Are all 1080 Ti's created equal?

[Damascus]

4 minutes ago, Enderman said:

No, like the fan curve and settings in afterburner or gpu tweak or whatever software you prefer.

That won't keep you under 60c on 95% of air cooled gpus.

[Enderman]

4 minutes ago, sazrocks said:

Unless you keep the card under water keeping temps under 60C even at 100% fan speed while overclocking is very difficult with pascal cards.

1 minute ago, Damascus said:

That won't keep you under 60c on 95% of air cooled gpus.

Have you never used afterburner or something?

Do you know how temp limit works?

[sazrocks]

2 minutes ago, Enderman said:

Have you never used afterburner or something?

Do you know how temp limit works?

Please enlighten me as to what purpose intentionally throttling my card serves when I am overclocking.

[Enderman]

2 minutes ago, sazrocks said:

Please enlighten me as to what purpose intentionally throttling my card serves when I am overclocking.

The setting allows you to increase the default temperature limit of the card so as to avoid throttling at a lower temperature such as what you were speaking about.

I suggest you actually read what I said earlier about "adjusting fan curve and other settings". This is one of those other settings.

I have a feeling you have not been overclocking for very long?

[Damascus]

18 minutes ago, Enderman said:

Do you know how temp limit works?

Yep, Pascal still throttles like an fx cpu at 60c.  Keeping my 1070 at 82c vs 40c (liquid) running identical oc profiles gave me a significant 5% increase in performance.

 

Pascal isn't like broadwell maxwell where temp doesn't matter.

[Enderman]

1 minute ago, Damascus said:

Yep, Pascal still throttles like an fx cpu at 60c.  Keeping my 1070 at 82c vs 40c (liquid) running identical oc profiles gave me a significant 5% increase in performance.

 

Pascal isn't like broadwell maxwell where temp doesn't matter.

That's why you increase the temp limit and up the fan speed and the crappy founders edition performs like a FTW3 or lightning card.

It's only limited by silicon lottery.

[sazrocks]

11 hours ago, Enderman said:

The setting allows you to increase the default temperature limit of the card so as to avoid throttling at a lower temperature such as what you were speaking about.

I suggest you actually read what I said earlier about "adjusting fan curve and other settings". This is one of those other settings.

I have a feeling you have not been overclocking for very long?

Have you ever actually overclocked a pascal card? If you set the temp limit to 92C the card will still start throttling at 60C.

 

As I said earlier, even with the fan running at 100%, 60C is difficult to maintain, especially when the power limit and voltage have been increased.

 

I’m no master overclocker but I have been overclocking for roughly a year and a half, and currently have the fastest superposition 1080p extreme score for a 1060 in the world (AFAIK).

[Wh0_Am_1]

13 hours ago, Enderman said:

But the performance doesn't change.

Whether you buy a founders edition with a crappy stock cooler or a ftw3 your max OC will be the same on average.

The only difference is that the founders edition will need to run the fan near max to keep from throttling.

But that's what I said, noise and aesthetics and warranty, the performance will be the same no matter what brand you buy from.

Sure founders edition could hit those clocks, and crash... The higher end  cards do have some performance gain. Plus they have larger cooler allowing them to turn their fans on less (extending the life of the card and also run a cooler card, which allows for less power loss, since a PCB performs differently at different temps, the higher the temp of the card the more power that is required to get the same work done. which leads to more heat, and then more power draw, and then back to more heat, and etc. Also a cooler card will last longer than a hot card since heat accelerates the degradation of components esp capacitors, and after the task is done the card needs to cool down, and then when you start your next task the card will need to heat up again, as the this occurs the solder will expand and contract, leading to eventual fractures that can kill the card.

[Enderman]

8 minutes ago, Wh0_Am_1 said:

Sure founders edition could hit those clocks, and crash... The higher end  cards do have some performance gain. Plus they have larger cooler allowing them to turn their fans on less (extending the life of the card and also run a cooler card, which allows for less power loss, since a PCB performs differently at different temps, the higher the temp of the card the more power that is required to get the same work done. which leads to more heat, and then more power draw, and then back to more heat, and etc. Also a cooler card will last longer than a hot card since heat accelerates the degradation of components esp capacitors, and after the task is done the card needs to cool down, and then when you start your next task the card will need to heat up again, as the this occurs the solder will expand and contract, leading to eventual fractures that can kill the card.

There are plenty of 1080 FE cards that do over 2100MHz, just as many as any other aftermarket 1080.

I already mentioned that a better cooler allows you to get less noise due to the better temps.

 

What is not true is what you say about "higher end  cards do have some performance gain"

This is not true, just a placebo effect.

Unless you're comparing a stock card to an aftermarket card with factory OC, then obviously a factory OC will perform better.

 

My point is that once you MANUALLY OVERCLOCK IT you are limited by silicon lottery, not temps or power delivery.

A FE cooler with the fan turned up is enough to prevent a 1080 from throttling, but obviously it will be loud.

And as I mentioned earlier, a lightning card with 450W of power delivery or whatever will not get you a higher overclock than a 250W reference design.

[Wh0_Am_1]

1 minute ago, Enderman said:

There are plenty of 1080 FE cards that do over 2100MHz, just as many as any other aftermarket 1080.

I already mentioned that a better cooler allows you to get less noise due to the better temps.

 

What is not true is what you say about "higher end  cards do have some performance gain"

This is not true, just a placebo effect.

