You're gonna wanna brace yourself for this - GPU sag might be killing RTX cards

[BachChain]

Summary

 

Tech repair YouTube channel KrisFix has reported receiving an increasing number of RTX 2080tis suffering VRAM damage. He attributes this damage with heavy GPU coolers causing excessive amounts of flex on the bottom of the card, straining the memory chips located there, and eventually breaking the solder joints.

 

Quotes

Quote

They have one very common problem. and this common problem is right here: the PCIe slot. And the problem comes from there that these three ram chips are very close to the PCI slot. When the card is plugged into the motherboard, in the front, there are one or two screws holding the card, here in the middle we have the motherboard, and on the back side we have nothing. And what happens is all this weight from here causes pressure right here, and from this pressure the card bends very slightly, and from this bending some of the solder joints under these three ram chips crack, and after that they they don't have a good connection between the board and the chips.

 

My thoughts

I'm surprised that 1) this hasn't become more of a problem sooner, and 2) including card brackets hasn't become standard practice for these monster cards.

 

Sources

 

https://www.tomshardware.com/news/rtx-2080-ti-dying-from-gpu-sag

[GuiltySpark_]

This the same guy that said RX6000 cards dies were cracking due to drivers? 
 

IMO this is a non-story. These cards have been out for over two years and preventing sag has well known prevention options. 
 

Fake drama is boring. 

[IkeaGnome]

23 minutes ago, GuiltySpark_ said:

This the same guy that said RX6000 cards dies were cracking due to drivers? 
 

IMO this is a non-story. These cards have been out for over two years and preventing sag has well known prevention options. 
 

Fake drama is boring. 

Yes. Same guy. 
If this was about 3000 series, I could get behind the story. Weird he’s doing about sag and not dying vram chips…..

[jaslion]

Severe gpu sag can cause thermal pads to stop making contact. I see this yearly a couple times. But we are talking about severe sag where its visibly detaching.

 

As always I recommend people prevent it but normally its a non issue for the lifetime of the card.

 

I dont doubt a handfull of cards are epxeriencing this issue but unless it gets real common I dont see this being a problem.

[Tan3l6]

Thanks for reminding me - I've got a high tech stick to hold the gpu tail up:

https://www.deepcool.com/products/Accessories/graphiccardaccessories/2021/13758.shtml

 

Just put it to use - works it seems, for now.

[Quackers101]

we have always known that trend of sag has not been good or pressure on the PCIe slot. so not much news.

gpus be goofy sports cars with the price to match

[thrasher_565]

solution buy a $10 stand... or make your own... you think would be worse on 30x and 40x?

[Quackers101]

57 minutes ago, thrasher_565 said:

solution buy a $10 stand... or make your own... you think would be worse on 30x and 40x?

40x being a 3-4 slot, already hitting the floor

just water cool the floppy fish of a card.

Edited April 14, 2023 by Quackers101

[Kisai]

2 hours ago, BachChain said:

 

 

My thoughts

I'm surprised that 1) this hasn't become more of a problem sooner, and 2) including card brackets hasn't become standard practice for these monster cards.

 

The problem is that few, if any computer chassis actually have the mounting points for full length cards in the right place, if at all, because that space is often taken up by 3.5" hard drive bays. The original PC/XT/AT chassis had them, because most of the expansion cards for the original PC/XT/AT were full length cards. The spec is defined as 312mm (12.283 inches.)  

 

Yet...

 

A popular case like the Lian-Li O11 doesn't have these brackets because they have glass front panels.

 

That said, aside from the RTX x70/x80/x90 parts of the 30 and 40 series, it's really going to depend on the card design (which are all 3 slot designs.) The high end RTX cards, if they are FE cards, or ASUS TUF cards are unlikely to experience this problem just because of the rather solid mounting (eg the cards have entire backplates) so it's unlikely to sag if it has a backplate.

 

But my guess is there are probably some cards out there where the backplate doesn't cover the entire PCB, only the GPU die back side. Something to keep in mind if your transport your PC around, but probably not going to be encountered that often.

 

It's not like this is real or anything:

(This is an April Fools video, in case anyone visits this thread from outside the forum)

 

Because the FE cards don't actually have a PCB that goes end-to-end so it can have the type of cooling fan it uses, there is probably some potential for sag, if only due to surface area (Eg a cooler mount is spread over less area.)

 

But I don't imagine this to be an issue unless something like that April Fools joke video ends up becoming the reality for GPU's.

 

 

 

[da na]

Haven't 2080 series cards had VRAM problems for years? This is nothing new. Also, GPU sag is easily preventable with literally any moderately long stiff object

[Arika]

my last 2 cases have been horizontal mount cases. never have to worry about sag ever!

[da na]

3 minutes ago, Arika S said:

my last 2 cases have been horizontal mount cases. never have to worry about sag ever!

I just don't own any GPUs big enough to sag lol, RTX A4500 isn't the heaviest of cards compared to the 4 slot monstrosities we see now

[Arika]

Just now, da na said:

I just don't own any GPUs big enough to sag lol, RTX A4500 isn't the heaviest of cards compared to the 4 slot monstrosities we see now

and my a4000 is single slot

[da na]

1 minute ago, Arika S said:

and my a4000 is single slot

Based pro card user

 

I also have a Tesla P4 accelerator that is basically a 75w single slot GTX 1080 lol

[StDragon]

Ever take a paperclip and bend it back and fourth several times? It eventually breaks from metal fatigue from the constant thermal cycling of being off (room temperature) to pull 100% duty cycle of gaming sessions. Eventually the solder points will crack.

This isn't a new thing, and in fact very common with laptops with powerful GPUs.

[tibtob]

Now i really want a horizontal case for all these new GPUs coming out. 

