News Defective Vapor Chamber May Be Causing RX 7900 XTX Overheating Issue

[zx128k]

Makes me think of all those thermal pad mods on the nVidia RTX 3090 to fix GDDR6x memory temps (not just mining, some people reported 110c temps in games). I got an AIB RTX 3080 TI with better memory cooling and had no issues from the start. Sounds like AIB models with custom cooling are the way to go with the AMD RX 7900XTX. Imagine getting a card over $1000 with these kind of problems. Then have to pay another $25 to ship it back for RMA.

 

[digitalgriffin]

Makes me think of all those thermal pad mods on the nVidia RTX 3090 to fix GDDR6x memory temps (not just mining, some people reported 110c temps in games). I got an AIB RTX 3080 TI with better memory cooling and had no issues from the start. Sounds like AIB models with custom cooling are the way to go with the AMD RX 7900XTX. Imagine getting a card over $1000 with these kind of problems. Then have to pay another $25 to ship it back for RMA.

The custom AIBs are always a lot more expensive and the reference is too expensive as is. So $&#-_ it!

 

[digitalgriffin]

That's one of the first departments to get trimmed down when companies get big to the point that moving volume starts to overshadow quality.
Even if publicity on a product is poor, it doesn't really hurt if profits are still being made.

Where's that statement about, "There's no such thing as bad products?"

Did we watch the same video? He had the card completely upside down, which did not work. Then he tried vertical on its side which worked OK until he turned it completely upside down again so the water would get trapped in the cold parts, after which it did not work when vertical again because the liquid was stuck so far away from the heat source. At no point did he try with the heatsink right side up which would drain the liquid to back over the die.

Heatpipes without wicks at all have to have the heat source at the lowest point in order to work. And unfortunately ATX puts the GPU die at the highest point when in a tower case.

Heat pipes are different than vapor chambers.
All heat pipes have a wick due to the copper sintering acts as a wick. Although it works in any position through natural hot cold convection, it does work better when the hot plate is at the lowest point. But the natural convection sources do overcome gravity.

The vapor chamber issue sounds more like potential boil off due to improper dual point set. A triple point is the specific point at which a substance can be all 3 phases at a very specific temp pressure volume. A dual point is the line it stradles during a temp pressure curve. If the liquid is the wrong amount it throws that curve off and creates imbalance and you get outside proper bounds.

That's 200 level thermodynamics for you. It's a classical constant volume heat problem. Energy turns liquid to gas. It expands and increases pressure which raises the boiling point. Too little liquid and there's not enough pushback to prevent liquid on hotplate from evapping. You use Boyle's laws to calculate the equilibrium.

 

[Ogotai]

How so?? The roughly 50 cases of the melting connector were from all user error from not properly plugging in their GPUs

sorry, but 50 cards with melting connectors, it NOT user error, but a design flaw. user error would be 10 or less.

Worse melting still occurs with regular PCIe connectors.

i have never had a conenctor melt with the 6 and 8 pin connectors, and it seems like no one else has, either.

 

[InvalidError]

That's 200 level thermodynamics for you. It's a classical constant volume heat problem. Energy turns liquid to gas. It expands and increases pressure which raises the boiling point. Too little liquid and there's not enough pushback to prevent liquid on hotplate from evapping.

Well, you do want the liquid to boil/evaporate from the hot plate so it can carry the phase transition heat away. The problem with too little liquid is possibly ending up in a situation where not enough of it remains in liquid phase to make its way back through wicks..

From Derßauer's vapor chamber milling video, the whole thing appears to be one single giant chamber. If so, there is a possibility that even if the amount of liquid is correct, the vapor may condensate and accumulate in a cold spot, starving the hot spot.

 

[systemBuilder_49]

It is only 110C when the vapor chamber is in a failed state, ~86C otherwise. The 50M$ question is whether the problem is a systemic design issue, a problem with specific batches (incorrect fill amounts, incorrect chamber pressure, bad seals, contamination, etc.) or seemingly random.

Re: implies that the card is energy-efficient because even when the vapor chamber is in a failed state the heat peaks at 110C?

This is an incorrect assumption. When the vapor chamber fails, it's 110C and the card is slowing down slowing down slowing down until its running so slowly the temperature is reduced. It might be at half-speed, even one-third speed when it's a 110C in order to keep from exceeding that number. Some people report their card powering off.

Most vapor chambers create a miniature climate where the card boils off water, which travels as vapor to the far reaches of chamber, where it condenses, and the water runs downhill back towards the gpu to be boiled off again. In a failure, either the vapor cannot reach the far reaches of the card, the condensed liquid cannot return to the GPU, or not enough liquid (or too much) liquid is in the vapor chamber. Modern cards have a vapor chamber because it's a really fast, cheap way to spread out the heat up to 10 inches away for dissipation by the fans.

The NVidia card-melting issue was obviously NVidia's mistake. You don't build a 600w card with a non-latching power connector. But if your company is staffed by monkeys who always make the new thing exactly the same way they made the old thing, then maybe you do, and maybe your company's name is NVidia.

 

[BX4096]

Actually no, user error is not the manufacturers problem. Car manufacturers still let people drive even though they mess up all the time.
It is up to the user to make sure they are using a device as directed.

Cars don't catch on fire if you fail to insert the ignition key perfectly. You people are getting it wrong: since there hasn't been a precedent for a poorly inserted PC cable causing fire damage, it was 100% unrealistic for NVidia to expect millions of people who weren't paying attention to this for decades NOT to make a user error like that.

I mean, every single PC technician I know, including myself, start their hardware troubleshooting from checking the connection of the cables. Your hard drive not working? Check the cables. No internet? Check the cables. The GPU won't POST? Check the seating and the power cables. And so on and so forth. The fact that poorly inserted cables is such a widespread issue alone makes any expectations of "perfect" user behavior by Nvidia unreasonable and negligent. And that comes from a lifelong Nvidia user.

 

[InvalidError]

Re: implies that the card is energy-efficient because even when the vapor chamber is in a failed state the heat peaks at 110C?

This is an incorrect assumption.

What part of what I wrote said anything about efficiency? IIRC, I haven't mentioned efficiency even once in any of my RX7900-related posts anywhere on this forum, so the "incorrect assumption" is whatever you imagined the efficiency-related part from. I was only commenting on how the temperature maxes out at 110C when the vapor chamber failure state occurs, nothing more, nothing less.

 

[-Fran-]

Seems like AMD will issue a recall and return process for all affected batches of the reference cards:

Hotspot problem of AMD’s Radeon RX 7900 XTX (MBA): Even several batches of the vapor chamber affected, replacement of thousands of cards starts | igor´sLAB

I deliberately waited until I could get congruent and plausible information through various information channels, which can now also shed some light after two to three weeks. While AMD’s support…

www.igorslab.de

Regards.

 

[Why_Me]

 

[Fates_Demise]

Cars don't catch on fire if you fail to insert the ignition key perfectly. You people are getting it wrong: since there hasn't been a precedent for a poorly inserted PC cable causing fire damage, it was 100% unrealistic for NVidia to expect millions of people who weren't paying attention to this for decades NOT to make a user error like that.

I mean, every single PC technician I know, including myself, start their hardware troubleshooting from checking the connection of the cables. Your hard drive not working? Check the cables. No internet? Check the cables. The GPU won't POST? Check the seating and the power cables. And so on and so forth. The fact that poorly inserted cables is such a widespread issue alone makes any expectations of "perfect" user behavior by Nvidia unreasonable and negligent. And that comes from a lifelong Nvidia user.

1) people die every day from improper handling of thier vehicle. 2) it's not widespread, out of millions of 4090s sold they had like 50 that had a issue.

It's always up to the user of any device to read the instructions. If you can't follow them, then pay somebody to do it for you.

 

[DRagor]

out of millions of 4090s sold

Last data published was 120k sold units. Considering 4090 is mostly out of stock all the time doubt the numbers went up significantly since then. Certainly not even a single million sold.

 

[InvalidError]

Last data published was 120k sold units. Considering 4090 is mostly out of stock all the time doubt the numbers went up significantly since then. Certainly not even a single million sold.

With Nvidia seemingly throttling 4000-series GPUs' availability to protect its artificially inflated prices, there must be a still quite significant inventory of 3000-series out there. Heck, Nvidia pushed the 3000-series for mainstream at CES. Wouldn't have said that if they were in a good position to transition everything to the 4000-series.

Also doesn't make much sense to cut back on wafer orders while seemingly unable to keep stock of most new GPUs on store shelves... unless you are still sitting on an alarming inventory of previous-gen stuff, which almost everybody agrees must be the prime reason for the insane pricing on the 4000-series. Can't allow too many people to get an RTX4080-4090 if you want to get more than pocket change for the remaining inventory of 3070-3090s.

 

[BX4096]

...people die every day from improper handling of thier vehicle...

Are you seriously comparing high velocity car accidents with some run-of-the-mill cable connections in an unmovable object that has had zero history of poorly plugged cables causing fire until now? Why not go to a bomb analogy right away?

"People die from mishandling landmines and guns all the time..."

 

[TJ Hooker]

These designs should always fail safe. A fuse should always blow before a connector or wire melts. Preferable (but much harder to design) would be that no current flows until the connector is fully seated.
All bets would be off if there were major accidental damage, like somebody spilling a soda on the gpu... But there's no excuse for unsafe design when we're talking about such a minor and easily missed user error.

A fuse protects against overcurrent, which isn't what's happening with the melting 12VHPWR connectors AFAIK.

Edit: I guess thermal fuses are a thing. But I don't know much about them, or if they'd make sense for this type of scenario.

 

[TJ Hooker]

Heat pipes are different than vapor chambers.
All heat pipes have a wick due to the copper sintering acts as a wick. Although it works in any position through natural hot cold convection, it does work better when the hot plate is at the lowest point. But the natural convection sources do overcome gravity.

