The GTX 1080-Ti Thread

I am on the same boat with tensai28.
No luck with the curve.
I achieved +35 on the core and +350 on the memory with the traditional way.
My card stays at 1949 while gaming though,
cause my temps are in the low 80s(fan 100%).
I noticed that tensai28 uses 135% power target, and as I am still learning the card I wonder if my 150% is the cause of the high temps.
I thought that maximizing power target is a precondition for o/c the card.

 

Not with pascal. You want to keep you power target as low as possible because the more energy you use, the hotter it gets and then it throttles. I've had better luck overclocking the memory. Try lowering to 125-135% and see if you're still stable.

 

You want to keep the power USE as low as possible as once you hit 120% the card will throttle.
These cards will throttle due to temps.
AND
Throttle if you hit 120% power.
So you use the curve to hit a set core speed but at a lower voltage.

So if my card throttles at 2100mhz at X volts (where X is a set voltage), then i use the curve to set X slightly lower. If i can run stable at 2100 with the voltage at the next increment down from X, chances are i wont hit the 120% cut off and the core wont throttle.

The key to OCing these cards is keeping temps low and the stable voltage used as low as possible for the speed you want to run at.

A fictitious example below.
My card hits 120% power limit at 2100mhz stable at 1.3v but throttles.
I set 2100mhz at 1.25v and it hits 115% and doesn't throttle and is still stable.

 

Thanks for the info. I didn't know 120% made it throttle. I thought it was just based on temperature. Unfortunately I had to put it at 135% to get it stable. I guess I'll play around with the curve more. Thanks.

 

Im not familiar with your card but i always thought all 1080Ti's had a 120% power limit. So if you can put 135% into your OCing software then that may well be your issue, the card will throttle when it hits the power draw limit so use the curve to keep it below that point.

Run a windowed stress test with the power curve visible so you can see the power draw bounce around dependant on whats onscreen.

Using the power curve is also a great way to lower temps if you are happy with a set core speed. If you are happy with say 1900mhz but have high temps (that may cause throttling) then just use the curve to drop the voltage at 1900mhz until you lose stability.

Again a fictitious set of setting may be.
1.3v and 1900mhz = 80c (stable, thermal throttling).
1.25v and 1900mhz = 70c (stable, no thermal throttling).
1.2v and 1900mhz =65c (unstable, no thermal throttling)

Ultimately and rather annoyingly, watercooling wont guarantee you any more speed as the card has two factors that limit speed, both thermal AND power draw. So WCing removed the thermal limit but not the power draw limit. all 1080Ti are the same in that respect (unless you use various bios tricks).

 

Anyone notice a 50% reduction in vulkan and dx12 drawcalls? I just noticed a huge drop on windows 10 regardless of what drivers I try. Dx11 remains unaffected.
Nevermind, removing my video card and putting it back in fixed it. Cat must have knocked it loose. I thought I was going insane for a second.

 

Anyone else here under-volting their 1080Ti?? I just gave it a try and I must say I'm impressed!! Card runs soo much cooler now.. this was near the start of the run but only hit 51C by the end of it. Not to mention it didn't change voltage/speeds at all the entire run!

1924Mhz @ 0.950v!

 

Which game?

 

No but on certain (older) games il put VSync on to limit to 60fps which drops core speed and temps by a lot. COH2 and Empire/Napoleon Total War spring to mind

 

Well I'm gonna post this again with more detail of what I have for an overclock right now in hopes that someone can help me out here because I'm really stuck. With this overclock, boost 3.0 has me throttling down to 1974mhz from 2ghz after a few minutes. Open to any suggestions on what to try. I've seen the tips on here for the voltage curve and did my own fiddling around but had no luck at all with the voltage curve. But then again I'm new to this kind of overclocking.
Here's what I got so far 100% stable:

Spoiler

Thanks in advance!

 

Have a Google for the Guru 3D YouTube video on setting up the voltage curve.

 

Yeah I've seen a few videos including the one by Hilbert. No luck though. My card just doesn't seem to like overclocking this way.

 

What speed do you hit if you just raise the temp and power limits to max? Also have you unlocked the voltage control?

 

It's not your card that doesn't like overclocking, that curve is just not going to be stable for majority of cards.

Do this.

Reset to default, bring up the curve

Hold shift and drag the whole curve down so that the points on the right are at max frequency of about 1900mhz.

Next, select a single point and drag it up to perhaps 2000mhz.
I would start off with a voltage of 1.025.

So 1.025v @ 2000mhz~.

All the points past 1.025v will be the same(this ensures it does not go past the first set point).

After that, the 3 or 4 points below 1.025, you'll want to reduce them by 13mhz step each.
so it would be 2000 @ 1.025, 1987 @ 1.012, 1970 @ 1.0v and so on.

You do not want to go beyond 1.06v~ for most cases as the average frequency will go down due to limited power target.

Reducing voltages reduces power usage which will lower temps while keeping average clock speeds up.

If above is not stable, maybe try 2ghz at 1.0375v or 1987@ 1.025v etc.

 

Thanks I'll try it out later.

 

I've tried as you suggested but for some reason my card is running at 1974mhz @1.025v when running the heaven benchmark.


Trying your second suggestion gives me this:


Card seems stuck at 1974 no matter what but at least the traditional method gives me 2ghz for a while before going down to 1974. This just shot to that right away. I did set the power limit to still be at 135% because I know that I need that to be stable at 2ghz otherwise it's an instant crash.
I do really appreciate the help but it seems this card really doesn't respond well to voltage curve overclocking.

 

It's normal behavior.

Pascal has one boost bin drop at 35c, then another bin drop somewhere around 50-60~ degrees, hence 1974.

To counteract that, you would want to set 2025(provided you are stable at 2025) because once temperatures stabilize, it will drop clock by a bin or two.

 

Hmm so you would suggest trying 2025mhz at 1.0375v? By the way, is there a way to create custom voltage points? I actually don't have 1.0375 but 1.031 and 1.043.

 

In a nutshell, Pascal can be very frustrating to OC, certainly for those with great cooling solutions as its power draw that is now the limiting factor rather than temps

 

Yeah I just failed again. Honestly right now even the slightest deviation from my current curve will cause it to crash no matter what. I think if I did overclock via the curve method, it would turn out exactly the same as what I am getting the traditional way. In other words; not worth the trouble. Thanks for the help anyways guys.