RTX 2080 Super - crashing, freezing, artifacting

[OneMoar]

There are other components that overheat too you know. Either way the mounting of the waterblock is definitely wrong, so a remount is in order...and yeah if the VRM is fried then there is nothing to salvage...

instantaneous crashing/artifact is not thermal thats a hard-failure
if the vrm was 'fried' it would be on fire or completely dead
you do not know what you are talking about you are taking a little bit of knowledge and making faulty assumptions about how this stuff works

its neigh impossible to damage a vrm(or core for that matter) on modern hardware simply by removing cooling unless you actively go and disable the multitude of thermal safety systems (and no raising the slider in msi ab doesn't count)

these cards are perfectly happy to run all day at 90c at 1600 to 1800mhz

 

[nguyen]

instantaneous crashing/artifact is not thermal thats a hard-failure
if the vrm was 'fried' it would be on fire or completely dead
you do not know what you are talking about you are taking a little bit of knowledge and making faulty assumptions about how this stuff works

its neigh impossible to damage a vrm(or core for that matter) on modern hardware simply by removing cooling unless you actively go and disable the thermal safety

these cards are perfectly happy to run all day at 90c at 1600 to 1800mhz

Yeah i'm gonna ask you the same really, have you ever deal with watercooling anything lol. My watercooling loop alone probably cost more than your PC...

I have had experiences with failing VRM with my R9 290X, yes the card could still run in windows but artifacted within seconds in games. VRM just don't combust into fire unless it reaches its thermal runaway point, although they still fail before reaching that if left alone for extended period of time.

Eitherway OP fack up the remount of the waterblock so everything was his fault, not the card...Although he could still try remount the Waterblock and see if the card still works.

Anyways just don't believe 60C is fine for GPU temp as it's only an average value, other area of the die could be cooking itself when the waterblock don't make proper contact with the GPU die, as indicated by the sudden rise in temperature. You can literally turn off your GPU fans and the GPU temp won't rise that quick if the cooler is making proper contact.

 

[Dinnercore]

You can literally turn off your GPU fans and the GPU temp won't rise that quick if the cooler is making proper contact.

Because an air cooler has a lot more mass and passive dissipation.

I did the remount, and I just did it again. The thermal paste was squished out to a fine layer as it should. It shot up in temperature like that from the day I got it, the fault is the waterblock design. Inno3D partnered with alphacool but alphacool did not want to cut into their own waterblock line-up so they took an existing design but cut a hole in the flow path so that most of the water is going in and out of the block without going across the whole area:



I took that picture back when I first used the card. It was already back then hitting 60°C under load, and that was with the rather flimsy X5690 CPU.

The block has stand-offs that touch the PCB when you screw the block on the card. These stand-offs limit how far the contact-areas can go down to avoid cracking the die. I can take a picture if you dont believe me, but all of these screw-standoffs touch the pcb so the cooler is mounted properly all the way down.

I'm going to RMA it, be polite and wait what they say.

 

[nguyen]

Because an air cooler has a lot more mass and passive dissipation.

I did the remount, and I just did it again. The thermal paste was squished out to a fine layer as it should. It shot up in temperature like that from the day I got it, the fault is the waterblock design. Inno3D partnered with alphacool but alphacool did not want to cut into their own waterblock line-up so they took an existing design but cut a hole in the flow path so that most of the water is going in and out of the block without going across the whole area:
I took that picture back when I first used the card. It was already back then hitting 60°C under load, and that was with the rather flimsy X5690 CPU.
The block has stand-offs that touch the PCB when you screw the block on the card. These stand-offs limit how far the contact-areas can go down to avoid cracking the die. I can take a picture if you dont believe me, but all of these screw-standoffs touch the pcb so the cooler is mounted properly all the way down.
I'm going to RMA it, be polite and wait what they say.

Nope, that is still a good full-cover waterblock design, what kind of pump do you have ?
My Heatkiller IV waterblock has the same design and it's one of the 2 best 2080 Ti wb out there

Yeah I wouldn't believe even with the worst wb design that a 200W GPU can shoot up from 25C to 56C in a few seconds.