[Kisai]

21 minutes ago, StDragon said:

Ever take a paperclip and bend it back and fourth several times? It eventually breaks from metal fatigue from the constant thermal cycling of being off (room temperature) to pull 100% duty cycle of gaming sessions. Eventually the solder points will crack.

This isn't a new thing, and in fact very common with laptops with powerful GPUs.

True, but usually the cooler itself fails before it gets there.

[Deadpool2onBlu-Ray]

Oh no! Anyway....

[DildorTheDecent]

44 minutes ago, da na said:

Haven't 2080 series cards had VRAM problems for years? This is nothing new. Also, GPU sag is easily preventable with literally any moderately long stiff object

The initial Micron equipped cards had artefacts. Samsung cards however are fine.

[MageTank]

1 hour ago, Kisai said:

The problem is that few, if any computer chassis actually have the mounting points for full length cards in the right place, if at all, because that space is often taken up by 3.5" hard drive bays. The original PC/XT/AT chassis had them, because most of the expansion cards for the original PC/XT/AT were full length cards. The spec is defined as 312mm (12.283 inches.)  

 

Yet...

 

A popular case like the Lian-Li O11 doesn't have these brackets because they have glass front panels.

 

 

That said, aside from the RTX x70/x80/x90 parts of the 30 and 40 series, it's really going to depend on the card design (which are all 3 slot designs.) The high end RTX cards, if they are FE cards, or ASUS TUF cards are unlikely to experience this problem just because of the rather solid mounting (eg the cards have entire backplates) so it's unlikely to sag if it has a backplate.

 

But my guess is there are probably some cards out there where the backplate doesn't cover the entire PCB, only the GPU die back side. Something to keep in mind if your transport your PC around, but probably not going to be encountered that often.

 

It's not like this is real or anything:

(This is an April Fools video, in case anyone visits this thread from outside the forum)

 

Because the FE cards don't actually have a PCB that goes end-to-end so it can have the type of cooling fan it uses, there is probably some potential for sag, if only due to surface area (Eg a cooler mount is spread over less area.)

 

But I don't imagine this to be an issue unless something like that April Fools joke video ends up becoming the reality for GPU's.

 

 

 

This has already been solved, ironically by Lian Li, the manufacturer of the O11 Dynamic, lol.

 

https://lian-li.com/product/gb-002/

 

This bracket is actually installed through the motherboard standoffs on standard ATX motherboards (the two bottom right standoffs). The bracket is adjustable both vertically and horizontally to avoid fans and contact the card itself.

 

This design is far more universal than the ones coming from card manufacturers since the ATX standard has been around for years. ITX or Micro ATX boards won't be able to use it, but you are far less likely to pair a 5lb 4090 with a micro or ITX board/chassis. At that point, I'd recommend the Bracing Retention Apparatus, or "BRA":

 

[StDragon]

2 hours ago, Kisai said:

True, but usually the cooler itself fails before it gets there.

Regarding, laptops, not in my experience at least. Though a fan failure being the primary would be true of crypto mining GPUs were the temps remain at a constant load.

And while I don't advocate for the use of lead, from a physics perspective, RoHS (lead free) compliant solder isn't as durable under extreme thermal cycling; and the industry has switched over to RoHS some time ago.

[starsmine]

2 hours ago, StDragon said:

Ever take a paperclip and bend it back and fourth several times? It eventually breaks from metal fatigue from the constant thermal cycling of being off (room temperature) to pull 100% duty cycle of gaming sessions. Eventually the solder points will crack.

This isn't a new thing, and in fact very common with laptops with powerful GPUs.

that requires bending it beyond the yield point. 

Metal bends and comes back just fine without fatigue (elastic deformtion). Its onces you make a plastic deformation that you have now changed the structure and made that area harder/more birttle. 

 

I would have to look further into the OP story, but given their history im suspicious about his heuristics that he uses for troubleshooting and overconfidence. 

[StDragon]

8 minutes ago, starsmine said:

that requires bending it beyond the yield point. 

Metal bends and comes back just fine without fatigue (elastic deformtion). Its onces you make a plastic deformation that you have now changed the structure and made that area harder/more birttle. 

That is incorrect! Solder fatigue is absolutely an issue under extreme thermal cycling. If it wasn't then there would never be a need to reflow the solder using either a heat gun or the oven method.

Note: specifically in regards to BGA ICs

[starsmine]

Just now, StDragon said:

That is incorrect! Solder fatigue is absolutely an issue under extreme thermal cycling. If it wasn't then there would never be a need to reflow the solder using either a heat gun or the oven method.

yes, but that has nothing to do with your paper clip example or sag

[StDragon]

5 minutes ago, starsmine said:

yes, but that has nothing to do with your paper clip example or sag

It does, it's has to do with the crystalline structure within the metallurgy.

Here, right from Wiki
 

Quote

A disadvantage of BGAs is that the solder balls cannot flex in the way that longer leads can, so they are not mechanically compliant. As with all surface mount devices, bending due to a difference in coefficient of thermal expansion between PCB substrate and BGA (thermal stress) or flexing and vibration (mechanical stress) can cause the solder joints to fracture.
 

Thermal expansion issues can be overcome by matching the mechanical and thermal characteristics of the PCB to those of the package. Typically, plastic BGA devices more closely match PCB thermal characteristics than ceramic devices.
 

The predominant use of RoHS compliant lead-free solder alloy assemblies has presented some further challenges to BGAs including "head in pillow"^[3] soldering phenomenon, "pad cratering" problems as well as their decreased reliability versus lead-based solder BGAs in extreme operating conditions such as high temperature, high thermal shock and high gravitational force environments, in part due to lower ductility of RoHS-compliant solders.^[4]