Hey guys,
 
I've had my ftw3 for about 2 months now. I've noticed something strange that no one else seems to have brought up in the forums. Basically, when I'm benchmarking and, say, only overclock the memory, I'll get a score, call it 4000. Then I go back in XOC and add some offset to the core clock, bump it by +13mhz. Then when I benchmark again, I actually get a lower score, less than 4000! And it's a difference that one wouldn't feel is within the margin of error. So I try adding more core clock, now I bump it to +26mhz. I get an even lower score than +13mhz gave me. Now I take away the offset on the memory, so that I only have an offset on the core clock. Boom, now my score goes back up to 4000 again!
 
So I look at my hardware monitoring on XOC and I observe that when I have both the offset on the memory and core clock, the power usage of the card goes way down, like it will sag into the 80% range, and that's why I get the lower score. Now, the funny thing is, the monitoring still says the gpu usage is 99%. If I push the card even further it's like it totally gives up, eventually settling around 50% power usage and just totally tanking. By this point, a restart of my PC is needed to get it to behave normally again. 
 
Is this not weird??? It's like the card just gives up and stops using full power once I start asking for too much offset. It gets lazy. 
 
I'm using an EVGA ftwk mobo, 7700k @ 5ghz, EVGA 750 watt gold psu. I thought it might be that my power supply was inadequate at first, since after all it is the power draw of the card that goes down, but 750 watts should be plenty for this card, right? It is just so strange that the card performs worse as you clock its hardware faster. Got all the settings tweaked for max performace in NVidia control panel, I have the latest driver, etc. Using the slave bios. Power target in XOC is 127%, have the voltage slider all the way up as well. Any ideas?

There are two things which could potentially be happening.
 
The first is that the 1080Ti uses ECC memory or self correcting memory. If you overclock the memory too far it causes errors to occur which the card then corrects which costs valuable time.
 
The second is that the card might be hitting the power limit due to the faster clocked memory which then throttles the card down to remain under that limit. I believe the card will still show the utilisation for that clock speed though, otherwise the 100% limit would never be known, or seen, since cards can be overclocked much higher, like on LN2 for example.
 
The 1080Ti FTW3 can use 355W but it can draw more from the wall if your PSU isn't very efficient. That is also a high overclock on the 7700K but it shouldn't be pulling more than 350W anyway.

Valtrius Malleus
There are two things which could potentially be happening.
 
The first is that the 1080Ti uses ECC memory or self correcting memory. If you overclock the memory too far it causes errors to occur which the card then corrects which costs valuable time.

The 1080Ti does NOT feature ECC, nor has any GeForce card from the past 7 years, Titans included.

schulmaster

Valtrius Malleus
There are two things which could potentially be happening.
 
The first is that the 1080Ti uses ECC memory or self correcting memory. If you overclock the memory too far it causes errors to occur which the card then corrects which costs valuable time.

The 1080Ti does NOT feature ECC, nor has any GeForce card from the past 7 years, Titans included.

Ah yes that's correct. Thought I read it somewhere. My mistake.

Odd. The power % shouldn't go down when overclocking too high. Clocks & voltage should be reduced when running into/exceeding the max power target value.

forthegains
Should I give EVGA a ring?  

Only if the issue occurs on both vbioses.