Overclocking a GTX 1070 Ti with Afterburner, pushing it higher... suggestions?

[Sasqui]

So last night, I sat down and watched a few YT videos on overclocking a GTX 1070 using Afterburner. Pretty simple to use once you know the basics and interface

The only controls I messed with so far are:
1. Power and temp limits (MAXED out 120% and 97c)
2. Core clock. The final stable OC = +160 or 2025Mhz +/-
3. Mem clock. The final stable OC = +850 or 4850Mhz or 2x2425 (wow)

In all the test I ran, the temp never gets over 62c ...and that is with the default fan curve! I was disappointed that I could only get my core up to that amount. The high memory overclock helped make up for that. I've gone over stock Vega 64 scores in all of the 3D mark tests

It appears that the power limit is squashing the core overclock any higher. GPUz shows the max TDP at 135%, and max core voltage at 1.06. This is running Furmark, Time Spy Extreme and Firestike Ultra.

Question - is there any way to push the power limit higher??? I know they limit it for safety reasons, but given the headroom for temps, I'd like to push it higher.

Here are the controls I messed with (this is a generic screen capture):

 

[infrared]

Unfortunately you can't really get around the power limit without either a bios flash or physical modification, neither of which I'd recommend.

You might be able to eek a bit more out of it using the frequency/voltage graph (click the symbol to the left of the Core clock text. or Ctrl + F) If you find it's crashing at a certain frequency you can go to that point on the graph and adjust the curve so that it'll only go to that frequency on the next voltage point. Although this isn't gonna help much if you're already on the power limit :/ Maybe setting a higher frequency at slightly less than stock voltage .1.062v is stock (well, stock on 1080ti, I might be wrongly assuming it's the same on a 1070ti), try setting your max frequency at 1.043v and all points after at the same frequncy so it doesn't jump to 1.062v, this might help it stay under the 120% tdp limit.

I hope that helps, have a play and see how you get on

 

[Sasqui]

Unfortunately you can't really get around the power limit without either a bios flash or physical modification, neither of which I'd recommend.

Yea, I wasn't planning on going there. I was wondering if afterburner itself has some possible mods, or is the power limit burned into the BIOS and that's what Afterburner reports and is limited to?

Maybe setting a higher frequency at slightly less than stock voltage .1.062v is stock (well, stock on 1080ti, I might be wrongly assuming it's the same on a 1070ti), try setting your max frequency at 1.043v and all points after at the same frequncy so it doesn't jump to 1.062v, this might help it stay under the 120% tdp limit.

When I first played with MSI AB overclocking , I tried messing with the curves and still couldn't get the core to behave over 2050. I didn't really know what I was doing, since I was also playing with the voltage slider ...and have no clue how the voltage slider and the speed/voltage curves interact, do they? I think the 1.062v is the same for the 1070 Ti and the 1080 Ti... at least is seems so and makes sense.

 

[EarthDog]

MSI AB will not do it as was said earlier.

You need to crank the power limit and use the voltage to find the best clocks without throttling due to power limit. Welcome to locked down NVIDIA, for the 2nd........or is it 3rd...... generation.

Ditch furmark, would ya? If you didn't notice, Furmark throttles the cards clocks way back. Not a good barometer at all as you are not even remotely testing the clocks you intended to test.

 

[Sasqui]

You need to crank the power limit and use the voltage to find the best clocks without throttling due to power limit.

That's basically my method. If I goto +165 on the core, I start noticing the scores going down slightly.

Ditch furmark, would ya? If you didn't notice, Furmark throttles the cards clocks way back.

It worked quite well to find the top Core / Mem clocks. Saw artifacts in Kombustor at +950 mem, so that's when I dialed it back. Also, testing with 3DMark can be a complete PITA when the system crashes and you have to reboot. With Kombustor, it crashes "gracefully" ... so it's a great tool for getting close to the "zone"

 

[EarthDog]

It throttles the CORE back, not the memory. Run 3D FS and watch the GPU clocks, then run furmark and see if it comes close to the same clocks.

Ditch furmark.

 

[Sasqui]

It throttles the CORE back, not the memory.

What throttles the core back is any load that exceeds the power limit. Kombustor worked great for finding the sweet spot, so no, I'm not going to "ditch" it.

 

[EarthDog]

...and Furmark. Test it since you don't seem to believe me.

Look where your core clock settles in Heaven, 3DMark, etc, then run Furmark and see where those clocks settle. You should notice the core clocks are fairly different, thus invalidating what Furmark can really do as you are not really testing against the intended clocks.

