Changing voltage doesn't stabilize overclock ?

[gamingmonkey]

Hey guys.

Recently, I got to OC my GPU (MSI GTX 980 Ti Gaming), and I went slow and increasing the core speed in +10 increments (haven't touched the memory yet), so far I'm stuck where core +115 (1469 MHz) is my most stable speed (low I know), and when I try to move it by +5 or more it causes a driver crash.

Now all of this I'm doing is done at stock voltage, and I've heard that it's possible to go higher with overclocking by maxing out the core voltage. I do that (maxes at 87 mV) and then try the +120 offset only for it to crash again.

I'm wondering but shouldn't the increase in voltage give a little more stability, as I have noticed the voltage bump when monitoring but the heat too ?

If anyone is curious my ASIC is 77.5% , I'm on stock cooling (Twin Frozr) and stock BIOS.

Thanks.

 

[Mr.Scott]

GPU temp is?
You voltage and power limit settings seem excessive for the clocks you're running.

 

[trog100]

i run mine (sli) at +96 on the core +296 on the memory.. everything else is at stock..

going much higher on the core will bring in the odd signs of instability.. mine have been rock solid this way now for nearly two years..

trog

 

[Vayra86]

Back it down to +100 if you want this stable for longer periods of time and prolonged gaming loads where GPU temp will cross the 80 C barrier.

Silicon lottery.

 

[JalleR]

sounds like you are having temperature issue.

 

[gamingmonkey]

GPU temp is?
You voltage and power limit settings seem excessive for the clocks you're running.

GPU Temps are from 68-74C on +115 core, without voltage increase. With voltage increase the temps jump from 72-78C.

Also what do you mean voltage and power limit seem excessive ? My power limit is at max (107%), and the voltage can go up to 87mV.

Back it down to +100 if you want this stable for longer periods of time and prolonged gaming loads where GPU temp will cross the 80 C barrier.

Silicon lottery.

Why would I want to put it only at +15 core speed ? That seems awfully low.

And no thanks, I don't want to cross 80C temps on my GPU.

sounds like you are having temperature issue.

How ? My temps seem fine where they are 68-74C on load.

 

[trog100]

+115 on the core was where my stable point seemed to be.. but over a couple of weeks or so of real gaming use i dropped it down to +96..

do the same with your card and it will be stable..

trog

 

[vega22]

binning is a thing, the better cores went into review samples and their top version.

 

[Vayra86]

GPU Temps are from 68-74C on +115 core, without voltage increase. With voltage increase the temps jump from 72-78C.

Also what do you mean voltage and power limit seem excessive ? My power limit is at max (107%), and the voltage can go up to 87mV.



Why would I want to put it only at +15 core speed ? That seems awfully low.

And no thanks, I don't want to cross 80C temps on my GPU.




How ? My temps seem fine where they are 68-74C on load.

I'll give you a quick guide for OC on Pascal (air).

- Pascal's GPU Boost 3.0 works with a base clock > boost clock and the boost clock is finely grained into 13 mhz steps. These are called 'boost bins'.
- Pascal's BIOS hard limits GPU Boost 3.0 to 83 C. This means: it will do whatever is needed to the boost clock and voltages, to stay under 83 C.
- Every 5 degrees C past 55 C, Pascal tries to add one or more boost bins at the SAME voltage for that temperature point, if it can apply higher voltage, then it will, and again determine if it can add boost bins, until it reaches the cap for either the voltage it gets, or the temperature limit.
- Once you cross 75 C, in most cases, you can add additional GPU core clock, but in the end you'll see that you lose the same frequency in dropped boost bins: at the current voltage, Pascal cannot sustain the boost bin without exceeding the temperature limit for that boost bin, or exceeding the voltage that is currently set.

So, if you think about this, at some point around 70 - 75 C, once you hit that wall where your temps go towards 80 C, it is more effective to LOWER the voltage and/or the clock OC, so that GPU Boost 3.0 gets the headroom to play with that last boost bin. If you push maximum volts, the temps will go past 75 C, and you'll see the clocks fluctuate constantly: temp goes up; boost bin drop 13-26mhz or more, temp goes down > it tries to reapply your maximum OC > rinse and repeat.

This is why it is better to back off the OC by -15, which is just about one boost bin - if the GPU has the headroom, it will GIVE you that boost bin back by itself. Even more effective, is to balance out the maxed out Core voltage with the 75 C temp target, so even if you 100% stress the card at that voltage, it won't exceed that temp target (this is also why overclocking apps allow you to set temperature instead of power targets!!)

Note also: OC on memory is very effective on Pascal, and if you do not give the memory +10% or so, you can OC the core but you won't see it translated into extra performance. These cards are very very well balanced that way.

About the 80+C - there is no harm done in running the card at or over this temperature at all, so do not worry. But for the best OC efficiency, you'll want to stay under.

And about binning: when all Pascal cards manage to score between 2000-2180mhz, this is a bullshit statement. Binning only exists in the minds of professional overclockers these days, there are no real Pascal 'duds' because even the worst ones get to 1950 given sufficient cooling and since they drop boost bins at temp targets, on air there is no way you'll ever notice binned cards vs not binned ones. If you buy into binned cards at a premium, in my book you've lost the plot

 

[P4-630]

I'll give you a quick guide for OC on Pascal (air).

