RTX 5090 FE Molten 12VHPWR

arestavo said:
The PCIE slot is rated for 75W max power draw for 12V. EVGA got in a bit of trouble a few years back because some of their 30 series GPUs were pulling more, and in some cases a LOT more. I even RMA'd my 3090 FTW3 Ultra because their VBIOS fix didn't keep the power draw under 75W and I didn't want to risk my RAID card / array.

I remember marveling at how little PCIE slot power was drawn when I upgraded to a 4090 FE, something like 8W, compared to the 3090. I still miss EVGA because their US customer support was second to none.
Click to expand...

75W slightly overstates it actually. It's 65W of 12V and 10W of 3.3V. The latter is what legacy parallel PCIe used, providing it eased the transition by allowing makers to take existing designs slap a one dollar bridge chip into a bit of dead space and port their old design forward.

AFAIK very little modern hardware uses the 3.3V anymore. It's still there for back compatibility though and ends up driving the amount of 3.3V needed to be delivered by PSUs (since very little on the mobo uses it anymore either).

If PCIe7 drops 3.3V support, and ATX 12VO remains MIA on enthusiast boards maybe 2040 PSUs will be able to drop it like they did with -5V 15 or 20 years ago (and long after the last ISA cards that needed it were gone).
 
DanNeely said:
75W slightly overstates it actually. It's 65W of 12V and 10W of 3.3V. The latter is what legacy parallel PCIe used, providing it eased the transition by allowing makers to take existing designs slap a one dollar bridge chip into a bit of dead space and port their old design forward.

AFAIK very little modern hardware uses the 3.3V anymore. It's still there for back compatibility though and ends up driving the amount of 3.3V needed to be delivered by PSUs (since very little on the mobo uses it anymore either).

If PCIe7 drops 3.3V support, and ATX 12VO remains MIA on enthusiast boards maybe 2040 PSUs will be able to drop it like they did with -5V 15 or 20 years ago (and long after the last ISA cards that needed it were gone).
Click to expand...
Oh yeah! I'd forgotten about 3.3V because GPU-Z didn't differentiate.
 
DooKey said:
Damn, I'm missing out on the fun since my release 4090 is doing just fine and my near release 5090 is doing fine.
Click to expand...
littlegreenmenfromhome said:
My poor release 4090 from 2022 is also feeling left out.
Click to expand...
ZeroBarrier said:
My near release 5080 FE is also in the "feeling left out" club.
Click to expand...

The evidence and engineering presented in this thread and its references indicates the above have simply been lucky. Lucky in that the connector resistance for each pin is close enough that they actually share the load. The luck may also run out with age or if one starts running higher demand applications. Properly engineered things don't rely on luck, they rely on science, engineering theory and testing. The engineering theory applied to this connector and the power feed scheme for the board indicates it is poor. Testing by users is also confirming an issue.

arestavo said:
My 5090 was going to be another statistic, but I caught it with the WireView Pro II and replaced the cable. Crazy, that.
Click to expand...
Tsumi said:
Cool. So out of 10? 4090/5090 owners in this thread, 1 had a blow up, and 1 managed to catch a blow up before it occurred. A 20% failure rate seems unacceptably high to me, but what do I know.
Click to expand...
This connector will not age well. With time, more and more are going to fail. That's practically guaranteed at this point. <-- I bet this statement will age better than "mine's fine".

Spend your money, take your chance.
 
HockeyJon said:
My understanding, from what I’ve read, is that board partners have to use it. They are not allowed to use 8-pin connectors on a custom PCB. Why they’re not allowed to at least use two on a 5090 when they’re clearly redlining it is beyond me. You have 85W coming from the PCI slot and a cable rated at 600W feeding a GPU that, at times, will exceed 600W. There isn’t a good reason to limit the design to a single connector that’s already at its design limit.
Click to expand...
This has me suspecting that stuff like this, among other things, is part of the reason EVGA gave up on being a board partner with Nvidia.

The AI valuation has me suspicious that Nvidia's top engineering talent has been directed to those efforts with gaming getting the scraps. If you are a top engineer in a company, you want to be where the money is.
 
TXE36 said:
The evidence and engineering presented in this thread and its references indicates the above have simply been lucky. Lucky in that the connector resistance for each pin is close enough that they actually share the load. The luck may also run out with age or if one starts running higher demand applications. Properly engineered things don't rely on luck, they rely on science, engineering theory and testing. The engineering theory applied to this connector and the power feed scheme for the board indicates it is poor. Testing by users is also confirming an issue.



This connector will not age well. With time, more and more are going to fail. That's practically guaranteed at this point. <-- I bet this statement will age better than "mine's fine".

