RTX 5090 FE Molten 12VHPWR

Captain Newmackwa said:
Auto shutdown can be configured either on the device itself or by software. By the way, all settings are saved on the device so the software doesn't have to be running for it to do its thing. The auto-shutdown configuration options are comprehensive but I stick with the defaults so it would be less likely for Thermal Grizzly to deny warranty claims for melted connectors while their product is being used. Actually, that's why I chose the WireView Pro II, they've been fairly explicit and transparent about their warranty on melted connectors and how that has factored in the pricing of their product. That, and having multiple thermal sensors on the GPU connector side.

Here's the software config menu:

View attachment 832957

Actually, I only got around to installing the F_Panel header Y-connector a few days ago after several months of using my WireView Pro II without it because I was too lazy to look for an F_Panel extension break-out cable since the connector that comes with the case is the unified type. The hardware availability and pricing situation right now convinced me not to take any further risks.

I tested the auto shutdown by temporarily setting the temp limit to 45C and then running a GPU load to artificially trigger it and auto-shutdown works great. The 10 second default shutdown wait time is okay with me so if ever something is wrong, the damage can be minimized. There are two shut down modes, soft shutdown and hard shutdown with the former doing a Windows shut down to minimize data loss/corruption.

I also recently stumbled upon a third party MSI afterburner plugin that allows me to monitor the telemetry from the Wireview in the in-game OSD. Very useful when testing DLSS5 which tends to push the graphics card wattage to the max.

View attachment 832958
Click to expand...
Good to see it has soft power.

If the WireView is good enough, I would use a tight tolerance on Current Balance as a leading indicator. The precise current balance will vary greatly from system to system. However, on any individual system, it should not, as the wire lengths are not going to change, neither should the individual contact resistances in the connectors. A change in current balance is an indication a failure is going to happen. Your current balance above has a low of 7A and a high of 8.1. I'd start with a balance limit of 1.2A and 28%. Adjust up to get rid of nuisance failures. It all depends on precise the WireView can be -- I would also leave the other alarms in place.

I also would not trust windows to shut the system down.

Where did you find the Afterburner plugin?
 
littlegreenmenfromhome said:
So you keep posting like you got a point.
Care to tell me what percentage you are at now?
Click to expand...
gigabyte 5080 gaming oc
16 pin 12vhpwr melted

1792960519567.png

https://x.com/unikoshardware/status/2114446812065189918
 
erek said:
yea, almost every day there's multiple new reports of a failed connector now
Click to expand...
So stlil no answer and still under 0.5%?
You know the math right?

If 1 report comes in per day, it will take 2 years to rach 0.5% failure rate. (750 failed card out of 150.000 unit (low estimate).

You hunt the internet like a madman, yet your results are more tham meager.

Your bias is blinding your for the numbers.

You are actuly proving yourself wrong, the irony is hillarious.

I guess math is hard.
 
littlegreenmenfromhome said:
So stlil no answer and still under 0.5%?
You know the math right?

If 1 report comes in per day, it will take 2 years to rach 0.5% failure rate. (750 failed card out of 150.000 unit (low estimate).

You hunt the internet like a madman, yet your results are more tham meager.

Your bias is blinding your for the numbers.

You are actuly proving yourself wrong, the irony is hillarious.

I guess math is hard.
Click to expand...

nvidia 5090 fe
16 pin 12vhpwr melted

1792976852237.png

https://x.com/unikoshardware/status/2114396385395224643
 
littlegreenmenfromhome said:
So stlil no answer and still under 0.5%?
You know the math right?

If 1 report comes in per day, it will take 2 years to rach 0.5% failure rate. (750 failed card out of 150.000 unit (low estimate).

You hunt the internet like a madman, yet your results are more tham meager.

Your bias is blinding your for the numbers.

You are actuly proving yourself wrong, the irony is hillarious.

I guess math is hard.
Click to expand...

"Second MSI RTX 5090 Seemed Fine On Arrival Until Safeguard+ Helped User Identify Uneven Load On Pins Before The Connector Melted​


MSI's new Safeguard+ equipped power supply helped this user identify the problem his new RTX 5090 had.