 

[mobiuus]

is that ti or the regular 1070? the afterburner pic u posted shows 1070...

 

[Sasqui]

...and Furmark. Test it since you don't seem to believe me.

Look where your core clock settles in Heaven, 3DMark, etc, then run Furmark and see where those clocks settle. You should notice the core clocks are fairly different, thus invalidating what Furmark can really do as you are not really testing against the intended clocks.

It worked damn well for me to find the clocks that proved stable in all of the other tests. For the core, it finds the power limit and stability at +165 (due likely to power limit) For the memory, it found stability at around +950

is that ti or the regular 1070? the afterburner pic u posted shows 1070...

Yea, that's just a stock MSI afterburner image I grabbed from Google. ...Mine is a Gigabyte Aorus GTX 1070 Ti.

 

[John Naylor]

As a means of comparison....also 62C.

http://www.guru3d.com/articles_pages/msi_geforce_gtx_1070_ti_titanium_8g_review,43.html

Core Clock: 1,757 MHz
Boost Clock: ~2,038
Memory Clock: 8,996 MHz

 

[mobiuus]

how much if u dont mind me askin,, did u pay and where??

p.s.u can safely push voltage slider to max (nothing dangerous will happen due to nvidia boost 3.0 safe restrictions) try that

i put power and voltage slider to max and i get 2114mhz gpu-ingame and mem 9500mhz

 

[Sasqui]

As a means of comparison....also 62C.

http://www.guru3d.com/articles_pages/msi_geforce_gtx_1070_ti_titanium_8g_review,43.html

Core Clock: 1,757 MHz
Boost Clock: ~2,038
Memory Clock: 8,996 MHz

I guess I can't complain too much That's almost EXACTLY what I'm getting.

Edit: no, my memory clocks are +850, theirs are +500 (I win)

Why do they see the power limit at 133%... I can only get up to 120%. The only difference I see is they have turned on voltage slider and maxed it out.

how much if u dont mind me askin,, did u pay and where??
p.s.u can safely push voltage slider to max (nothing dangerous will happen due to nvidia boost 3.0 safe restrictions) try that
i put power and voltage slider to max and i get 2114mhz gpu-ingame and mem 9500mhz

Newegg for $590 ouch. ...those clocks are on your MSI 1070?

 

[infrared]

You'd be better off testing with firestrike, or just games that you intend to play tbh. Furmark will always slam a card against the power limit and as a result the core clock will drop more than it normally would in real world situations.

And as dark stalker said, you can put the voltage slider to +100%, it's a very small increase tho, it'll take your card from 1.062 to 1.093. This would increase power draw so you might end up worse off, which is why I suggested playing with the slightly lower voltages. Try a real game and see what your tdp does, you may not be as power limited as you think

 

[Sasqui]

You'd be better off testing with firestrike, or just games that you intend to play tbh. Furmark will always slam a card against the power limit and as a result the core clock will drop more than it normally would in real world situations.

Between you and Earthdog, I'm going to put a bullet in my head.... Do I need to scream this out? I'm using it for short-term stability testing!!! Ultimately, I'm doing all significant stability testing with 3DMark Firestrike (Ultra, Extreme & Standard) + Time Spy (Standard + Extreme). Kombustor saved me HOURS!

And as dark stalker said, you can put the voltage slider to +100%, it's a very small increase tho, it'll take your card from 1.062 to 1.093. This would increase power draw so you might end up worse off, which is why I suggested playing with the slightly lower voltages. Try a real game and see what your tdp does, you may not be as power limited as you think

Yes, that's my next try.

 

[EarthDog]

Nobody is saying it doesn't work. What I am saying is even though you are on +265, when running Furmark it will test at a lower clock due to throttling then when testing with a different application. You are robbing yourself of higher clocks.


I took the liberty of running 3DM Fire Strike Ultra and Furmark with QHD defaul settings. Notice the difference in clocks? You will also notice I hit the power limit withg Furmark and wasn;t close with 3D FS Ultra. Perhaps now you see what I mean and what infra supported.

You'd be better off testing with firestrike, or just games that you intend to play tbh. Furmark will always slam a card against the power limit and as a result the core clock will drop more than it normally would in real world situations.

And as dark stalker said, you can put the voltage slider to +100%, it's a very small increase tho, it'll take your card from 1.062 to 1.093. This would increase power draw so you might end up worse off, which is why I suggested playing with the slightly lower voltages. Try a real game and see what your tdp does, you may not be as power limited as you think

Exactly what I am saying.


