1080Ti Bug - any advice please

[leekeene]

hello

i have an EVGA SC 1080ti. had it for a while and its been running fine but long story short i thought id try some more OC on it this evening and i seem to have hit a bug

scenario

Pc started logged in to windows and left for 10 mins... Open Heaven Benchmark 4.0 and set to extreme.... open EVGA Precision and start my OC

My OC settings in evga precision
Power = 120%
Temp = 90degrees set to priority (max temp seen is 60 degrees)
Volts = 100%
GPU clock = +75mhz
Mem Clock = +650

i start the bench mark and it crashed within seconds.... narrowed down to my GPU clock and anything greater than +50 mhz seems to crash my card. i also noted that my card is sitting at Total power draw at 110 to 120 most of the time during the test

Next

so after all this i loaded 3dMark... i dunno why i have always been drawn to Skydive benchmark. set it off with my +75 mhz GPU clock... runs like a dream and the power draw barely gets over 90. odd i thought but its an old test maybe something to do with that and not over taxing the card.... anyway i dunno what inspired me but i loaded Heaven Benchmark 4.0 with the same settings as before (+75muz gpu clock going straight from 3dmark skydive to heaven benchmark) and hey presto runs like a dream. power not going over 80%!!!

i reboot the machine and load Heaven Benchmark 4.0 again and back to crashing. load 3dmark skydive runs fine... then back to heaven bench mark runs fine

Video of the issue

https://youtu.be/6ZasB07SVGA

its repeatable.... and its confusing me.

anyone seen anything like this before?

any help is greatly appreciated

 

[snips86x]

Reduce the memory overclock or start from scratch. The card is crashing as the overclock is too much to be stable.

Also, have you set the priority to temp or power?

 

[leekeene]

Priority is temp

Yea I guess that's the total fix.. but why does running sky dive then fool the card in to running the increased overclock that failed previously?

 

[snips86x]

Priority is temp

Yea I guess that's the total fix.. but why does running sky dive then fool the card in to running the increased overclock that failed previously?

Set priority to power and run it again.

 

[leekeene]

Tried that but temp doesnt get above 60

It's almost like running skydive loads some setting that makes the card more efficient

I took a video of it but need to edit it lol

 

[leekeene]

https://youtu.be/6ZasB07SVGA

Edited the video so here it is

 

[deuse]

Priority is temp

Yea I guess that's the total fix.. but why does running sky dive then fool the card in to running the increased overclock that failed previously?

The only way to get the temps down on a 1080ti is to water cool it.
And I would have GPUz running at the same time and only use 1 monitor.

 

[leekeene]

Cheers maybe its not clear in my post. The 90 degrees temp is my overclock limit.. I am no where near that.

Max temp seen is 60 degree. I'll edit first post to show this.

I know the reason my card crashes on heaven benchmark is because my GPU clock is set to high.

What I am trying to understand is why if I run my failing setting from heaven benchmark on 3mark skydive not only does it run but total power limit is not hitting 120% and drops to more like 100%... 3mark passes and then if I rerun the heaven benchmark the highest power limit I see is like 80% even though its with the exact same settings, no changes.

Less than 4 mins previous, it wouldn't go further than 5 seconds before crashing

 

[Finners]

Your RAM speed seems too high for starters and you are not mentioning what core speeds you are actually seeing. Just using an offset will see you get different speeds under different loads.

On my titan Pascal I overclock using the voltage curve instead of an offset so I can control the power so I don't hit limits and run a fixed clock speed.

I run 1950Mhz @ 0.95v, when you just max everything on Pascal I think it will top out at 1.08v if I remember right and then you hit power limits.

Having a really high memory OC will add quite a bit to your power draw too.

As you are not stating clocks speeds what might be happening in skydive is that the card is actually crashing but recovering and the clocks will be staying lower from then on. That's why you can run heaven after until you reboot.

 

[deuse]

Cheers maybe its not clear in my post. The 90 degrees temp is my overclock limit.. I am no where near that.

Max temp seen is 60 degree. I'll edit first post to show this.

I know the reason my card crashes on heaven benchmark is because my GPU clock is set to high.

What I am trying to understand is why if I run my failing setting from heaven benchmark on 3mark skydive not only does it run but total power limit is not hitting 120% and drops to more like 100%... 3mark passes and then if I rerun the heaven benchmark the highest power limit I see is like 80% even though its with the exact same settings, no changes.

Less than 4 mins previous, it wouldn't go further than 5 seconds before crashing

I wouldn't worry about it.
But I would use GPUz and not rely on the bench mark programs
60 c is very good. My Titan X pascal used to reach 80c when playing a game until I watercooled it.

 

[leekeene]

Your RAM speed seems too high for starters and you are not mentioning what core speeds you are actually seeing. Just using an offset will see you get different speeds under different loads.

On my titan Pascal I overclock using the voltage curve instead of an offset so I can control the power so I don't hit limits and run a fixed clock speed.

I run 1950Mhz @ 0.95v, when you just max everything on Pascal I think it will top out at 1.08v if I remember right and then you hit power limits.

Having a really high memory OC will add quite a bit to your power draw too.

As you are not stating clocks speeds what might be happening in skydive is that the card is actually crashing but recovering and the clocks will be staying lower from then on. That's why you can run heaven after until you reboot.

GPU clock is hitting 2050mhz and memory clock 6156mhz

volts are hitting 1.094 down to 1.081.

i agree the memory clock does seem high although that doesnt appear to be my problem e.g i do a +50mhz on GPU clock i get to 2025mhz instead of 2050mhz and leave the memory clock at +650mhz which is 6156mhz then heaven benchmark runs fine

its literally something is happening when i run Skydive 3dmark. the power limit drops drastically and i get a solid 2050/6156 in any test i run. your comment in red above would you still expect it to be showing the higher clock speeds after it recovered?

literally just retested now. 2050/6156 fails with 5 seconds 2025/6156 passes the heaven benchmark

2025/6156 score = 4943 = FPS=196.2
my old OC 2012/6003 score = 4820 FPS = 191.3
2050/6156 score = failed, run skydive and then retest score = 1341 = FPS 53.2 this clearly falls inline with what you also mentioned in RED

what is the MAX memory clock i should be able to achieve?

I wouldn't worry about it.
But I would use GPUz and not rely on the bench mark programs
60 c is very good. My Titan X pascal used to reach 80c when playing a game until I watercooled it.

cheers i am surprised by this card considering its not even a FTW. i do run the fans pretty much flat out cos the noise doesnt bother me as i have noise cancelling headphones so maybe something to do with it?

 

[leekeene]

so did some testing this morning

2025/6156 seems to be my sweet spot.
i can go 2025/6264 and it runs fine but i start losing FPS and SCORE on benchmark. i presume thats because there are errors on the memory its having to resend?

 

[FlyingScotsman]

so did some testing this morning

2025/6156 seems to be my sweet spot.
i can go 2025/6264 and it runs fine but i start losing FPS and SCORE on benchmark. i presume thats because there are errors on the memory its having to resend?

Yeah it is memory errors that are causing the score to drop.

My 1080ti does 2012/6132 all day long but causes crashes in games if I go any higher on the core.