GTX 1080Ti crashing at 1940 Mhz

[trinxo]

Last month i bought a second hand gtx 1080ti (msi armor oc), i did some test and the card performed well (average with other stock gtx 1080ti by userbenchmark) , i also bought a 4k monitor and i wanted to overclock the card for extra fps (needed in 4k, nowodays games are design to work with dlss).
My surprise comes when i saw that i coud only do a +60 core overclock with the power limit set the highest (117). If i try +70 kombustor stress test crash in seconds. All of this with no memory overclock. This card is well-known to rich high tempertures so when i amb playing the frequency starts at 1920-25 and drops below 1900 as the temperature get high.
More interesting here is that the power limit do almost nothing, with 100% i can apply a +40 core and only +60 with at 117%. If i set the vcore limit to maxium, frequency can get to 1940 in soft gaming, but with this card is not a option since it gets pretty hot.
For what i have seen gtx1080ti in theory can go at least to 2000 mhz. I tryed a few things like change software (afterburner to asus gpu tweak) , reinstalling drivers, change power managment options in nvidia control panel,all makes no difference, I also look in bios for some power saving option, but everything looks fine.
My specs:
i7-6700k
Msi armor oc GTX 1080 Ti
asus z170 pro gaming
corsair 750W with gold standard ( i don't think it can be the power supply)

If someone had or have a problem like this or you think you may know the answer, I will appreciate some help.

thanks,
Marc

 

[P4-630]

2 things, cooling and silicon lottery.

 

[Deleted member 193596]

my 1080 ti Strix OC does not do anything above stock.
+15 = Stuttering and artifacting, +30 = crashes.

 

[trinxo]

That's nice to know, searching on internet it seems that each card was getting a good overclock, but your strix it's probably more overclocked by factory.

Thanks,
Marc

 

[FinneousPJ]

According to MSI that card is specced at 1645 MHz / 1531 MHz (Boost / Base)

MSI Global - The Leading Brand in High-end Gaming & Professional Creation

As a world leading gaming brand, MSI is the most trusted name in gaming and eSports. We stand by our principles of breakthroughs in design, and roll out the amazing gaming gear like motherboards, graphics cards, laptops and desktops.

www.msi.com

 

[Vayra86]

Reduce your power target. Don't increase it.

Heat = current leakage = more voltage needed = more heat = GPU gets unstable. That is how your OC works right now. Or, doesn't work.

Start with +0 Core OC, 95% power target and work your way up slowly - ONLY on the core overclock. When you reach the limit of stability, dial it back 10-20 mhz and put +300-500 mhz on Memory. Save. Done. Happy gaming.

Why does it work like that? Because GPU Boost will already bring your clock to near-maximum, the cards come with a predefined voltage/clock curve and adapt on it based on temperature. Higher temps will lead to lower boost clocks, so if you OC the core too much, boost can't dial 'down' properly within the same voltage bracket and becomes unstable.

Some other points of interest:
- Delete Kombustor and use something that stresses the GPU properly, like 3DMark. All you get with Kombustor is heat. Not stability or a good OC (ex: your experience).
- An OC won't turn 4K into playable - reducing quality settings can.

 

[trinxo]

Reduce your power target. Don't increase it.

Heat = current leakage = more voltage needed = more heat = GPU gets unstable. That is how your OC works right now. Or, doesn't work.

Start with +0 Core OC, 95% power target and work your way up slowly - ONLY on the core overclock. When you reach the limit of stability, dial it back 10-20 mhz and put +300-500 mhz on Memory. Save. Done. Happy gaming.

Why does it work like that? Because GPU Boost will already bring your clock to near-maximum, the cards come with a predefined voltage/clock curve and adapt on it based on temperature. Higher temps will lead to lower boost clocks, so if you OC the core too much, boost can't dial 'down' properly within the same voltage bracket and becomes unstable.

