Info Geforce 10 and up - 'just power limit the things'


... not quite, as there's one extra step before that.
Forget undervolting, and forget downclocking - especially that latter one, as there is no need to downclock AT ALL.
Msi Afterburner has a very handy Nvidia OC Scanner. USE IT. The downside is the wait time. [I do not know if other software like Gpu Tweak, Aorus Engine, etc have it.]
Once you get your results, click apply on Afterburner's main HUD, then save the profile. Go back to the OC Scanner window and click revert; if you don't, every time you start up the PC, the settings get reapplied automatically.
Now you can start experimenting with lower power limits, but don't forget to unlink the power and temperature limit on the OCed profile, or else you end up dragging the temperature limit down too.
FYI: the voltage slider does nothing on these gpu series; the algorithm ignores it.

Unigine Superposition custom settings, and the curve the scanner has found and set. [Superposition was used due to being faster to run and get results.]
FYI, there's always a dominant limiter, as well as always saying 'results are considered unstable'. The average core overclock will vary a little bit depending on a warmer/cooler day.
[Does GDDR6/R6X memory have that much headroom over GDDR5/R5X with the 1000mhz OCs that I sometimes see online, or are folks being too aggressive? Only 200mhz from GDDR5X here...]

Current gpu is a Gigabyte 1080Ti Gaming OC with a Kraken G12 + Celsius S36 mounted on it. Here are the results as follows...
Default gpu profile - 100% power:

Default, max power(120% for this model):
Not a lot of change between these two...

OC Scanner profile, max power:
Great, but the goal is not moar powah. How far down can I go 'till it matches the gpu's defaults?

OC Scanner, 100% power:

OC Scanner, 90% power:

OC Scanner, 80% power:
More or less have a match here.

70% power, for the heck of it:

Reasons for this: Seeing some folks downclocking(please don't), others are undervolting too much via the Curve Editor, then there's both undervolting and downclocking.
The algorithm can LIE. When you go too high on gpu clock or memory clock, those symptoms are pretty obvious - driver crashes, black screens, artifacts...
But when attempting to undervolt? There's no obvious visual symptoms like that, meaning it can be overdone and you won't know it.
Not even Gpu-Z's PerfCap Reason and hwinfo's Limit Reasons flag anything. The core clock and memory clock don't go down, even if it seems like it's too little voltage;
once before, I worked my way down to a -0.110v offset from my card's typical 1.043v @ 1949mhz, and it behaved like there was nothing wrong.
The only oddity was Superposition's benchmark scores, but I wasn't able to use them to tell that I was using too little voltage, WHEN THEY BARELY CHANGED.

1)Afterburner has an OC Scanner option. Open AB, click OC Scanner, then click Scan and wait for it to finish. [Took around 35mins for me the last few times I ran it.]
2)When the Scanner is finished, click Apply on the main HUD, then save the profile to one of the numbered buttons(for quick access). After, you click the reset button, then click Revert in OC Scanner.
3)Run benchmark/game on gpu's default. Record, or write down what it gets.
4)Apply the saved profile from the Scanner, click the unlink button(chain link icon), enter a lower value in the Power Limit box, and click Apply.
[This step will need to be repeated until you match your card's performance at default. Use intervals of 5 and 10%, to save some time. For my 1080Ti, I found it at 80% - this doesn't mean it's the same for your card, but it might be a start.]
5)Once you've found a match, save it to a numbered profile. Don't forget to click the Reset button when finished. Done.

I hope some find this helpful.
Last edited: