News Sagging GPUs Could Break VRAM on 20- and 30-Series Models: Report

[InvalidError]

I think 30 and 40 series GPUs might be immune to this. Not only do they have a very short PCB, but the design for the finstack and frame is basically a rigid body made of metal. I wonder if that is what prompted the design change knowing that they were going to use such massive coolers.

Keeping the PCB straight relative to the HSF is only half the battle.

No matter how short the PCB is, if most of the weight still rests on the PCIe slot, the same torque and shear force will be applied to the PCB connector edge and that could still be enough to pop the GDDR6X chips along the PCIe connector.

 

[Flyfisherman]

There is something here that don't adds up. A cold solder joint does not happen after the initial soldering.

The manufacturing of the board with automatic machines, starts with screen printing the solder paste (which includes flux in the solder paste), then mounting the components on the PCB (Printed Circuit Board) and lastly reflow soldering in an oven, that's makes a PCBA (Printed Circuit Board Assembly).
Sometimes samples of the PCBAs goes through inline or off line X-ray machine to check component placement and their solder joints.
If all the solder joints are well done - it cannot be a cold solder joint afterwards.

On the other hand if the manufacturing processes has problems, so that a cold solder joint occurs (happens immediately), then there will be one or more poor connections that will cause failure over time (latent failures), as it shows on the video, and any flex of the PCBA will further hasten this failure.

So yes it's important to not have any bending of the PCBA, but the main problem is poor quality in the manufacturing process that is causing cold solder joints from the beginning.

Best regards from Sweden

 

[InvalidError]

If all the solder joints are well done - it cannot be a cold solder joint afterwards.

He calls it "cold solder joint" because the solder separated from the BGA pad and whatever remained of the pads oxidized like they might from a cold solder joint.

The way oxidized pads still have a shiny ring around them could be an indication that plating got ripped off.

 

[russell_john]

GeForce RTX 2080 Ti's are dying from damaged GDDR6 connection points made from heavy GPU sag late in the card's life cycle.

Sagging GPUs Could Break VRAM on 20- and 30-Series Models: Report : Read more

Actually this has been a known problem for a long time, I experienced as far back as 2006. It's not just the sag that causes it but became a known problem with the advent of lead free solder which isn't as elastic as lead compound solder. Heat also plays a part as evidenced by the early PS3 and the need to 'reball' the main APU IC. While the lead free solders did get better they still aren't as good as the old lead solder compounds. I really didn't come across the sag problem again until 2019 when I bought a RX5700 and I had to custom 3D print a support which seems to have done the trick because My nephew has my old system and it's still going strong

 

[russell_john]

I think 30 and 40 series GPUs might be immune to this. Not only do they have a very short PCB, but the design for the finstack and frame is basically a rigid body made of metal. I wonder if that is what prompted the design change knowing that they were going to use such massive coolers.

With them it is more of a problem with the PCIe connector being stressed and causing problems over time. It can be alleviated by making sure the mounting bracket is fastened to the metal backplate in 2 or 3 places. With the unit he showed they attached the mounting bracket to the PCB instead of attaching it to the backplate so it didn't have enough support to stop it from sagging

 

[Deleted member 2838871]

I think 30 and 40 series GPUs might be immune to this. Not only do they have a very short PCB, but the design for the finstack and frame is basically a rigid body made of metal. I wonder if that is what prompted the design change knowing that they were going to use such massive coolers.

Time will tell. I did not have a support bracket with my 3090 FE... but the 4090 came with one and I definitely installed it.

It's solid for sure and I don't expect any issues. Cable clears the front glass too.

 

[Eximo]

Somehow I managed to leave out the founder's edition part of that...

 

[romeucapelasa]

Why not just fix the cooler to the backplate instead of the pcb it would relief alot of stress in that area if installed correctly

 

[InvalidError]

Why not just fix the cooler to the backplate instead of the pcb it would relief alot of stress in that area if installed correctly

Not attaching the HSF to the PCB isn't an option since it is necessary to apply contact pressure to everything that needs to transfer heat into the heatsink.

IO brackets are made by punching shapes out of a thin metal strip and then bending the tip over for the screw hole. Proper mechanical support would require having the screwed-in part extending at least across the PCIe slot, which means the IO part would need to be stamped as a separate piece that tees into the support plate and be fastened either with screws or welding.

Supporting the card properly (or at least 10X better than most of them currently do) would add maybe all of $3 to manufacturing cost.

Then again, not all cases use screws for the IO brackets. Some may not be able to properly secure a 20" long 5lb GPU no matter how much stronger the bracket may be.

 

[Colif]

Then again, not all cases use screws for the IO brackets. Some may not be able to properly secure a 20" long 5lb GPU no matter how much stronger the bracket may be.

this could help others

just a free work around