. They always run at full load.

Now, I got two possible solutions:

a. Accelero Extreme IV. But this will exhaust a lot of hot air inside the case. Moreover, the cards are very tightly spaced.

b. A couple of H55s with nzxt G12. This should be a good solution, but I'm a bit worried about vram/VRMs. Look at what that guy on reddit did:
This leaves the original fan/heatsink for cooling vram and vrms, and should be finer than just a custom fan that blows over them, as you would have with the G12. It looks a bit too much DIY for my tastes, though.

So, in the end I'm asking:

  • Do you have experience with a solution based upon H55 and g12? Pros and cons to report?
  • Will the stock fan included with the g12 be sufficient for cooling the parts? Is it sufficiently quiet? If not, which 92mm fan would you suggest?
  • The H55 seems to be one of the few compatible 120/140mm AIOs (e.g. the h75 2018 is no longer compatible), but if you have any other suggestion, it would be welcome.

Thanks!

I would use an H80.
 
Option 3: full cover water blocks.

Option 4: Sell them and get a 2080ti.


Have you blown out the heatsinks? That cooling performance is no where near normal. You can also try cutting off any extra bracket around the exhaust.
 
  • Like
Reactions: Auer
like this
I would use an H80.

Yes I thought about it. H80i v2 is compatible. But it costs almost twice than the h55, and all I want is to stay below 84C. I'm not in need to pursue some 5, or even 10 Celsius.
 
Option 3: full cover water blocks.

Option 4: Sell them and get a 2080ti.


Full cover WBs are too expensive (cosidering all the other parts too).

Yes I could sell them on ebay and get one 2080ti for the same price, but a single 2080ti won't replace two 1080ti for deep learning. I'll have half the memory and about 20-40% less speed (depending on the network type). Actually, on CNNs, two 1080ti still beat a single tesla V100.

Have you blown out the heatsinks? That cooling performance is no where near normal

Yes I thought it so. But I even took out the heatsink (the main ones, upon the core, little or no dust), and repasted them. Result: nothing. Same temps. Maybe 1C less. Mind that I have a good case, and a lot of quality fans..

Don't know.. Any suggestion that makes me stay aircooled would be precious...
 
Look:

The two front 120mm fans blow less than 10cm from the air intake of the cards. They blow at 100% rpms.

https://www.dropbox.com/s/0kgis3p6zg6vd7t/111.jpg?dl=0

The cards are tightly spaced, but a FE design should not suffer from that:

https://www.dropbox.com/s/txezpo540nfe25t/112.jpg?dl=0

The upper card runs hotter. I think it's because it suck in the hot air from the lower card's backplate (you cannot touch it under load). Anyway, the upper one is EVGA, the lower one I bought directly from Nvidia.
Finally, this is the heatsink after 5 months: little or no dust:

https://www.dropbox.com/s/yfd9qx4shdiulyh/113.jpg?dl=0
 
I use an Accelero Xtreme III with my 1080ti FE.
Its very good, temps are around 45 to 55C depending on the game and is so quiet.
My card isnt a good clocker, it gets around 2000MHz sustained, my main reason for buying it was silence as well as better cooling.
Its been running 24/7 for almost 2 years, the fans run like new still.
I also used one on a 290x which was a supreme triumph given how loud it was before that!

I used the Xtreme IV on a 980ti and it cooled ram + VRMs through the PCB, not such good cooling for them.
The backplate clamps bend the PCB to hold it on.
I decided not get the IV again despite the III not being removable.

To use 2 in one machine (and get the best from them) it will need very good case air flow and wide enough apart PCI-e slots to not restrict air flow.
These heatsinks are wide.
I would blow a large fan over them to reduce the chance of hot air immediately recirculating unless case air flow is supremo.
Perhaps keep the case side open and use a very quiet 200mm fan, or fit the fan to the case side.
Remove all unused blanking plates etc to maximise ways for air to exit the case.

I built my PC into an open back coffee table (made from a dining table chopped short), it has an air gap under the surface all the way round except where it mounts.
I route cold air to the gfx card from under the table with 2x150mm air hose and dont worry about how air exits, it makes its own way out.
 
Last edited:
Try changing the thermal paste. Bet the top one is pretty dry from heavy temp swings. Get a fan to throw air directly in between the cards. If you are sandwiching them, you will never escape the heat from the top card, it will choke unless you put a block on it.


I would try changing paste out and putting a fan on the side.


What case do you have, plus give me your fan confgurations.. .
 
I used the Xtreme IV on a 980ti and it cooled ram + VRMs through the PCB, not such good cooling for them.
The backplate clamps bend the PCB to hold it on.

This is truly valuable info. Plus, like you said, using two of them upon tightly spaced cards could even be unfeasible. Thanks!

Try changing the thermal paste.

First thing I did. No result. Used arctic MX-4. I even polished the thermal surfaces with that arctic polisher.

One thing I forgot to mention: none of the cards is connected to a monitor. Nonetheless, in idle the EVGA one consumes 13-14W (and it is the one running hotter under load), while the Nvidia one consumes just 8W in idle. I think the hotter one is just an unlucky chip.

and putting a fan on the side

Yes this would help a lot. But.. How? I cannot conceive a reliable contraption to hold a fan sidewise.. Any suggestion/link?

What case do you have, plus give me your fan confgurations.. .

I got a Sharkoon AI7000. I have two 120mm fans in the front (they blow directly upon card's asses), plus a 140mm fan and two other 120mm (rear and top, respectively).