- Pascal's GPU Boost 3.0 works with a base clock > boost clock and the boost clock is finely grained into 13 mhz steps. These are called 'boost bins'.
- Pascal's BIOS hard limits GPU Boost 3.0 to 83 C. This means: it will do whatever is needed to the boost clock and voltages, to stay under 83 C.
- Every 5 degrees C past 55 C, Pascal tries to add one or more boost bins at the SAME voltage for that temperature point, if it can apply higher voltage, then it will, and again determine if it can add boost bins, until it reaches the cap for either the voltage it gets, or the temperature limit.
- Once you cross 75 C, in most cases, you can add additional GPU core clock, but in the end you'll see that you lose the same frequency in dropped boost bins: at the current voltage, Pascal cannot sustain the boost bin without exceeding the temperature limit for that boost bin, or exceeding the voltage that is currently set.

So, if you think about this, at some point around 70 - 75 C, once you hit that wall where your temps go towards 80 C, it is more effective to LOWER the voltage and/or the clock OC, so that GPU Boost 3.0 gets the headroom to play with that last boost bin. If you push maximum volts, the temps will go past 75 C, and you'll see the clocks fluctuate constantly: temp goes up; boost bin drop 13-26mhz or more, temp goes down > it tries to reapply your maximum OC > rinse and repeat.

This is why it is better to back off the OC by -15, which is just about one boost bin - if the GPU has the headroom, it will GIVE you that boost bin back by itself. Even more effective, is to balance out the maxed out Core voltage with the 75 C temp target, so even if you 100% stress the card at that voltage, it won't exceed that temp target (this is also why overclocking apps allow you to set temperature instead of power targets!!)

Note also: OC on memory is very effective on Pascal, and if you do not give the memory +10% or so, you can OC the core but you won't see it translated into extra performance. These cards are very very well balanced that way.

About the 80+C - there is no harm done in running the card at or over this temperature at all, so do not worry. But for the best OC efficiency, you'll want to stay under.

And about binning: when all Pascal cards manage to score between 2000-2180mhz, this is a bullshit statement. Binning only exists in the minds of professional overclockers these days, there are no real Pascal 'duds' because even the worst ones get to 1950 given sufficient cooling and since they drop boost bins at temp targets, on air there is no way you'll ever notice binned cards vs not binned ones. If you buy into binned cards at a premium, in my book you've lost the plot

Recently, I got to OC my GPU (MSI GTX 980 Ti Gaming)

 

[neatfeatguy]

My 980Ti AMP! Omega, I found the best spots for overclocking (on top of the factory OC) was:

+100 on core clock
+295 on the memory clock

I can push these clocks without adjusting any voltage and with the massive cooler on this thing, temps stay keep under 75C for heavy gaming

Max OC I can push and keep stable (with the minimal testing I did) was:
+135 on core clock
+405 on memory clock
I only tested these clocks on benchmarks such as Heaven 4.0 and Firestrike. I didn't actually try to game at these clocks.

(found the gpuz validation on my max OC for my 980Ti: https://www.techpowerup.com/gpuz/details/4puaf )

Remember that not all cards will overclock well. I had two identical 8800GTS 512MB cards years back; same model, same BIOS. One card I could ramp up the clocks by 100/150 and the other wouldn't run without crashing with the clocks adjusted even just a tiny bit. Sometimes you get a beast of a card you can OC the crap out of. Other times you get a card that utterly fails at OCing.

I'd recommend you trying to keep the core around +80-90 since 115 seems to be the max you hit without issues and then work on the memory. Hopefully you can get it up at least an extra 100-150. You should get a much better result by backing down the core OC a tad and then pushing up the memory.

 

[Vayra86]

HA HA
Did I miss that WOW

Regardless, much of the same principles apply to GPU Boost 2.0 - except for the temperature targets for every 5 C - there is just the 83 C hard limit.

We call it Paxwell for a reason ^^

 

[gamingmonkey]

HA HA
Did I miss that WOW

Regardless, much of the same principles apply to GPU Boost 2.0 - except for the temperature targets for every 5 C - there is just the 83 C hard limit.

We call it Paxwell for a reason ^^

So I assume I can lower the core speed , and it would still hit 1460 MHz (just dependent on the temps) ?

 

[Vayra86]

So I assume I can lower the core speed , and it would still hit 1460 MHz (just dependent on the temps) ?

Likely. Or you will be stuck to 1447 or so but it will be actually stable at that clock, instead of black screening your game at some point causing you to back down anyway

General rule of thumb for any 24/7 OC is you find the max, and back down a couple notches for 24/7 use.

 

[trog100]

when it gets near 80 C it will start to throttle down.. these things are built to be idiot proof.. they have to be..

try running furmark and see what it throttles down to.. he he

trog

 

[eidairaman1]

Abient temperatures have a major play on them too.

Not all gpus oc the same.

 

[EarthDog]

binning is a thing, the better cores went into review samples

Review samples are binned a lot less than you think. Nearly every card ive bought out clocked my review sample. I can think of about two instances of around 10, where this happened.

 

[xorbe]

Eventually you hit a clock wall where extra V does nothing. It doesn't help that boost moves the freq and voltage all over the place.