Another question is did you put in the correct screws when remounting the wb, they could be different in heights...therefore not the correct pressure was apply to the core.
Here is how my GPU temp go up in Unigine Valley and my GPU idle at 31C (room temp 27)

 

[Dinnercore]

My Heatkiller IV waterblock has the same design

That is just plain wrong. It is not the same design, the Heatkiller blocks force the flow through all of the finstack while my block has a cut in the barrier close to the inlet and outlet. On my block the water can just turn right instantly, only going through 1/4th of the fins instead of the full row.
I had a Heatkiller IV on my Vega and I know they perform very well, I use their CPU block. I just did not want to install one on my GPU as I feared exactly this case with the RTX cards failing. A friend of mine had his 2080ti fail after a day, so I did not feel like spending additional money on a Heatkiller block that would instantly rip my warranty.

Another question is did you put in the correct screws when remounting the wb

I always lay out my screws in the order they belong on the card, but in this case they even are the same length all around. And again, the stand-offs touch the PCB, it phyiscally can't be any tighter on the card or I would break it.

what kind of pump do you have ?

I have a dual D5 reservoir with two D5s. They get the job done.

 

[Decryptor009]

It's dead jim.

Error 404

 

[Chrispy_]

If the RMA fails then you should do what der8auer does and put it in the dishwasher with hot water only for about 5 hours.

If you're using conductonaut or even just arctic silver, that stuff can get into places out of sight where it doesn't belong and can cause some truly inexplicable behaviour.

 

[Caring1]

If the RMA fails wait until the 4th of July and put it in your Microwave.

 

[delshay]

and yeah if the VRM is fried then there is nothing to salvage...

That's not true. I change MOSFETS here but I grab most of my parts from other dead devices. If I owed this card & it was out of warranty I would grab the parts from another similar card from EBAY or buy the MOSFETS direct, but I'm the kind of person that will fit even better MOSFETS (if available) & crank up the power, but as another user has pointed out in this thread, I don't think it's a VRM fault, but I could be wrong.

 

[Caring1]

My guess is the memory chips are faulty due to the heat.

 

[delshay]

My guess is the memory chips are faulty due to the heat.

I'm reading very carefully what the OP is saying. This card was working perfectly until he made modifications by upgrading to a better cooling system. So there was nothing wrong with card to start with, until changes were made.

I think it's possible when making changes OP may have crack solder joint somewhere as it looks like the OP put the card back together & now it's even worse.

& no, I never put a fully populated PCB in an oven, too many sensitive components. I only use that for reballing here, nothing else goes in there.

@Dinnercore

How are the memory modules cooled.

What is the state of the thermal pads (if any).

 

[xtreemchaos]

its clear to me that the card is duff, and its classic card breakdown most likey it has nothing to do with stripping down just badluck that it went at that time, ive heard of plenty with the same m.o. the sooner the op RMAs the sooner the good times come back .

 

[delshay]

its clear to me that the card is duff, and its classic card breakdown most likey it has nothing to do with stripping down just badluck that it went at that time, ive heard of plenty with the same m.o. the sooner the op RMAs the sooner the good times come back .

But the card was working perfectly. There was nothing wrong with it until the OP made hardware changes.

OP did something wrong that broke it.

 

[xtreemchaos]

yes mate but dodo happens, i RMA a card for a customer a month ago which was doing the same, thay sent a new one in 9 days and everythings good, id put it down to sods law it happen at the same time as he cleaned it out.

 

[Dinnercore]

How are the memory modules cooled.

What is the state of the thermal pads (if any).

The memory has grey pads on them, the very soft and squishy kind. Roughly 1.5mm thickness making direct contact to the waterblock.
It's all guesses from here, no one can be sure whats wrong without the card in hand and laboratory testing equipment. Even then it might need an X-ray to find cracked solder joints.

I strongly support the idea that I did something wrong, but there is always the chance that it might just have failed without my fault. Maybe there was a weak connection somewhere from the start and me moving things made it worse.

I have talked to the retailer, they said they will handle it. I can send it in and they will run a burn-in test suit that they originally run on the cards before shipping them. IF it fails the test there they will take a closer look at it to see if I did something wrong and if they don't find anything broken I will get a replacement immediatly and they handle the RMA with the manufacturer. I won't have to worry too much about the sticker either, as long as that is the only thing out of place on the card they will pass it.

 

[Dinnercore]

Update:

Full props to @nguyen as he was right with the proper connection issue. Also everyone questioning the mount were right I think.