Note that I installed windows for doing some tests. While palying BF V, the hotter card scarcely hits the 84C limit. Playing less demanding games makes it stay under 80. Conversely, if i run Furmark (extra burn-in), the card behaves like I'm training a neural network on it (hitting ~90C within a few minutes).
 
Last edited:
Either water cool them or find the optimal voltage curve. You could probably run 1080Ti's at like 1800mhz-1900mhz with 850mV-900mV or something which would be about 20% less power draw which would make less heat.
 
Either water cool them or find the optimal voltage curve. You could probably run 1080Ti's at like 1800mhz-1900mhz with 850mV-900mV or something which would be about 20% less power draw which would make less heat.

Nope! Under load, they run at less than 1750 MHz with the standard power limit (250W). Only if I raise the power limit to 300W each, they run at >1850 MHz. This till they hit the 84C limit. Then they run @14xx Mhz. Note that even so, one card goes straight to 90-91C. Then the fan hits some 88% max-rpm, and the card stays there. At this point, if I manually throw the fan at 100% rpms, the card slowly descends to 84C, where it does boost/no-boost, and stays there.

Don't know about voltages, but the freq is not so high.


Man this is quite a valuable advice. Found it on amazon, too.

If I'm not misunderstanding something, from the illustrations I see that you need two of these brackets to hold a fan in place, am I right?

Thanks!
 
How many intakes/exhausts do you have? You should have positive pressure in your case.
 
Nope! Under load, they run at less than 1750 MHz with the standard power limit (250W). Only if I raise the power limit to 300W each, they run at >1850 MHz. This till they hit the 84C limit. Then they run @14xx Mhz. Note that even so, one card goes straight to 90-91C. Then the fan hits some 88% max-rpm, and the card stays there. At this point, if I manually throw the fan at 100% rpms, the card slowly descends to 84C, where it does boost/no-boost, and stays there.

Don't know about voltages, but the freq is not so high.

Is this with a flat curve or with GPU Boost? I'm pretty sure you can get better power efficiency with a flat curve if you're going for 100% load.
 
Nope! Under load, they run at less than 1750 MHz with the standard power limit (250W). Only if I raise the power limit to 300W each, they run at >1850 MHz. This till they hit the 84C limit. Then they run @14xx Mhz. Note that even so, one card goes straight to 90-91C. Then the fan hits some 88% max-rpm, and the card stays there. At this point, if I manually throw the fan at 100% rpms, the card slowly descends to 84C, where it does boost/no-boost, and stays there.

Don't know about voltages, but the freq is not so high.



Man this is quite a valuable advice. Found it on amazon, too.

If I'm not misunderstanding something, from the illustrations I see that you need two of these brackets to hold a fan in place, am I right?

Thanks!


One would work just fine - Just put some rubber grommets on the top or bottom so it doesn't rattle off the GPU.
 
How many intakes/exhausts do you have? You should have positive pressure in your case.

Good point. I have 2 intakes and 3 exhaust. But the intakes run at 12V whereas the exhaust run at 5V. This should ensure positive pressure, but as you know we got the two 1080ti, whose FE design do exhaust the air outside the case. And when they are close to 100% rpms, that's a lot of air.
I honestly don't know if I got true positive pressure inside the case..

Is this with a flat curve or with GPU Boost? I'm pretty sure you can get better power efficiency with a flat curve if you're going for 100% load.

With default profile, which I think is a boost profile.
Consider that under linux you don't have these fancy tools like msi afterburner. The only thing you got is control over the power limit, from 150 to 300W. Theoretically, you could overclock/underclock the cards, but that would require monitors to be attached to them, a thing I'd rather avoid.
Any suggestion aout how to switch to a flat profile under linux? Thanks!
 
With default profile, which I think is a boost profile.
Consider that under linux you don't have these fancy tools like msi afterburner. The only thing you got is control over the power limit, from 150 to 300W. Theoretically, you could overclock/underclock the cards, but that would require monitors to be attached to them, a thing I'd rather avoid.
Any suggestion aout how to switch to a flat profile under linux? Thanks!

Unfortunately I don't think there is a way to do it in linux unless Nvidia X Server has been updated with more features.
 
I have had great results strapping AIO's to GPU's, never did it with a 1080ti but did with a 670 and 780.

I would go with an H80 (thicker rad) as mentioned above, but if you look at the EVGA hybrid kits they use thin rads and cool well. Just make sure the vrms are also getting air. I think the Krakens have a fan for that though.
 
Ok, thanks guys!

I'll go with the fan bracket first. Should I still go over 84C, I'll switch to liquid..

I'll let you know!
 
Ok, it was worth a try, but it didn't work. I gained some 4-5C, but they still go over the throttling threshold.

I will go liquid. h55+g12.

I just want to ask you a couple of questions:
1. Do I have to unglue some thermal pad from the VRM/vram chips prior to mounting the g12? Do I have to remove the backplate too?
2. Can you provide a rough estimate for the failure rate of a 50$ AIO? I mean both liquid spill and pump failure. The machine will work for hours/days with both GPUs in full load (often with me away from home), and I really fear the possibility of starting a fire.
3. I understand that a pump should draw some 20-30W, so the overrall power draw for the whole system will increase of some 50W. I got a seasonic focus platinum 850W, and a Xeon e5-2680v2 with six 16gb ram sticks. Do I risk to kill the psu?

Thanks mates..!
 
Last edited:
Back
Top