Rtx 2080 Ti - small overclock crash in firestrike

[andrewiksman]

Halo. Meine karte 2080 Ti Aorus . Stock boost 1695mhz.

I am currently ocing testing Fire strike gpu test 1.
Here is my settings:
Power limit max - 133%
Temp limit - 88C
Voltage - 0+ or +100
Memory - 800+ or 0+

Stock boost 1695mhz. When i dont overclock core +0 ,clocks during use is 1980mhz-to 1890mhz when temps go to max 87C in firestrike gpu test 1 loop.
And 3dmark and games are stable.
But when i overclock core +65,clocks are 2050-1950 and stable in all games. But when i loop for 3-4 hours normal firestrike gpu test 1, it get crash after 3 hours on 1950mhz.


When i dont overclock,and left stock boost in 1695 but only max power limit to 133% my max boost is 1980 at start and never get crash.

But like i said when i overclock core even 50-75+ core i can get crash in looped firestrike gpu test 1 after few hours. Clock during crash is like 1920-1950mhz.
What to do now?
Graphic card bios is F3 updated. Also i am using FAN 100%. Card is plugged by 2 pcie cables each 8 pin.

 

[kastriot]

What is the point overclocking this beast?

 

[andrewiksman]

Because all peoples telling that 2080 ti overclock easy on 100+core and clocks average 2000-2100mhz. But mine running 1900-1950mhz stable.

 

[Vayra86]

Sometimes less is more. You are temp limited, 87C is far too high. With Pascal/Turing cards, what really matters is your clockspeed at temp. equilibrium. So in your case, your actual OC is 1890Mhz. Use that for your comparisons with new settings, then follow this method:

> Use Power Target = 100% and try again. Memory +500. Core +100. See what happens. If you crash, go to 110% power target and try again. If you run into temps anywhere over 80C, drop back to 100% and use a lower Core OC, such as +80. Work your way through like that and you will find your max stable OC. High temp will still be an issue though. If you can only keep temps under control at 100% fan speed... painful. It might be a good idea to drop some power in favor of lower fan speeds/temps, this is a personal choice.

Aaand we have another example of a Gigabyte GPU with shit cooling If you can still return it, might be a good idea.

 

[Naito]

Could be just the luck of the silicon draw. Have you tried the OC scanner? It generally gets very close to what manual overclocking will get you

Edit: From the TPU reviews, the average max boost clock over 8 samples is ~2106MHz, while memory is 2010MHz

 

[andrewiksman]

Yeah i try oc scanner and after apply profile i have 1770mhz ( stock 1695mhz ) but my clocks are lower than if i manually make 1770mhz offset.
So oc scanner profile adding 1770mhz but this is not real like i add manually. Because i have 1905mhz max minimum clock when temp go to 87C in firestrike loop.


Also in games my temps are max 75-80C. Only firestrike gpu test 1 loop making card 87-88C. Weird.
And like in topic i have crashing only in firestrike gpu test 1 after few hours when i add some oc,even +30core and crash.

 

[Vayra86]

Yeah i try oc scanner and after apply profile i have 1770mhz ( stock 1695mhz ) but my clocks are lower than if i manually make 1770mhz offset.
So oc scanner profile adding 1770mhz but this is not real like i add manually. Because i have 1905mhz max minimum clock when temp go to 87C in firestrike loop.


Also in games my temps are max 75-80C. Only firestrike gpu test 1 loop making card 87-88C. Weird.
And like in topic i have crashing only in firestrike gpu test 1 after few hours when i add some oc,even +30core and crash.

Your card is unable to keep a core clock at high temps, its probably a leaky chip and high temp will cause it to fall short. While it is Fire Strike crashing on you now, that is only because it presents a 100% load and your games are likely not a fulltime 100% load, so you get lower temps and lower temps = stable.

What this means is your OC is simply not stable. Its not weird, but very normal to work like this.

In addition, you are temp limited and already at 100% fan speed. So your ONLY option is to drop voltage/lower power target. You cannot keep your OC by adding more power, because the GPU simply doesn't offer it, it actually wants to REDUCE power to push temperature back under 80C.

 

[andrewiksman]

Ok i will try on lower power limit 100 or 120? But clocks are more fluctuating

 

[Naito]

Also in games my temps are max 75-80C. Only firestrike gpu test 1 loop making card 87-88C. Weird.

Sustaining maximum clocks and maximum temperature threshold while at what I assumed to be very high utilisation level, is going to take a toll on the chip. I'd stick with the OC Scanner profile, as it gives decent results while being unlikely to cause long term degradation.

On the other hand, as Vayra86 mentioned, games rarely maintain 100% utilisation over long periods, so if you're game enough, perhaps keep your maximum clocks, as the card seems to cope at the lower temperatures, but tolerance of said clocks may slip over the lifetime of the card. Nvidia boost will keep everything else in line.