Its gone through testing and an issue was confirmed. However the tester stated it was not the block or the mounting that was wrong. It was my paste. You can't use Kryonaut on this waterblock as it is too 'fluid'. You need a thick and dry paste that can not flow.

Official explanation: The block has a clearance betweend die and coldplate of 0.X mm, it is not complelty flush. So my paste when getting hot started to move slowly and created small pockets with bad contact. Eh. Sounds strange to me, I would connect this to a mounting issue. Maybe I tightned the screws in the wrong order.

Anyway I got the same card back, threw out the inno3D block and ordered a watercool heatkiller. I'm tired of the block that came with the card.

Temps are now better then the original OEM-block by ~6°C. Instead of using a backplate I taped some sensors to the back of the core and a VRM just to fight my newly found paranoia.

Card seems not to be bothered much.

 

[SpikeHob]

He has a watercooled 2080 Super, the idle temp in his image is 25C
I have a 2080 Ti watercooled and the temp rise to a maximum of 44C with maximum load in 20 minutes, not in seconds like his. Seems like the WB is not making proper contact...
Look closely at the sudden rise in GPU temp, no watercooled GPU would react that way

I have to agree with nguyen , those temps are seriously wrong . only time I have seen temps act like that was my friends water-cooled 1080 when the pump failed . I can overclock the snot out of my RTX 2080-Ti , and even on a warm day ( well warm for England ) temps never go higher than very low 50 deg C . There is definitely something wrong with block/ pump . I would RMA it

oops did not see your update , my apologies ( walks away shame faced , head down )

 

[nguyen]

@Dinnercore Nice that see that your card is working fine, and with a better Waterblock too , usually with thermal paste applications, I load up the GPU or CPU to get them warmer then apply a bit of rotation to the screws just so that the cooler is making good contact once the thermal paste and thermal pads heats up, maybe being a bit OCD with cooling paid off in the long run ^^.

 

[Chrispy_]

Why the hell would you design a block so that the coldplate doesn't make contact with the silicon? That seems beyond stupid, unless the goal was always to use one of those gum pads or intentionally use too much goop in there.

 

[SpikeHob]

maybe trying to multi-purpose a block , save on manufacturing costs ?

 

[Dinnercore]

Why the hell would you design a block so that the coldplate doesn't make contact with the silicon? That seems beyond stupid, unless the goal was always to use one of those gum pads or intentionally use too much goop in there.

I wish I knew, I was feeling scammed from the start. The picture on the box of the card shows a completly different block design then what is in the box. I bought it because of the product pictures that show a neat guided flow path and got that crap thing with the weird cut-down flow guides.

I mean, compare these two images:

AD:

My block:

That is borderline false advertising.

 

[Chrispy_]

yeah, the change in block is really bad. looks like a bodge-job hotfix for an underperforming pump or something. why else would you want to short-circuit the path unless to reduce flow restriction on a cheap-ass pump that's not up to the job of the original design?

Having the coldplate not making contact with the die is another facepalm. If you're still in the RMA window I'd be sending it back. Nobody goes full waterblock unless they want decent performance and a rubbish block with an air-gap to the die is the exact opposite of decent performance, lol

 

[nguyen]

This is the exact reason why AMD use hotspot temperature for performance and fans tuning. With legacy temp reading base on only 1 sensor, the temperature reading could be 60C at the sensor location while the other part of the core is 200C, which cause instant artifacts...

 

[Chrispy_]

This is the exact reason why AMD use hotspot temperature for performance and fans tuning. With legacy temp reading base on only 1 sensor, the temperature reading could be 60C at the sensor location while the other part of the core is 200C, which cause instant artifacts...

Nvida GPUs have more sensors too, they just don't expose those temperatures to the user. The Navi temperature monitoring setup is really impressive though. Junction temps for fan control allow some really aggressive overclocking, if that's your thing.

Everyone got alarmed when the new 5700XT runs at 105C and everyone was going OMG SO HOT, but of course every GPU in the last decade has been running just as hot, if not hotter. AMD's average temps and target average temps are still what they've always been at 75-80C on desktop cards.

 

[cucker tarlson]

Everyone got alarmed when the new 5700XT runs at 105C and everyone was going OMG SO HOT, but of course every GPU in the last decade has been running just as hot, if not hotter

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 ?