Dajinn08
For overclocking a 1080 Ti SC2, I'm trying to understand what the behavior is supposed to be.
 
Deferring from technical explanations of turbo boost, I want to discuss observations and how it should perform because from witnessing the core clock during games/benchmarks, I am still confused.
 
Does the core clock correlate with the load on the GPU? e.g. 100% GPU load means the GPU tries everything it can to squeeze out as much performance from the parameters set? (power / temp limit, core clock offset, voltage) Or is it load + temp or strictly temp?
 
The settings for my SC2 are 120% power, 90c temp target, voltage maxed, and core offset of +50. Temps chill at around 70c. Whenever I make changes to these and click apply, I see the GPU go to it's theoretical "max" for a split second which is around 2025 mhz. However in a game like FFXV, it starts off reaching 2000 mhz and the card seems to settle back down to around 1936, 1942, 1960 mhz, never really going above those again usually, maybe for like 1 second out of a string of 20-30 minutes it'll boost up to 1987 mhz but it doesn't stick there. All the while, card is still sitting at 70c. 
 
The core clock of 1936/1942/1960 is basically what I see if I don't even overclock the card and I just set it to 120% power 90c temp targets. I have the priority set to temp however even though the temps are 20 degrees below the target, I cannot seem to get any overclock on the core to actually stick.
 
So that leaves me wondering if current core clock is associated with GPU load and if the clock would be higher if the game/application was more demanding?
 
If say, the temps of my card were actually 60 or 65c, would I see over 2000 mhz more frequently with the +50 core offset? If this is the case then what is the point of the temp target and setting it as the priority? You would think that the card would push itself to its max if it's under this target, otherwise to me it doesn't make much sense.
 
Or is it that GPU boost has reductions in maximum clock speeds based on temperature tiers? 

ty_ger07
When the frequency drops, what is the perfcap reason? Once you know what the perfcap reason is, you know what to adjust.

Dajinn08
I kind of wish the big OEMs would stop doing these weird boost things, at least with Intel if you overclock the whole turbo boost is essentially taken out of the equation. But with this it's just kind of awkward. This sounds weird but I really miss the days of just setting a slider and having a card work at the frequency set.

HeavyHemi
The main difference is being voltage limited to 1.093v. You can't add more voltage over that without mods. I never hit that. I'm at 2037 @ 1.0v using the voltage curve.

Valtrius Malleus

Dajinn08
I kind of wish the big OEMs would stop doing these weird boost things, at least with Intel if you overclock the whole turbo boost is essentially taken out of the equation. But with this it's just kind of awkward. This sounds weird but I really miss the days of just setting a slider and having a card work at the frequency set.

I couldn't agree more. With Intel, you get the clocks you want until it hits max Tj and then it throttles. If throttling doesn't work for some reason, like you forgot to put on a heatsink, then it would shut down the PC. nVidias Turbo boost thing is pathetic. Nothing more than a marketing gimmick. If AMD doesn't have it next year when I upgrade my cards then I'll be jumping on team red.


The voltage slider sort of shifts the temp/voltage curve higher so you will get higher volts at the same temperature by increasing the slider.
 

HeavyHemi
The main difference is being voltage limited to 1.093v. You can't add more voltage over that without mods. I never hit that. I'm at 2037 @ 1.0v using the voltage curve.

That's damn impressive. I can only get 1987MHz at 1.0v but I can get +175/+750 (on a FE) even at 1.093v so yours would be capable of hitting higher, at least in theory.


The best way to overclock pascal is to literally overclock every single bin on the voltage curve. Find out the max frequency for each voltage so it doesn't matter where Boost 3.0 draw the temp-line, the card is always performing at it's maximum. The benfits are neglible ingame though and it takes a while to find a stable overlock for each bin. On top of that, the curve can move on it own... thanks to Boost 3.0! So the offset you had set which would get you a stable overclock might be lowered, or it might go higher and cause a crash which is why I hate it so much.
 
 

Dajinn08
How do you change the voltage?

Dajinn08
So pointless. Boost is garbage. Just let me stay at the frequency I want if the temp is fine, jeez.