User Gets His RTX 5090 RMA'd, and Didn't Had Any Errors, But Swapping the PSU Helped Him Knowing the Pin Not Carrying Any Current​

The typical problem associated with the 12V-2x6 connector is that it easily melts due to uneven load. Many vendors have tried to mitigate the problem since it has caused the deaths of probably thousands of RTX 50-series GPUs, especially the RTX 5090, which is rated at a much higher power. From cables with built-in sensors to monitoring devices, we have a dozen such measures by different vendors to curb the problem, but not every solution is flawless in detecting the problem before the hazard occurs."

Source: https://wccftech.com/second-msi-rtx...l-safeguard-helped-user-identify-uneven-load/
 
erek said:

"Second MSI RTX 5090 Seemed Fine On Arrival Until Safeguard+ Helped User Identify Uneven Load On Pins Before The Connector Melted​


MSI's new Safeguard+ equipped power supply helped this user identify the problem his new RTX 5090 had.

User Gets His RTX 5090 RMA'd, and Didn't Had Any Errors, But Swapping the PSU Helped Him Knowing the Pin Not Carrying Any Current​

The typical problem associated with the 12V-2x6 connector is that it easily melts due to uneven load. Many vendors have tried to mitigate the problem since it has caused the deaths of probably thousands of RTX 50-series GPUs, especially the RTX 5090, which is rated at a much higher power. From cables with built-in sensors to monitoring devices, we have a dozen such measures by different vendors to curb the problem, but not every solution is flawless in detecting the problem before the hazard occurs."

Source: https://wccftech.com/second-msi-rtx...l-safeguard-helped-user-identify-uneven-load/
Click to expand...
So keep it up, in 2 years with daily posting you might have a case.
Remember to seperate the SKU's (the 750 numbers only applies to the 150.000 5090's sold, if you add in other SKU's you "case" gets even worse).

Enjoy your confirmation bias, but do not act like you have a valid piont, the math doesn't back you up so far.
 
littlegreenmenfromhome said:
So keep it up, in 2 years with daily posting you might have a case.
Remember to seperate the SKU's (the 750 numbers only applies to the 150.000 5090's sold, if you add in other SKU's you "case" gets even worse).

Enjoy your confirmation bias, but do not act like you have a valid piont, the math doesn't back you up so far.
Click to expand...
Yeah nothing to see here. Even when two of just the handful of youtube channels you watch have the same thing happen to them. :rolleyes:


View: https://www.youtube.com/watch?v=DwhtYL3F1go&t=1s


View: https://www.youtube.com/watch?v=ofcdHlKJ9W0
 
1_rick said:
Any reason you won't answer my question that I've asked a couple of times? How many does it take before you admit it's a problem?
Click to expand...
+0,5%
So wake me up when there are +750 reported cases in this thread.

Math is hard it seems.

EDIT: 750 x 5090's, if you add 5080 and other SKU, those needs to be held against sold units.
But very telling that a thread dedicated to this issue are having trouble with the numbers.

Facts > biased crying.
 
1_rick said:
It's useful to know you think 700 people's wrecked cards are no big deal.
Click to expand...
You fail at percentages, 750 out of 150.000 is nothing,
After 750 cases, the a root cause analysis needs to be done to exclude uers errors etc.
How many 5090 caes in this thread, do you even know or just rage-autopilot from you?

Stay away from numbers, your bias makes you fail hard. being an (angry) AMD user in an NVIDIA thread.
 
arestavo said:
While we will likely never get failure rate data from Nvidia, AMD, PCI-SIG, nor any board partners, a survey taken two (2) years ago found that more people reported 12VHPWR failures in one (1) GPU generation than in seventeen (17) years of 6/8 pin PCIE power connectors. That's nuts.

I wonder what a survey would look like today?
Click to expand...
A survey is biased, you need neutral statistics, a lot of people not having issues will never take the poll, they simply do not care.

I mean look erek , he searches the internet like a sniffer dog every day and stil have found less that 100 cases.
 
mullet said:
Time will be the 5090's biggest enemy.
Click to expand...
we can hope that enthusiasts will solve the issue aftermarket

as in some proven modification or conversion to the older style connectors

i doubt the GPU dies and Memories will be bad

and depending on timing… it’ll be a boon for repair shops to handle conversions and a decently reasonable cost to the consumer who owns the $10,000+ dollar GPU

—-

with how long Vera Rubin has been taking to come online even for the enterprises… seems like we’re a long way out for any NextGen consumer gaming cards at scale

something feels weird with Rubin, maybe i’m wrong but it’s taking a long time

so there’s an opportunity for enthusiasts to solve this 12vhpwr problem themselves
 
1_rick said:
"only 3 apartment complexes were taken over by Tren de Aragua, JD!"

At what point does it become a problem?
Click to expand...

He just ignores everything that contradicts what he wants to believe. He keeps posting one link from GamersNexus saying that if it were an issue, then Youtubers would be talking about it. Meanwhile, GamersNexus HAS talked about it. Many times. All tech Youtubers have done so. Several (like Der8auer) have explained and demonstrated why it was failing by using ammeters showing uneven current distribution. Steve at Hardware Unboxed actually had one fail on him and he made an entire video documenting it. Anyone with any engineering knowledge would argue redlining a connector like that will have predictable consequences, and the fact that the bulk of the failures are 5090s as compared to lower wattage 5080s or 9070XTs backs that up.

The best part is there are several companies creating and marketing solutions to a problem he claims doesn't exist, which they wouldn't devote the resources to if it weren't actually a problem.

Just ignore his posts and move on. It doesn't matter what he wants to believe and it doesn't change reality as much as he wishes it would.
 
The failure that is the 12VHPWR connector compared to previous in one picture (that was posted earlier):
1793031039561.png


Note the condition of the four old-school PCIe 8 pin connectors. They were subjected to the exact same load that destroyed that 12VHPWR connector and yet they look unscathed. That picture makes me think the soldered 4x PCIe 8 adapter will be enough to address the issue and work around Nvidia's faux paw.
 
Not sure why they didn't apply the standard 20% margin when it comes to safe continuous operation, pretty standard in most cases. If the peak the connection can handle is 600W, only use 480W on a continuous basis. The 5090 design, especially the FE, was a marvel of aesthetics and a failure of proper engineering standards, from the non functional PCI-E 5.0 due to daughter card related signal integrity issues (the person I sold it to had the same issue, had to revert to 4.0 for it to even post), to the impossibly large thermal load on a such a small albeit pretty cooler, nothing about it inspired much confidence. My PNY was built substantially better but reports of PNY just denying warranty claims period made me get rid of it, 2x working 5080/70tis >>>> a dead 5090 lol.
 
Last edited:
1_rick said:
Yup.

I could put up with a fact-based argument, but i'm more tired of the attitude.

I went like 9 years before blocking my first person here, and in the last week I've had to add two more. Annoying.
Click to expand...

Fact-based arguments don’t work on him. He’ll just come back with some stat like “it’s 0.5% of cases”, which frankly sounds made up and inconsistent with global R&D efforts to solve this problem. Recall these are the same people who claimed the 12VHPWR issues always broke down to “user error”, only to have the design formally updated to 12v-2x6 specifically to address the difficulty people were having ensuring all pins were connected. But that’s because they’re all idiots, of course.
 
undertaker2k8 said:
Not sure why they didn't apply the standard 20% margin when it comes to safe continuous operation, pretty standard in most cases. If the peak the connection can handle is 600W, only use 500W on a continuous basis. The 5090 design, especially the FE, was a marvel of aesthetics and a failure of proper engineering standards, from the non functional PCI-E 5.0 due to daughter card related signal integrity issues (the person I sold it to had the same issue, had to revert to 4.0 for it to even post), to the impossibly large thermal load on a such a small albeit pretty cooler, nothing about it inspired much confidence. My PNY was built substantially better but reports of PNY just denying warranty claims period made me get rid of it, 2x working 5080/70tis >>>> a dead 5090 lol.
Click to expand...

Sounds like you’re approaching this from an engineering standpoint, asking logical questions like “why are you redlining the connector so close to maximum design tolerances without doing at least the most basic of things like ensuring proper load balancing or creating fusing to prevent catastrophic card failure?”. Such questions are unwelcome here, and you’ll be sent to the Radeon Gulag for your thoughtcrime, sir!
 
undertaker2k8 said:
Not sure why they didn't apply the standard 20% margin when it comes to safe continuous operation, pretty standard in most cases. If the peak the connection can handle is 600W, only use 500W on a continuous basis. The 5090 design, especially the FE, was a marvel of aesthetics and a failure of proper engineering standards, from the non functional PCI-E 5.0 due to daughter card related signal integrity issues (the person I sold it to had the same issue, had to revert to 4.0 for it to even post), to the impossibly large thermal load on a such a small albeit pretty cooler, nothing about it inspired much confidence. My PNY was built substantially better but reports of PNY just denying warranty claims period made me get rid of it, 2x working 5080/70tis >>>> a dead 5090 lol.
Click to expand...
Right as an electrician it pisses me off to know that I have to power limit my card in the hope it lasts a few years because the connector is junk.

My bet is if you put 100 5090s under full load 24/7 for a year or two most of them would burn out at the full 600w and that is unacceptable to me.

To the guy that's going to say that doesn't prove anything well though shit no one cares what you think and most of us have you blocked now
 
tunatime said:
Right as an electrician it pisses me off to know that I have to power limit my card in the hope it lasts a few years because the connector is junk.

My bet is if you put 100 5090s under full load 24/7 for a year or two most of them would burn out at the full 600w and that is unacceptable to me.

To the guy that's going to say that doesn't prove anything well though shit no one cares what you think and most of us have you blocked now
Click to expand...
Seems like the ideal scenario peak load for the connector is around 684W so it can theoretically handle around 547W safely, the default rating on the 5090 at 575W still exceeds this.

The 80% rule holds up well across almost all electrical connections, my garage dryer nema outlet is rated for 30A max, it can safely charge my model 3 at 24A all day long even on 100f days, increase it to 26-27A and it will occasionally trip after a few hours on hot days (tested when I first got the car).

I bet most AI data centers are power limiting the cards, using only the best cables and/or just physically changing them back to 8 pin ones along with active cooling directed at the connectors, they have trained folks specializing in (reverse) engineering these things.
 
tunatime said:
Right as an electrician it pisses me off to know that I have to power limit my card in the hope it lasts a few years because the connector is junk.

My bet is if you put 100 5090s under full load 24/7 for a year or two most of them would burn out at the full 600w and that is unacceptable to me.

To the guy that's going to say that doesn't prove anything well though shit no one cares what you think and most of us have you blocked now
Click to expand...
in the enterprises with 100,000 gpu clusters

they’re having to chase down failures each hour like they used to have to hunt down burned out vacuum tubes in the ENIAC

https://www.theregister.com/systems...percluster-just-waiting-for-a-failure/5304269


View: https://medium.com/the-investors-handbook/gpu-failure-rates-and-the-3-clocks-that-set-ai-economics-cd0dafb17885

https://www.tomshardware.com/tech-i...ee-hours-for-metas-16384-gpu-training-cluster

https://www.linkedin.com/posts/denn...eering-gpus-activity-7495299278081363968-KJ6a

https://note.com/chemimotty/n/ne9cf6abcb846?hl=en
 
 
TXE36 said:
The failure that is the 12VHPWR connector compared to previous in one picture (that was posted earlier):
View attachment 833188

Note the condition of the four old-school PCIe 8 pin connectors. They were subjected to the exact same load that destroyed that 12VHPWR connector and yet they look unscathed. That picture makes me think the soldered 4x PCIe 8 adapter will be enough to address the issue and work around Nvidia's faux paw.
Click to expand...
That powerplug looks bent, like the user did not ensure no pulling load on the cable:
1793055998918.png


If so we are in the "user error" land.
 
serpretetsky said:
Its hard to tell whats going on, but it could also be the cable already melted, and the user had a helluva a time trying to yank it out, causing bent pins/wires in the process.
Click to expand...
So far from enough data to blame the design here, but this is being "sold" as another one in this "sub 100 issues thread."

And the the question arises:
Why pull it apart and simply not ship fused for RMA?
 
Last edited:
You must log in or register to reply here.
Back
Top