Some other points of interest:
- Delete Kombustor and use something that stresses the GPU properly, like 3DMark. All you get with Kombustor is heat. Not stability or a good OC (ex: your experience).
- An OC won't turn 4K into playable - reducing quality settings can.

Thanks for that, i've done what you said and with 95% of power limit i can do almost the same overclock than before but the temperatures still the same or even more, what makes no sense for me. Despite limiting the power, what i can see in the graph is that the power goes up constanly over 95, maybe afterburner can't control the spikes.

Marc

 

[Vayra86]

Thanks for that, i've done what you said and with 95% of power limit i can do almost the same overclock than before but the temperatures still the same or even more, what makes no sense for me. Despite limiting the power, what i can see in the graph is that the power goes up constanly over 95, maybe afterburner can't control the spikes.

Marc

No, its because the card will stlll boost, that is also why you're not losing much performance. You can lop off another 5% and see what happens, too and find a sweet spot where noise and temps are minimal but performance still great.

What are your core/mem OC settings now with the lower power target? Still trying to go hard on core? That might be the reason it edges over the specified range in AB.

 

[FreedomEclipse]

The thing is. Pascal starts to 'throttle' the moment temps start going over 60'c. Thats how the boost works. To get the best out of it youre going to need a better cooler. The one of the Armor OC isnt a good one

GN's article has a lot to say about the card, also lots of feedback from people who bought the card on newegg and amazon with 99% of them saying the cooler is crap and the card runs hot.

Get yourself an after market cooler. Either something from Raijintek like a MORPHEUS II or an Accelero Xtreme III. I got the same Accelero cooler on my 1080Ti and got +100 on the core and +660 on the memory though I do have heatsinks on the memory as well though I dont know if those make any difference to my OC or not. I cant remove them now as they are glued on with thermal epoxy.

Your card will probably turbo over 1940mhz if you get a new cooler or find a way to cool it better.

 

[trinxo]

No, its because the card will stlll boost, that is also why you're not losing much performance. You can lop off another 5% and see what happens, too and find a sweet spot where noise and temps are minimal but performance still great.

What are your core/mem OC settings now with the lower power target? Still trying to go hard on core? That might be the reason it edges over the specified range in AB.

My actual configuration:

I am testing the stability with heaven benchmark 4.0 now. The result of the benchmark does not change for 90 or 95 power limit, temperature is around 1 degree more hot for 95%, while playing cyberpunk the max temperature is 79 degrees and 78 for heaven benchmark. I tryed to open the pc case and tempertures are better, 72ºC for heaven benchmark, so i will have to figure out how to improve the airflow. What's weird is that yesterday playing cyberpunk with 107% of power limit I was getting 76 ºC (it could be an external issue like room temperature or different zone in the game ).

About the overclock, +60 doesn't seem as stable as +50 , +70 results in crash no matters the memory clock.

Thanks
Marc

 

[nguyen]

Card probably went through mining for quite some time, don't expect the capacitors to give stable power to the chip, so yeah the chip won't be able to reach any high clock.

 

[rtwjunkie]

2 things, cooling and silicon lottery.

@trinxo What @P4-630 says is correct, although simplistic. First, you may have gotten a lower-quality chip in the silicon lottery. I’d guess the vast majority are average, not special.

Second: temperatures. Pascal must be kept under 60 Celsius or it will begin to throttle. Focus on cooling. All raising the power limit does is increase heat unless cooling increases as well.

Keep in mind, with that card the only thing you are guaranteed is 1645 on the boost clock. Anything else is a bonus.

 

[trinxo]

Card probably went through mining for quite some time, don't expect the capacitors to give stable power to the chip, so yeah the chip won't be able to reach any high clock.

Probably, I think i got a good deal for 300 EUR , but nothing is perfect. The card works well anyway.