EDIT:

Between you and Earthdog, I'm going to put a bullet in my head.... Do I need to scream this out? I'm using it for short-term stability testing!!! Ultimately, I'm doing all significant stability testing with 3DMark Firestrike (Ultra, Extreme & Standard) + Time Spy (Standard + Extreme). Kombustor saved me HOURS!

I feel like it needs to be a large caliber bullet...

... all we are saying is that they are testing differently. You are testing the lower clocks with furmark and the higher clocks with whatever other app you are using. That doesn't make much sense.

 

[Sasqui]

Nobody is saying it doesn't work. What I am saying is even though you are on +265, when running Furmark it will test at a lower clock due to throttling then when testing with a different application. You are robbing yourself of higher clocks.


I took the liberty of running 3DM Fire Strike Ultra and Furmark with QHD defaul settings. Notice the difference in clocks? You will also notice I hit the power limit withg Furmark and wasn;t close with 3D FS Ultra. Perhaps now you see what I mean and what infra supported.
View attachment 96280View attachment 96281





Exactly what I am saying.

Yes, Kombustor pushes the power envelope more than 3DMark, you are overstating the obvious. What you don't seem to get... It's a great (and quick) tool to get near the stable core and men overclocks to then try in more realistic tests. Maybe it doesn't work for you, can't help you there

 

[EarthDog]

And what you don't seem to get is that its really not 'great and quick'. You are testing nearly 75 MHz lower clocks when running Furmark than you are when running other, sane, applications. Its like for a CPU you are testing at 4 GHz when you want to test 4.3 GHz.

You do you...not trying to force your hand, but to show what is what and perhaps save some time in your testing (my testing is fine, mind you). Just trying to help.

GL!

EDIT: In other words, I used to do it this way, however, due to the way NVIDIA GPUs boost, I have been stable with Furmark 100 Mhz lower, and not stable in games where is boosts higher.

 

[Sasqui]

And what you don't seem to get is that its really not 'great and quick'. You are testing nearly 75 MHz lower clocks when running Furmark than you are when running other, sane, applications. Its like for a CPU you are testing at 4 GHz when you want to test 4.3 GHz.

After establishing "max" clocks in Kombustor (aka Furmark), the next step is to test and max out clocks in "sane" applications.

To your point, what I found was +165 was max core in Kumbustor, but it crashed Time Spy. So I dialed it back to +160 and it's all good.

...every crash in 3D Mark meant about 5-10 minutes of rebooting and starting up applications all over again. And each run of 3D Mark is 10-15 minutes itself. So Kombustor saved a TON of time. And by running Kombustor, I'm talking about letting it runw 2-3 min each clock change.

 

[EarthDog]

I have no idea why it takes 5-10mins to reboot and start up applications... are you on a HDD from 2004? I literally reboot in less than 30s. I give it a minute to settle to idle, open up 3DM and go.

I also don't know why 3DMark is 10-15 mins.. TS is like a 3 minute benchmark. Maybe 4 with the demo....5 mins at most to reboot and rerun. If it takes 15-25 minutes to reboot and run as you say, something is wrong with your rig.


Anyway, you do you... just trying to help by giving it a bit of perspective on what you are actually testing versus what is believed to be going on.

 

[Sasqui]

I have no idea why it takes 5-10mins to reboot and start up applications... are you on a HDD from 2004? I literally reboot in less than 30s. I give it a minute to settle to idle, open up 3DM and go.

I also don't know why 3DMark is 10-15 mins.. TS is like a 3 minute benchmark. Maybe 4 with the demo....5 mins at most to reboot and rerun. If it takes 15-25 minutes to reboot and run as you say, something is wrong with your rig.


Anyway, you do you... just trying to help by giving it a bit of perspective on what you are actually testing versus what is believed to be going on.

Nah, you just seem to enjoy arguing. Lol

 

[infrared]

Just do whatever works mate, we've all got different methods and favorite test programs etc.. And we all think our own way is the best

Good luck

 

[EarthDog]

Exactly.

I am not forcing you to do anything. Just attempting to share the reasons why we say not to use Furmark/Kombustor as it doesn't remotely test the same clock speeds for stability and will ADD time to stress testing (you gave an example of how it adds time) in many cases.

GL. Unsubscribed.

 

[Sasqui]

Just do whatever works mate, we've all got different methods and favorite test programs etc.. And we all think our own way is the best

Good luck

Yes indeed, as evidenced by the various methods on Youtube as well, some used Heaven, some used Kombustor while doing a short test of given clock speed

In any case... thanks. I'll update if I have any luck breaking the current power limit barrier.

 

[RejZoR]

For testing, several loops of Unigine Heaven or Valley is still the most realistic workload.