Unless you're comparing a stock card to an aftermarket card with factory OC, then obviously a factory OC will perform better.

 

My point is that once you MANUALLY OVERCLOCK IT you are limited by silicon lottery, not temps or power delivery.

A FE cooler with the fan turned up is enough to prevent a 1080 from throttling, but obviously it will be loud.

And as I mentioned earlier, a lightning card with 450W of power delivery or whatever will not get you a higher overclock than a 250W reference design.

Once again some manufacturers do cherrypick their GPUs and heat does play a great role in how stable those overclocks are, on top of that those extra VRMs deliver more power allowing for those speeds to be maintained at a higher temperatures.

[sazrocks]

3 hours ago, Enderman said:

My point is that once you MANUALLY OVERCLOCK IT you are limited by silicon lottery, not temps or power delivery.

A FE cooler with the fan turned up is enough to prevent a 1080 from throttling, but obviously it will be loud.

And as I mentioned earlier, a lightning card with 450W of power delivery or whatever will not get you a higher overclock than a 250W reference design.

This is simply not true. Hitting the power limit is definitely a thing with pascal. 

 

Aside from that, once again, as I said above, GPU boost 3.0 is programmed at the bios level to actually start throttling the card at temps as low as 60C. There have even been reports that the first throttling level is somewhere in the neighborhood of 40C. Therefore if a card has a cooler that can keep the temperature below 60 or at least below the next few throttle levels (something like 65C, 70C, 75C, and 80C) it will be able to out-clock a founders card with the same silicon lottery. And yes, I can confirm these findings by my own many hours of overclocking on my 1060.

[Enderman]

4 hours ago, Wh0_Am_1 said:

Once again some manufacturers do cherrypick their GPUs and heat does play a great role in how stable those overclocks are, on top of that those extra VRMs deliver more power allowing for those speeds to be maintained at a higher temperatures.

Once again you don't understand that temps cause throttling not stability issues.

And no, adding more VRMs will not give you a higher overclock with pascal.

You can take a look at the average OC on any card and there is no pattern that indicates a higher OC on cards that have more VRMs.

And no, stock vs factory overclock does not count, we are talking about manual overclocks.

[Wh0_Am_1]

1 minute ago, Enderman said:

Once again you don't understand that temps cause throttling not stability issues.

And no, adding more VRMs will not give you a higher overclock with pascal.

You can take a look at the average OC on any card and there is no pattern that indicates a higher OC on cards that have more VRMs.

And no, stock vs factory overclock does not count, we are talking about manual overclocks.

Thermal throttling only kicks in when the card is near critical temps otherwise the card balances for power limitations, which are based on temps due to the fact that pcbs perform differently at higher temps, and and the stability of a card is often based on power, sure if you add too much then you will fry it, but if you ask a chip too perform a task and there simply isn't enough power then it can't properly accomplish a task, there is also interference. More VRMs means two things, less heat per VRM, and by extension, less waste, it also means more power with greater stability. And yes, I know that we are talking about manual overclocks, what made you think I wasn't?

[Enderman]

14 minutes ago, Wh0_Am_1 said:

Thermal throttling only kicks in when the card is near critical temps otherwise the card balances for power limitations, which are based on temps due to the fact that pcbs perform differently at higher temps, and and the stability of a card is often based on power, sure if you add too much then you will fry it, but if you ask a chip too perform a task and there simply isn't enough power then it can't properly accomplish a task, there is also interference. More VRMs means two things, less heat per VRM, and by extension, less waste, it also means more power with greater stability. And yes, I know that we are talking about manual overclocks, what made you think I wasn't?

1) which is why you raise the temp limit...

2) as i mentioned earlier, a 250W FE will get the same overclocks as a 400W lightning card, it literally does not matter for pascal.

3) the miniscule difference in "pcb performance at higher temps" is insignificant to a GPU, the VRM and core temp matters.

4) nvidia GPUs have a hard limit on voltage and power, adding more power phases doesn't let you move the slider any more to the right.

5) when the overclock is being limited by silicon lottery it doesn't matter how much more power or cooling you throw at it, it will always crash past the max frequency.

[Wh0_Am_1]

3 hours ago, Enderman said:

1) which is why you raise the temp limit...

2) as i mentioned earlier, a 250W FE will get the same overclocks as a 400W lightning card, it literally does not matter for pascal.

3) the miniscule difference in "pcb performance at higher temps" is insignificant to a GPU, the VRM and core temp matters.

4) nvidia GPUs have a hard limit on voltage and power, adding more power phases doesn't let you move the slider any more to the right.

5) when the overclock is being limited by silicon lottery it doesn't matter how much more power or cooling you throw at it, it will always crash past the max frequency.

And when the temps goes up, the more the card (sorry VRMs, MOSFETs, RAM, and GPUs specifically) becomes significantly less power efficient, which is why more power, and large coolers, offer greater performance. And yes the silicon lottery applies to most cards, a few but select higher end cards have cherry picked GPUs that binned well, and have surpassed the standard, all be it this is less significant than in previous years due to Nvidia getting really good at making GPUs. Also as you mentioned "thermal throttling" or in this case dynamic clock scaling prefers lower temps, which is why a larger cooler makes  such a huge difference. 

Edited March 1, 2018 by Wh0_Am_1
grammatical errors

[sazrocks]

1 hour ago, Wh0_Am_1 said:

dynamic clock scaling prefer lower temps, which is why a cooler makes  such a huge difference. 

+1