Undervolting my 1080 TI Fe

[Boykie]

Hello! I'm new to undervolting, but it's definitely a must for my GPU since I'm going to be putting it in a SFF case, and it runs pretty hot when under stress (even has some thermal throttling)… I've been using the Heaven 4.0 benchmark and MSI Afterburner when trying to configure the best undervolt… normally the stock settings seem to go up to 1900mhz, but run over 80 degrees (+thermal throttling). I'm still experimenting what would be the best undervolt for it and how far I can get the temps down (I'll gladly take any recommendations!)

 

What I've been doing is setting the curve down by -250 and then bringing up one of the voltage points up to around 1850mhz and prettying apply to make the new curve.

One weird thing that happens when I try this, is that a faint curve line shows up above the new curve I just applied... and even though it shows as 1850mhz on MSI Afterburner, on the benchmark it might display 2000+ mhz (similar to what the new faint curve line displays). Any idea why this might be?

[Hairless Monkey Boy]

The method you are using is okay for finding voltage/freq points for testing, but don't use that method for your final curve.

 

Instead, you should reset the settings in Afterburner, then open up the curve editor. Use the offset slider to shift the entire curve up so that your desired voltage/freq point is at the correct position. Then, hold shift and select all points to the right of your desired point, click on one of them, and drag them all down below your desired freq level. Next hit apply, and Afterburner will flatten out the chart past your desired volt/freq point.

 

The above method is superior because it ensures that volt/freq levels below your desired max are configured properly, and are not reduced compared to stock.

 

As for the faint line, it seems to indicate by how much each point differs from stock, or to indicate the guide that was used to create the final curve. You don't need to pay too much attention to it. I don't know why your benchmarks are displaying incorrect numbers, but that isn't new to me. The clock speed reported by Heaven has never been accurate in my experience.

 

On my wife's GTX 1080, I have a preset that uses 50%-60% of the power limit while delivering 90% of the performance, so undervolting Pascal can yield good results. I have other presets that OC the card to the limit, but the low power one is nice for hot summer days or more basic games.

[Boykie]

7 minutes ago, Hairless Monkey Boy said:

The method you are using is okay for finding voltage/freq points for testing, but don't use that method for your final curve.

 

Instead, you should reset the settings in Afterburner, then open up the curve editor. Use the offset slider to shift the entire curve up so that your desired voltage/freq point is at the correct position. Then, hold shift and select all points to the right of your desired point, click on one of them, and drag them all down below your desired freq level. Next hit apply, and Afterburner will flatten out the chart past your desired volt/freq point.

 

The above method is superior because it ensures that volt/freq levels below your desired max are configured properly, and are not reduced compared to stock.

 

As for the faint line, it seems to indicate by how much each point differs from stock, or to indicate the guide that was used to create the final curve. You don't need to pay too much attention to it. I don't know why your benchmarks are displaying incorrect numbers, but that isn't new to me. The clock speed reported by Heaven has never been accurate in my experience.

 

On my wife's GTX 1080, I have a preset that uses 50%-60% of the power limit while delivering 90% of the performance, so undervolting Pascal can yield good results. I have other presets that OC the card to the limit, but the low power one is nice for hot summer days or more basic games.

Thanks for the heads up! I'll try it when I get home!

[PDifolco]

What I would do practically is to observe max clock and voltage, and set a curve that tops at says 95% of your max clock but with 90% voltage

So say you have 1900MHz and 1V I'd use 1800MHz/ 0.9V

 

Test that, if unstable lower clock, if it works reduce voltage until you find the optimal point

You should get 8C less easily

[Boykie]

2 minutes ago, PDifolco said:

What I would do practically is to observe max clock and voltage, and set a curve that tops at says 95% of your max clock but with 90% voltage

So say you have 1900MHz and 1V I'd use 1800MHz/ 0.9V

 

Test that, if unstable lower clock, if it works reduce voltage until you find the optimal point

You should get 8C less easily

Sounds good! I'll try that later for a config!