Vapor chambers also have sintered copper wicks. I was under the impression that the mechanics of heat pipes and vapor chambers are pretty much the same, except that the former transports heat along a line while the latter does so across a plane.

 

[TJ Hooker]

The NVidia card-melting issue was obviously NVidia's mistake. You don't build a 600w card with a non-latching power connector. But if your company is staffed by monkeys who always make the new thing exactly the same way they made the old thing, then maybe you do, and maybe your company's name is NVidia.

12VHPWR connectors do have a latch, it just doesn't help if users aren't plugging it in far enough for the latch to engage...

 

[TJ Hooker]

Are you seriously comparing high velocity car accidents with some run-of-the-mill cable connections in an unmovable object that has had zero history of poorly plugged cables causing fire until now? Why not go to a bomb analogy right away?

"People die from mishandling landmines and guns all the time..."

You keep referring to fire, there is zero evidence of 12VHPWR connectors catching fire. They just melted. And it is trivial to find reports of other PC power connectors melting, far from "zero history",

 

[digitalgriffin]

Vapor chambers also have sintered copper wicks. I was under the impression that the mechanics of heat pipes and vapor chambers are pretty much the same, except that the former transports heat along a line while the latter does so across a plane.

100% correct. The previous poster implied that some heat pipes do not. He was also talking about heat pipes which is different than vapor chambers.

 

[digitalgriffin]

So what I imagine happening is this:

There is too little fluid in the vapor chamber. When tilted on it's side, none of it reaches the hot plate. The hot plate overheats past a critical threshold, and the ability of the liquid reaching the hot plate is null.

Example Illustrated:

Heat up a dry frying pan. Spritz some water from your fingertips onto the frying pan. The liquid water NEVER touches the frying pan because there is a pocket of gaseous steam under the liquid water keeping it from touching the hot surface! Thermal conductivity is poor. Thus it takes a while for the liquid to disappear. This is not a normal mode of operation for a vapor chamber.

If you had the water on the plate as it is heated in the first place, it would have better thermal conductivity across the surface, thus staying below the critical point.

 

[InvalidError]

12VHPWR connectors do have a latch, it just doesn't help if users aren't plugging it in far enough for the latch to engage...

The only major "flaw" here is simply the connector not being designed to make the latch's proper engagement (or lack thereof) more obvious, which would still be no guarantee people wouldn't still fail to fully insert it. A few reviewers pointed out the unexpected amount of force required to fully insert the HPWR connector in their RTX4070s too.

 

[BFG-9000]

Heat pipes are different than vapor chambers.
All heat pipes have a wick due to the copper sintering acts as a wick. Although it works in any position through natural hot cold convection, it does work better when the hot plate is at the lowest point.

Perhaps you did not notice that der8auer actually cut open one of the offending vapor chambers and was surprised to only find sintering under the die pad. The rest of the vapor chamber only had two layers of copper mesh--like a lower-cost heatpipe uses. Yes, a vapor chamber differs from a heatpipe in that it is just a wider heatpipe which can be made thinner using internal reinforcements. Neither needs a wick if you can put the heat source at the lowest point, and both will need a wick to oppose gravity by capillary action to return condensed fluid back to a higher heat source, because neither can work efficiently unless the working fluid boils. So they are functionally the same regardless of what your freshman-level course may have taught you (and if you are relying on a solid phase of your working fluid then you have chosen the wrong fluid). It was just curious to see such a low-quality wick in something that normally is expected to work upside-down.

 

[InvalidError]

Perhaps you did not notice that der8auer actually cut open one of the offending vapor chambers and was surprised to only find sintering under the die pad.

Probably not surprising: you want the wick to be as close to saturated as possible to get the most flow from the most saturated area to the hot spot which should be on the verge of drying out to draw the fluid in.

The pillars without wick material are primarily structural so the vapor chamber doesn't get crushed/bowed-in by mounting pressure and the inside-outside pressure differential over time.

 

[digitalgriffin]

Perhaps you did not notice that der8auer actually cut open one of the offending vapor chambers and was surprised to only find sintering under the die pad. The rest of the vapor chamber only had two layers of copper mesh--like a lower-cost heatpipe uses. Yes, a vapor chamber differs from a heatpipe in that it is just a wider heatpipe which can be made thinner using internal reinforcements. Neither needs a wick if you can put the heat source at the lowest point, and both will need a wick to oppose gravity by capillary action to return condensed fluid back to a higher heat source, because neither can work efficiently unless the working fluid boils. So they are functionally the same regardless of what your freshman-level course may have taught you (and if you are relying on a solid phase of your working fluid then you have chosen the wrong fluid). It was just curious to see such a low-quality wick in something that normally is expected to work upside-down.

If the sintering only exist around the hot plate then that by definition is not a vapor chamber. The cold plate side needs sintering back to the hot plate to allow capillary action to work.

 

[Coffee Fueled Curmudgeon]

2) it's not widespread, out of millions of 4090s sold they had like 50 that had a issue.

Not to mention half a million people claimed to have a melting 4090 on T'Internet while fanboy band wagoning was afoot. Curiously Nvidia have only heard from fifty of them.