SLI EVGA GTX 1080 FTW Cooling - ~20c difference between cards

What do you think the problem is?

  • The bottom card is radiating heat towards the top card, causing it to run 20c hotter

    Votes: 7 77.8%
  • Manufacturing variation in the GPU means it does not run as cool

    Votes: 1 11.1%
  • The thermal interface material between the GPU and HSF on GPU 2 is flawed/not there at all.

    Votes: 0 0.0%
  • Airflow inside the case is not good enough

    Votes: 1 11.1%
  • Other (leave in comments)

    Votes: 4 44.4%

  • Total voters
    9

|oR|Spetsnaz

2[H]4U
Joined
Jul 20, 2004
Messages
2,797
I am running SLI EVGA GTX 1080 FTW's and I have noticed that the temperature of the cards can sometimes be 20c different. I have a Corsiar 750D Airflow case (with the front mesh high airflow grille) and I have swapped the stock fans for 3 high airflow Fractal Design HF140 non-PWM models, which are 118 CFM compared to the stock fans which are 67CFM. My CPU cooler is a Corsair H110i GTX with stock fans and it is exhausting out the top of the case. The 1080's are in slots 1 and 4 of my ASUS X99A, so there is 1 slot of spacing between them. So basically, airflow inside the case should be pretty damn good.

Just so it's easier to visualize:

Intake:
2 x 140mm 118 CFM Front Fans behind mesh grille

Exhaust:
1 x 140mm 118 CFM Above cards
2 x 140mm 47 CFM on H110i GTX

When running Unigine Valley, you can see how GPU 1 (bottom) is running about 20c cooler than GPU 2 (top).

I can think of the possible scenarios being:
  • The bottom card is radiating heat towards the top card, athough 20c sounds like a pretty insane difference.
  • Manufacturing variation in the GPU means it does not run as cool.
  • The thermal interface material between the GPU and HSF on GPU 2 is flawed/not there at all.
  • Airflow inside the case is not good enough.
What do you think?

gpu 1.PNG gpu 2.PNG px gpu 1.png px gpu 2.png

I feel like I am taking crazy pills here.
 
If they aren't blower style cards the bottom one is going to be exhausting hot air into the intake of the top card.
 
I will give this a try later tonight and report back.

Thanks for the advice. I don't know why I didn't think to try that.

If that doesn't help, a fan in the side of the case blowing air onto the ACX cards would help tremendously. You can verify that by taking the side panel off and placing a house fan to blow on the cards - if temps drop drastically for the top card, the top card is just recirculating hot air over the heatsink which causes the higher temps.
 
The card that the monitors are hooked up to will run hotter than the other. This is always going to be the top card.

It has been this way ever since Crossfire and SLI came about.

Even if you have double spaced PCIe slots AND intake fans giving cool air directly to the cards this will happen unless something has changed in the latest gen that I am unaware about.
 
The card that the monitors are hooked up to will run hotter than the other. This is always going to be the top card.

It has been this way ever since Crossfire and SLI came about.

Even if you have double spaced PCIe slots AND intake fans giving cool air directly to the cards this will happen unless something has changed in the latest gen that I am unaware about.

What happens if you connect the monitors to the bottom card instead?

Is it possible that the bottom card is just not working very hard, and that's why it's running cooler? Maybe try a game that has really good SLI support, and see if the difference is less. My understanding is that a difference in temperature is normal with SLI setups, and that as long as you're getting satisfactory performance, and not overheating, you don't necessarily need to worry.
 
Top card usually runs hotter because there isn't a lot of airflow between cards.
 
Always had a problem like this with sli - blower style or multiple fan style. Top card gets hotter air from the bottom card and also, usually, has a slighlty higher load than the bottom card (could be 2%, could be 30%). That may add up to a large temperature difference.
In your screenshots there is a 13% load difference (88% vs 75%).
 
Back
Top