Valtrius Malleus

ty_ger07
GPU Boost 3.0 is not garbage. Boost is great. As said, it's self-overclocking. Even better, it's adjustable overclocking which changes as the the card's overclocking limits change.

If all you had was a slider -- like back in the good old days -- you could overclock it to 1950 MHz. But then lets say that you discover that your overclock works nicely until the card reaches 70°c, and then it crashes. So, you lower your overclock to 1925 MHz. Now it works without issue. See the problem? Your total overclock is now actually lower than it could have been with GPU Boost 3.0. With GPU Boost 3.0, it overclocks to 1950 MHz when it can, so you get extra FPS when you can, but lowers to 1925 MHz when it has to in order to keep the card from crashing. Better. Certainly.

The issue is that Boost 3.0 overcompensates so you're not getting a better overlock. It's just something else that's been dumbed down to appeal to the masses.

Dajinn08
So pointless. Boost is garbage. Just let me stay at the frequency I want if the temp is fine, jeez.

HeavyHemi

Valtrius Malleus

ty_ger07
GPU Boost 3.0 is not garbage. Boost is great. As said, it's self-overclocking. Even better, it's adjustable overclocking which changes as the the card's overclocking limits change.

If all you had was a slider -- like back in the good old days -- you could overclock it to 1950 MHz. But then lets say that you discover that your overclock works nicely until the card reaches 70°c, and then it crashes. So, you lower your overclock to 1925 MHz. Now it works without issue. See the problem? Your total overclock is now actually lower than it could have been with GPU Boost 3.0. With GPU Boost 3.0, it overclocks to 1950 MHz when it can, so you get extra FPS when you can, but lowers to 1925 MHz when it has to in order to keep the card from crashing. Better. Certainly.

The issue is that Boost 3.0 overcompensates so you're not getting a better overlock. It's just something else that's been dumbed down to appeal to the masses.

You must also realize that automatic clock control is of course going to have some margin. Otherwise RMA rates go through the roof.  This is exactly the same concept that occurs with CPU. Pretty much any recent CPU will over clock by ~20%. But  you have to do that yourself by trial and error. By MANUALLY adjusting for your individual part you can typically gain some performance. The same for a GPU. Also! These GPU's are designed to operate within a certain power draw. Increasing the clocks increasing power draw quite substantially. That's another purpose of GPU Boost 3.0. Even at my moderate overclock, I'm pulling ~300 watts peak. And lastly, of course, it is 'dumbed down for the masses'. Who do you think buys the vast majority of their products? The masses that just stick one in their machine, and that it. Sometimes, I think *enthusiasts have unrealistic expectations of what should be provided for a mass produced product.
 
Custom costs....did you consider the KingPin model?

ty_ger07

Valtrius Malleus

ty_ger07
GPU Boost 3.0 is not garbage. Boost is great. As said, it's self-overclocking. Even better, it's adjustable overclocking which changes as the the card's overclocking limits change.

If all you had was a slider -- like back in the good old days -- you could overclock it to 1950 MHz. But then lets say that you discover that your overclock works nicely until the card reaches 70°c, and then it crashes. So, you lower your overclock to 1925 MHz. Now it works without issue. See the problem? Your total overclock is now actually lower than it could have been with GPU Boost 3.0. With GPU Boost 3.0, it overclocks to 1950 MHz when it can, so you get extra FPS when you can, but lowers to 1925 MHz when it has to in order to keep the card from crashing. Better. Certainly.

The issue is that Boost 3.0 overcompensates so you're not getting a better overlock. It's just something else that's been dumbed down to appeal to the masses.

Not in my experience. The initial boost may be a little low, sure. The initial power safety margin may have some want tor a little more aggressiveness, sure. But once you increase the power % where you want it and increase the boost % to the max stable, the way the boost drops as temperature increases seems just about right to me.

Try it for yourself. Boost to +100 (or whatever your max stable is) see the core frequency is 2000 at 50°c (or whatever it actually is), see that the core frequency drops to 1970 at 70°c (or whatever it actually is), and then without allowing core temperature to drop, boost to +130 (or whatever) so that core frequency is back to 2000 MHz (or whatever) at 70°c (or whatever). Crash. See? These are just example numbers, but my point should be clear and easily substituted for the max stable values of your video card to prove the point.