Spend your money, take your chance.
Click to expand...
i'm thinking the luck is gonna run out eventually, honestly
 
TXE36 said:
The evidence and engineering presented in this thread and its references indicates the above have simply been lucky. Lucky in that the connector resistance for each pin is close enough that they actually share the load. The luck may also run out with age or if one starts running higher demand applications. Properly engineered things don't rely on luck, they rely on science, engineering theory and testing. The engineering theory applied to this connector and the power feed scheme for the board indicates it is poor. Testing by users is also confirming an issue.



This connector will not age well. With time, more and more are going to fail. That's practically guaranteed at this point. <-- I bet this statement will age better than "mine's fine".

Spend your money, take your chance.
Click to expand...
Bollock, if you claim the majority of users have this issue we have left reality...with +900.000 RTX 4090/5090's sold , the amount of coonectors burned are very low.
Again , Nexus Steve is quiet.
 
littlegreenmenfromhome said:
Bollock, if you claim the majority of users have this issue we have left reality...with +900.000 RTX 4090/5090's sold , the amount of coonectors burned are very low.
Again , Nexus Steve is quiet.
Click to expand...
Who said majority? Who cares about Steve?

I'm saying those not having problems are depending upon luck due to inadequate engineering -- frankly, Nvidia is too. Time will tell.
 
  • Like
Reactions: kac77
like this
littlegreenmenfromhome said:
Bollock, if you claim the majority of users have this issue we have left reality...with +900.000 RTX 4090/5090's sold , the amount of coonectors burned are very low.
Again , Nexus Steve is quiet.
Click to expand...
One stop lumping 4090s with 5090 not at all the same power load. 4090s are falling because the connections have loosened up an aged a bit 5090 are already falling so by the time 5090s are the age of 4090s they will be falling more then 4090s now.

Next according to stream hardware statistics .42% of people have a 5090 if 900000 5090s when sold obviously most of them didn't go to gamers or that percentage would be higher.
 
tunatime said:
One stop lumping 4090s with 5090 not at all the same power load. 4090s are falling because the connections have loosened up an aged a bit 5090 are already falling so by the time 5090s are the age of 4090s they will be falling more then 4090s now.

Next according to stream hardware statistics .42% of people have a 5090 if 900000 5090s when sold obviously most of them didn't go to gamers or that percentage would be higher.
Click to expand...
Actually... that puts it right about 900k. Steam has roughly 200m active monthly users. So 0.42% of 200m = 840,000.

Now I don't know how accurate Steam's data is, I don't think they check every user, I think they use stratified random sampling and you have to opt in, but the numbers roughly match, particularly considering we are talking like 1-2 digits of precision in all these numbers.
 
TXE36 said:
This has me suspecting that stuff like this, among other things, is part of the reason EVGA gave up on being a board partner with Nvidia.

The AI valuation has me suspicious that Nvidia's top engineering talent has been directed to those efforts with gaming getting the scraps. If you are a top engineer in a company, you want to be where the money is.
Click to expand...

Design flexibility may have been an annoyance for EVGA, but a PM recently came out and said the primary driver was competition from Nvidia’s FE cards and a requirement for board partners to offer at least one card at MSRP, but then selling them a chip package at a price that meant they were running that model at a loss. It wasn’t sustainable.

https://www.techpowerup.com/353192/...unts-nvidia-fe-squeeze-and-leaker-witch-hunts

As for AI, the cards we get have been trickle down from AI research for some time. Tensor cores were brought in to help with AI deep learning, which also enabled Nvidia to start pushing things like DLSS. I don’t necessarily think this is a problem. A GPU is just a compute device at the end of the day. I’m more concerned about crappy vibe-coded drivers.
 
LodeRunner said:
While I think the connector is poorly designed, based on CPU dying at same time I’d be more inclined to blame the PSU.
Click to expand...
Psu probably popped after it started melting. That melt down was going on for a while. If the PSU would have sent a spike or a dip ECT the card would have just died without the connection melting
 
LodeRunner said:
While I think the connector is poorly designed, based on CPU dying at same time I’d be more inclined to blame the PSU.
Click to expand...
Going in the comments, he said he had a Seasonic Prime PX 1000w. Hate to think a Seasonic would take out multiple components.
 
The CPU may be a coincidence, hard to believe as it may be. And maybe it's not even really dead, it's not like we have pictures of the damage or extensive testing or anything (like maybe a clear CMOS and the CPU could boot again?), the melted GPU connector we can clearly see though.

Either way the "murderboards" are still around and some AM5 CPUs are mysteriously dying from those (mostly ASROCK, like he has) though, so there's that.
 
One of the problems with the official statistics on melted connectors is that these days it’s not always a good idea to send the card back for an RMA.
There’s a high chance the RMA will be rejected due to a “non-original adapter” or if the card was bought a few years ago at a low price, just to get a refund and resell it at today’s prices. So, waiting 20–30 days with an uncertain outcome… it’s better to just go to a local repair shop to have the connector replaced.

So, the people who complain about the lack of official statistics will never see them.
 
"An MSI GeForce RTX 5090 Ventus 3X OC owner discovered that the bundled 12V-2x6 power adapter had melted. The graphics card had been in use since March 2025. According to the owner, the cable was fully seated and had never been disconnected. Undervolting caused no visible stability issues.Testing with a Thermal Grizzly WireView Pro II showed that, at 473.42 W of power draw, the sixth power contact carried 0.0 A. Each of the other five lines carried between 5.7 A and 8.6 A. The damaged plastic was discovered only when the adapter was removed for inspection.The connector on the graphics card reached 43.7 °C. Switching to a direct Corsair power cable and changing the power supply port did not resolve the current imbalance. The exact cause has not yet been established, and the graphics card was sent in for warranty service.#MSI #RTX5090 #NVIDIA #GeForce #12V2x6 #WireView #GraphicsCards"

1791755196581.png

https://x.com/GameGPU_com/status/2109215213614452921
 
erek said:
"An MSI GeForce RTX 5090 Ventus 3X OC owner discovered that the bundled 12V-2x6 power adapter had melted. The graphics card had been in use since March 2025. According to the owner, the cable was fully seated and had never been disconnected. Undervolting caused no visible stability issues.Testing with a Thermal Grizzly WireView Pro II showed that, at 473.42 W of power draw, the sixth power contact carried 0.0 A. Each of the other five lines carried between 5.7 A and 8.6 A. The damaged plastic was discovered only when the adapter was removed for inspection.The connector on the graphics card reached 43.7 °C. Switching to a direct Corsair power cable and changing the power supply port did not resolve the current imbalance. The exact cause has not yet been established, and the graphics card was sent in for warranty service.#MSI #RTX5090 #NVIDIA #GeForce #12V2x6 #WireView #GraphicsCards"

View attachment 830678
https://x.com/GameGPU_com/status/2109215213614452921
Click to expand...
Why does this image reek of AI generation?
1208505_1791755196581.png


Oh wait... Because it's a fake image...

Screenshot 2026-10-11 163549.png
 
ZeroBarrier said:
Much like the whole melting connector manufactured drama.
Click to expand...
Almost just as sad as those making excuses and pretending it is not an issue. I only watch a handful of youtube tech channels and it sure is strange how many of those have also had the melting issues.

But do not argue with the usual suspects on here that will defend anything from Nvidia of course. Not going to reply just simply putting put my 2 cents in and moving on.
 
erek said:
what do you mean?

mine melted,

https://videocardz.com/newz/first-user-reports-rtx-5090-with-melted-connector-after-testing-dlss-5
https://www.techpowerup.com/352412/...ed-rtx-5090-connector-power-shoots-past-600-w
https://www.tomshardware.com/pc-com...s-up-to-50-percent-more-power-draw-in-testing
Click to expand...
I can find just as many or more 8 power pins that melted, you need more than anecdotes:
https://www.google.com/search?sca_e...86KeIQtKgLegQIGxAB&biw=2195&bih=1203&dpr=1.75

The LAST large study I saw was this:
https://gamersnexus.net/gpus/12vhpwr-dumpster-fire-investigation-contradicting-specs-corner-cutting

Based on this, there was a 0.04%-0.05% failure rate relative to approximately 125,000 RTX 4090s sold in the first month.
Click to expand...

I remember this first round, you wuld think it was 90% of all RTX 4090 failing be the crying on forum, but reality/data was something else.
This seems like the same thing, until you provide a large study, even a 100 posts would not break the 0.09% mark.
 
  • Like
Reactions: erek
like this
TXE36 said:
This has me suspecting that stuff like this, among other things, is part of the reason EVGA gave up on being a board partner with Nvidia.

The AI valuation has me suspicious that Nvidia's top engineering talent has been directed to those efforts with gaming getting the scraps. If you are a top engineer in a company, you want to be where the money is.
Click to expand...
The "other" things are they got tired of competing with Nvidia themself and the founders cards that were always the cheapest. They also got tired of making $5 profit on each card they sold.
 
jobert said:
Almost just as sad as those making excuses and pretending it is not an issue. I only watch a handful of youtube tech channels and it sure is strange how many of those have also had the melting issues.

But do not argue with the usual suspects on here that will defend anything from Nvidia of course. Not going to reply just simply putting put my 2 cents in and moving on.
Click to expand...
I have 2 5090 in two systems. No problems. Just saying
 
You must log in or register to reply here.
Back
Top