 

[Vayra86]

My actual configuration:
View attachment 179664
I am testing the stability with heaven benchmark 4.0 now. The result of the benchmark does not change for 90 or 95 power limit, temperature is around 1 degree more hot for 95%, while playing cyberpunk the max temperature is 79 degrees and 78 for heaven benchmark. I tryed to open the pc case and tempertures are better, 72ºC for heaven benchmark, so i will have to figure out how to improve the airflow. What's weird is that yesterday playing cyberpunk with 107% of power limit I was getting 76 ºC (it could be an external issue like room temperature or different zone in the game ).

About the overclock, +60 doesn't seem as stable as +50 , +70 results in crash no matters the memory clock.

Thanks
Marc

1, Increase the temp limit to 92C. Test it for clocking performance. It might improve because now the card can just run on its hardware limits. It will still hard cap voltage when it hits 84C so no harm done.

2. If your card has the option, unlock the core voltage (that greyed out top slider) and give it a little kick, see what happens to clocks and OC capability now. Temps will suffer.

Beyond that you're pretty much out of options, don't go about changing BIOS please.

Probably, I think i got a good deal for 300 EUR , but nothing is perfect. The card works well anyway.

Well said, I'd agree especially in these times.

 

[trinxo]

Well, I set the vcore limit to +50% and keep the power limit at 95%. That's all I can do and I'm okay with that. I will look to improve the airflow and that will be all.

Thanks guys for your answers.

Marc

 

[mouacyk]

If you really want to crank this thing beyond 2GHz, you'll need an Accelero Extreme or AIO, if not going full water route. Keeping the GPU under 40C, I can get 2114MHz at 1.081v stable for 99% of games, and 2100MHz for 100% of them. Mine is essentially an MSI Gaming X, but not sure if they bin it better than the Armor edition.

 

[Fangio1951]

Well, I set the vcore limit to +50% and keep the power limit at 95%. That's all I can do and I'm okay with that. I will look to improve the airflow and that will be all.

Thanks guys for your answers.

Marc

hi trinxo,

I have a Gigabyte gtx1080Ti Extreme Waterforce card in my rig and it clocks to 2045mhz using Afterburner.

My system has a custom water loop and the card usually sits around 26C during normal work load = Photoshop, AutoCAD, etc.

It goes up to about 50-55C when gaming/stress testing/benching.

As others have said - I don't think you're going to do much better on air cooling + silicone lottery.

regards

 

[Nuckles56]

Well with my Gigabyte Aourus xtreme, I've never been able to go above +80, just because of thermals and silicon lottery, even though it is one of the top cards from gigabyte, so sometimes you get luckier then others with how good a card you get.

 

[Glitch_Neutrino]

Hi.

Using GV-N108TAORUSX-W-11GD-rev-1

Power target merely allow card to USE more power if needed. Please calculate available PRODUCTION.

Consult your PSU.

Strategy :
Fan maximum
Temperature Maximum
Voltage maximum
Power Maximum

Run test. (0 Core/Memory boost)
Note result.

Cont'n'd :
Fan maximum
Temperature Maximum
Voltage maximum
Power Maximum

Core +1 Mhz until crash, yes. Do 49 times if required.
Compare result.

You have ascertained stability.
Set fan to aggressive Auto.
Save in profile.

Future improvement only available with COOLING.

Current leakage in ALL electronics, that is why brain voltage low and activity compartmentalised.

All energy same.

You are all wrong. Pascal reduce clock @>25°C

 

[Dr. Dro]

This is a two year old thread, my friend. It might not be as relevant as it used to be. However, to address the timeless:

Core +1 Mhz until crash, yes. Do 49 times if required.
Compare result.

This is pointless, NVIDIA hardware operates in ~13 MHz clock steps (otherwise known as GPU Boost bins). This has been roughly the same since Kepler.

You are all wrong. Pascal reduce clock @>25°C

It's common knowledge that the lower the operating temperature of a processor, the better the frequencies it will achieve. I don't see how the others are wrong. You won't be achieving 25°C load temps on a GP102 unless you're using a chiller or even more exotic cooling methods.