 

[Vayra86]

Ok i will try on lower power limit 100 or 120? But clocks are more fluctuating

Read again. You haven't got a choice, you are temperature limited, any temperature above 84C is far too hot and costs a LOT of performance - more than you gain from an OC. You've seen that already, you get 1980mhz on stock and all that extra power only gets you to 2050mhz *peak*, and you cannot even hold that clock very long. So what you are doing right now is pump a shitload of volts into the chip, get it warmed up fast, and then lose your OC because of that extra power.

Start with 100% so you know what the card can do with stock voltage in terms of boost and clocks. Work your way up from 0 > 20/40/60 Core or down from +100 > 0 in steps of 20.

As for the OC scanner... meh. It won't help your temperatures at all, not something I'd do in this case because it will still apply an overclock that does not take into account your actual temp at max load.

 

[andrewiksman]

Really appreciate for help. I add 70+core but power limit 120% and temps max 79-80C. Clocks fluctuating a little but its ok. I left now running and see if crashes.

 

[Vayra86]

Good luck. All the info is there now, you can play around with it

 

[andrewiksman]

yeah i am very greatful for your answers . You are experienced users. I will take your knowledge and play with settings of gpu. Thank you for explaining

 

[Naito]

As for the OC scanner... meh. It won't help your temperatures at all, not something I'd do in this case because it will still apply an overclock that does not take into account your actual temp at max load.

Won't help with temps, but will probably help squeeze a safe bit of performance by tuning the clock profiles to the specific card and ASIC quality. At the end of the day GPU Boost dictates the clocks. If OP can get the temperatures down, it may reduce voltage leakage, which in turn will generally drop power and voltage requirements, and thus allow GPU Boost to aim for higher clocks.

Basically, keep the card as cool as possible and GPU Boost will do the rest.

 

[andrewiksman]

Last question. So if in firestrike temp 79-80C with power limit 120% and clocks are 1920-1980mhz its ok? Its not the worst 2080 ti example?

 

[Naito]

Last question. So if in firestrike temp 79-80C with power limit 120% and clocks are 1920-1980mhz its ok? Its not the worst 2080 ti example?

I haven't had any experience with Turing, but seems within expectations. Every chip is different. If you have GPU-Z open and logging, monitor the PerfCap Reason value, this will tell you what aspect of GPU Boost is currently holding back clocks. For example, there is no point throwing extra power at it, if the limit is voltage...

 

[Vayra86]

Last question. So if in firestrike temp 79-80C with power limit 120% and clocks are 1920-1980mhz its ok? Its not the worst 2080 ti example?

Compare it to your stock example of the topicstart. You haven't gained anything meaningful from +20% power.

Again: temp limited. Its probably not the worst 2080ti example, but its not far off, and the problem is not the chip but the cooler around it. Good candidate for a reseat/repaste and/or replacing the cooler altogether.

 

[andrewiksman]

gpuz showing only green PWR at 120%. So left like it is? Core is +85
Clocks are 1920-1980 balancing with low voltages 0.900-1.030.


Temps 80C. Not 88C like with 133% power limit.With 133% pwr/vrel gpuz was showing. But it was not stable even if i add 50+core after few hours loop.And temp was 88C

Okies it crashed on power limit 120% on +85 core ( 1780 boost ). So what now turn down power limit or clock -10?

 

[Vayra86]

You really need to just scroll up and re-read my earlier posts about what to do. It still applies, and that never changes. We can't sit next to you every step of the way.

Its a very simple principle:
- more power = higher temps = lose boost clocks
- high temps = require more power to keep the same clock

Your adding power has no use whatsoever. You need less power, not more, and then see what clocks you can get out of it.

 

[trog100]

turning the power limit up allows the card to boost higher but it dosnt make it more stable at the higher boost speed..

you can have a stable card at 100% power.. turn the power limit up and it will crash..

your problem is you have the power limit set too high there really isnt much point in running at 133%.. it just heats up the card and lowers the amount you can raise the core clock before the card becomes unstable..

these cards have enough grunt not to need running on the ragged edge except maybe for benching and for benching you dont need long term stability..

trog

 

[londiste]

https://www.aorus.com/GV-N208TAORUS-X-11GC
This one?
One of the heaviest factory-OCd 2080Tis. 2080Ti stock boost clock is 1545MHz. Yours has 1770MHz. It has a hefty overclock already.
This specific model takes 300W at stock. The cooler on it is not bad but increased power limit and clocks can easily drive it over the edge of coolers ability.

OC scanner is very very useful if you want the card to end up stable. It tends be be awesome for that. While manually increasing clocks you will get higher overclock, it is likely to have some instability in specific applications. My 1080Ti at 2050MHz crashes in Firestrike and some games. Similarly, my RTX2080 at around the same frequency crashes in Timespy and some games. With the curve from OC Scanner (that is a bit lower, about 40MHz), no OC-related crashes any more.

Problem with adding power limit is that GPU Boost thing on Nvidia cards can see additional headroom and tries to boost to there. This will cause crashes, in most cases due to insufficient voltage at these clocks.

 

[Tatty_Two]

Sorry guys, going to have to shut this one down, it is our old friend Litwicki from Poland and his 38th account Appreciate all your efforts and help.