[SOLVED] Why does my GTX 2080 Ti run faster when underclocked?

[Andrew_St]

I have a 2080 Ti in which I replaced the stock fans with a Morpheus II VGA cooler. Heatsinks are properly attached and placed where they need to be, and temperatures are much cooler.

However, upon overclocking with MSI Afterburner and every tweak acting as it is supposed to, my "Power Limit (%)" setting is acting backwards. The sweet spot I have found for maximum FPS in benchmarks is the 78% setting; Any lower or higher and performance worsens.

I assumed it might be a problem with power delivery to the card, but every cord is properly attached both to the PSU and GPU. My PSU is a gold certified Corsair RMx 850 W.

Any help would be much appreciated!

 

[Lutfij]

My thoughts are perhaps the lower limit means not enough power to the GPU while a higher limit would mean that the VRM's are heating up affecting the amount of power deliver to the GPU effectively. Just to rule out my theory, you might want to slap the stock cooler on and see if your results are the same or are affected.

Aftermarket coolers tend to either cool everything properly or cool certain areas less effectively. Make and model of the GTX2080Ti in question?

 

[Phaaze88]

Not unusual for 10 series and up, which are temperature sensitive. Higher power limit doesn't equate better performance, unless thermals can be kept way down.

Also, if you benchmark those cards from cold boot, they'll produce a high, but inaccurate result. The next one after will be worse and any attempts to try and match the first usually won't succeed.
It's because the card had finally 'warmed up'.
That's why it's recommended to warm it up first, then bench it.

The cooler they run, the higher clocks they can sustain, and the higher bench scores you attain - bar power limits, anyway.

 

[Andrew_St]

My thoughts are perhaps the lower limit means not enough power to the GPU while a higher limit would mean that the VRM's are heating up affecting the amount of power deliver to the GPU effectively. Just to rule out my theory, you might want to slap the stock cooler on and see if your results are the same or are affected.

Aftermarket coolers tend to either cool everything properly or cool certain areas less effectively. Make and model of the GTX2080Ti in question?

I have one heatsink on the card's PCB I know is not very well attached, but still hasn't fallen off the card so I figured it makes enough contact.

I could attempt getting some sort of glue as opposed to the not-so-sticky strips the heatsinks came with, but there is no way I am putting the stock cooler back on, it was an absolute pain taking it off.

The card is a FE 2080 Ti.

 

[Andrew_St]

Not unusual for 10 series and up, which are temperature sensitive. Higher power limit doesn't equate better performance, unless thermals can be kept way down.

Also, if you benchmark those cards from cold boot, they'll produce a high, but inaccurate result. The next one after will be worse and any attempts to try and match the first usually won't succeed.
It's because the card had finally 'warmed up'.
That's why it's recommended to warm it up first, then bench it.

The cooler they run, the higher clocks they can sustain, and the higher bench scores you attain - bar power limits, anyway.

From MSI Afterburner monitoring, the GPU core temperature maxes out at 71 C.

You are correct about the cold-starting, I am familiar with this and generally disregard the first benchmark.

 

[Phaaze88]

At even lower than 71C, your card can sustain even higher frequencies.

Before I replaced the stock cooler on my 1080Ti Gaming OC:
-It would never see over 65C under load even with the power slider maxed out.
-It would boost to 1949mhz, from 1480mhz, but would not stay there, often dropping 85mhz.
-Now that it's under a Kraken G12 + Fractal Celsius S36, it never sees over 40C and pretty much sits at 1949mhz, save for menus and titles that don't really require the gpu.

Dropping the power slider will do little in my situation, because I'm pretty sure the power limit is what's stopping this card from going any higher. +469mhz from base is already good anyway.

 

[spentshells]

Less power = less heat, less heat higher boost, the math involved with the boost tables may just be working out in your favor.

Is your performance better or worse than the stock cooler with an unaltered power limit?
The stock cooler looks awesome but apparently it wasnt that great.