RTX 2080 Super - crashing, freezing, artifacting

[nguyen]

really ?
is that verified ?
every gpu as hot or hotter ?
is my 2070 super trio hitting 105 degrees when the software is reporting 60-61 ?

Could be, you never know. My guess is Nvidia use an average value from all the sensors. That's why when you re-paste and temperature seems higher than it should be but well within the limits, and the card exhibiting weird behavior then it could be that the core is not making proper contact with the cooler.

I have tried re-pasting many times before and found that the pea and line method don't work with GPU (due to lower mounting pressure on the GPU), the temperature would just rise by 1-2C compare to manually spreading the paste but when I took off the cooler, a part of the core is not covered with thermal paste, and there would be some shuttering in-game.

Moral of the story is don't trust the temp reading too much...Also the card can die if its VRAM or VRM is overheating, which Nvidia is not monitoring. If I remember correctly the early batch of Turing have had high failure rate due to either VRAM over-heating or bad solder joints...

 

[cucker tarlson]

Moral of the story is don't trust the temp reading too much...Also the card can die if its VRAM or VRM is overheating, which Nvidia don't monitor...

true
not gonna happen on better custom versions though.

 

[Chrispy_]

really ?
is that verified ?
every gpu as hot or hotter ?
is my 2070 super trio hitting 105 degrees when the software is reporting 60-61 ?

EVGA confirmed in an interview with the Verge that Geforce 10-series use an average result from 10 sensors on the die. If Nvidia's target temperature is set to 80C, then parts of the chip must be hotter than that average because GPU logic is not uniformly active, uniformly clocked, or uniformly dense across the whole die.

Taking the 5700XT as an example, the average delta is 20-25C for me so 105+C hotspots are pretty much guaranteed for an Nvidia GPU if the same delta applies, and I see no reason why it shouldn't.

Your 2070 super frio has an overbuilt cooler for its power limit because it's not reaching the temperature target defined by the nvidia driver - which for a 2070S is 83C, I believe. Either raise the power limit as high as it'll go for more performance, or do what I do with my 2060S and reduce the fan speed because there's no need to suffer more noise than necessary if the GPU is cooler than it needs to be.

Remember, the coolers, fan curves, and target temperature are all designed to survive the warranty duration in much hotter countries. If you have the luxury of an air conditioned room or a cooler climate, you can take those 'free' 20C and use them for your own benefit, either a faster or quieter GPU.

 

[nguyen]

EVGA confirmed in an interview with the Verge that Geforce 10-series use an average result from 10 sensors on the die. If Nvidia's target temperature is set to 80C, then parts of the chip must be hotter than that average because GPU logic is not uniformly active, uniformly clocked, or uniformly dense across the whole die.

Taking the 5700XT as an example, the average delta is 20-25C for me so 105+C hotspots are pretty much guaranteed for an Nvidia GPU if the same delta applies, and I see no reason why it shouldn't.

Your 2070 super frio has an overbuilt cooler for its power limit because it's not reaching the temperature target defined by the nvidia driver - which for a 2070S is 83C, I believe. Either raise the power limit as high as it'll go for more performance, or do what I do with my 2060S and reduce the fan speed because there's no need to suffer more noise than necessary if the GPU is cooler than it needs to be.

Remember, the coolers, fan curves, and target temperature are all designed to survive the warranty duration in much hotter countries. If you have the luxury of an air conditioned room or a cooler climate, you can take those 'free' 20C and use them for your own benefit, either a faster or quieter GPU.

I did some trouble shooting for my friend's PC and he has the Strix 2070S, at default the card already boost beyond the capability of the chip (2070mhz/1.05V) and it was unstable (ain't silicon lottery a bi*ch). So I set -30mhz clock offset so that the card would boost to 2040mhz/1.05V and it would run stable all day long. Sometime manufacturer overclocked BIOS are just a bit too much for the chip and this would off course be an instant RMA if I had not been there .

I also have the Aorus 2060S in my spare PC, that thing has an insane max power limit of 310W, the chip would never use that much even at 1.093V though.

 

[Vayra86]

Moral of the story is don't trust the temp reading too much...Also the card can die if its VRAM or VRM is overheating, which Nvidia is not monitoring. If I remember correctly the early batch of Turing have had high failure rate due to either VRAM over-heating or bad solder joints...

VRAM and temps are seriously underestimated. Those ICs don't like heat at all and they are sandwitched between power stages and the die.

As for the VRM temps... its not uncommon but when you see FLIR imagery of Nvidia cards you don't usually see hot spots over 90 C. Not even on an FE. When Toms Hardware DE tested the EVGA GTX 1080 FTW they saw hot spots over 100C... and cards died and EVGA issued thermal pad fixes.

So no, 105C hot spot is not nice and it will not make the 5700XT more durable, let's not beat around the bush here.

The bigger picture: its not a secret the silicon is pushed closer and closer to the limit and power delivery/temp management is super tight these days. These things will and actually DO happen more often, and will keep happening in the future. Why do you think Turing has all those super fat coolers?

ASUS GeForce RTX 2080 Super Dual EVO OC review

We review the GeForce RTX 2080 Super Dual EVO review from ASUS. The super card has been beefed up, received a hefty cooler and a moderate tweak. Will it be enough the dress and impress?... Graphics Card Thermal Imaging Measurements (FLIR)

www.guru3d.com