Can 12VHPWR issues be fixed?

[oxrufiioxo]

Been running my 4090 for about a year now no fires yet and it comes from the uber reliable gigabyte that usually can't get anything power supply related right....

When I got the card I actually tried to get the cable to burn setting it to 600w running it at 100% for days and not fully seating it I could not get it to go up in smoke sadly..... (cablemod cable)

After the return period ended I inspected the cable, fully seated it, and honestly haven't checked it since. Maybe I should.....

Personally I don't really care one way or the other if it had 4x8 pin or the one 12VHPWR connector. Maybe down the line there will br a class action like the 970 and I will get a cool 30 bucks until then it is what it is.

 

[A Computer Guy]

I don't understand the "low safety tolerance" argument, seems like high-quality cables are more than capable of meeting the spec's physical and electrical requirements. The latch design just needs to be improved somewhat and this will not be any behind the traditional 8-pin IMO.

To my understanding (per the video) der8auer attempts to clarify that the 8-pin 150w specification is actually very outdated and this is why that cable has a much higher safety tolerance. It seems plausible to me the technology and manufacturing around the 8-pin improved over the years thus creating this higher safety tolerance and allowing the 8-pin to actually do more than it was originally designed for. As an aside to me it seems this is why (to a certain point) you can get away with using those daisy chained 6+2 cables with 2 connectors instead of actually using 2 x 8-pin cables for GPU's that require two 8-pin connections.

Charts from the video...

On the other hand the 12HPWR being new has to live up to a higher manufacturing and installation expectation that is comparatively less forgiving of mistakes when used to it's max power specification. The 12V-2x6 design does a lot to mitigate the problems of the original design but the 12HPWR connector itself is still subject to handle the 600w max specification which I think der8auer sees as a potential issue and perhaps it's too much risk to put into a single connection for a high power drawing card like the 4090.

If 12HPWR was underrated for only 450w (even though it could do more) this would be a total non-issue as high powered cards would have 2 connectors installed in order to meet spec and you get that cushion of tolerance. Likewise if your not pulling max power then it's likely not an issue either except perhaps for the smaller latching mechanism and cable bending guidelines.

Just my late late late night thoughts.

 

[ty_ger]

Arguably these are flaws that already exist in the 6- and 8-pin ATX connectors that look like the EPS plug, so I don't really see the big deal. Sure they've got a "lock" on top, then again, so does my cable...

Again, the video covers this. I suggest you watch it.

The major point is that 6 and 8-pin connectors can have the same problems, but since they have a factor of safety of 180%, for example, instead of 110%, they are more tolerant to defects and errors. That's really the root of it. The major point to the whole topic. Idiot-proofing stuff doesn't create a factor of safety that isn't already part of the overall design.

 

[R-T-B]

All reviewes suddenly forgot about the 30 mating cycle this connector has.

So does 8-pin.

 

[trparky]

oes 0.05% failure rate enough to warrant a redesign? Probably not.

Northridge Fix would beg to differ. He claims that he gets pallets of video cards to fix on the daily. I don't know about you, but that's an epic failure of the connector.

 

[iameatingjam]

Northridge Fix would beg to differ. He claims that he gets pallets of video cards to fix on the daily. I don't know about you, but that's an epic failure of the connector.

Yeah but without a sample size... I mean... it doesn't really tell us a whole lot. How many cards did it take to create that many failures? Thats a pretty big missing variable. Though... I do lean towards der8auer here, I think he has a valid point.

----

Anyway on topic, knowing nvidia is nvidia and doesn't like to admit mistakes... I have a hard time imaging them abandoning the connector altogether. However, including a second connector on the next card that can deliver up to 600w, thats something I can see actually materializing. So I voted 'Yes - that would be fine' but I'd be fine with 8 pins as well.

 

[nguyen]

Northridge Fix would beg to differ. He claims that he gets pallets of video cards to fix on the daily. I don't know about you, but that's an epic failure of the connector.

He got hundred of dead 6900XT and claim there were core cracking issue, what's your point?

 

[chrcoluk]

Assuming DB has researched this right very tempted to vote the first option, but more 8 pins is also an option and we know that is a tried and tested method so I voted for it as the sound option.

 

[evernessince]

I don't understand the "low safety tolerance" argument, seems like high-quality cables are more than capable of meeting the spec's physical and electrical requirements. The latch design just needs to be improved somewhat and this will not be any behind the traditional 8-pin IMO.

The problem is that the spec barely calls for more than high end cards already consume (as pointed to in the video). There's no wiggle room for anything including the cable being bent, an overclock, environmental variables, age, wear, or manufacturing defects. The current form of the 12V2X6 is nowhere near as good as the 8-pin. The 8-pin is designed so that all of the above could be unfavorable and it'd still operate just fine. The polar opposite of the 12VX6, which is teetering on the edge on any 4090 unless you UV.

Yeah but without a sample size... I mean... it doesn't really tell us a whole lot. How many cards did it take to create that many failures? Thats a pretty big missing variable. Though... I do lean towards der8auer here, I think he has a valid point.

----

Anyway on topic, knowing nvidia is nvidia and doesn't like to admit mistakes... I have a hard time imaging them abandoning the connector altogether. However, including a second connector on the next card that can deliver up to 600w, thats something I can see actually materializing. So I voted 'Yes - that would be fine' but I'd be fine with 8 pins as well.

A single pallet of broken 12VHPWR connectors is vastly more cards than 8-pin has in over a decade. In fact I don't think I've ever seen those video card repair channels have a melted 8-pin. Just goes to show you the difference in frequency. Even if we assume melted 12HPWR connectors is rare, it is magnitudes worse than 8-pin regardless. Just another thing that could potentially nuke your GPU when it shouldn't given we had the connector completely solved with the 8-pin.

He got hundred of dead 6900XT and claim there were core cracking issue, what's your point?

You are definitely taking that out of context like some news outlets did, go watch his video.

 

[nguyen]

A single pallet of broken 12VHPWR connectors is vastly more cards than 8-pin has in over a decade. In fact I don't think I've ever seen those video card repair channels have a melted 8-pin. Just goes to show you the difference in frequency. Even if we assume melted 12HPWR connectors is rare, it is magnitudes worse than 8-pin regardless. Just another thing that could potentially nuke your GPU when it shouldn't given we had the connector completely solved with the 8-pin.

You are definitely taking that out of context like some news outlets did, go watch his video.

Just so you know that 4090 can be pushed to 900W with XOC Bios, the fact that the repair channel get those 4090 means their users did something that break the warranty.

Anyways it's just hysteria at this point, even if people do something stupid like pushing 900W on the 4090, all it does is melting the connector, which can be easily replaced

 

[Bobaganoosh]

The thing that gets me is there are other options out there. Why did they need to invent a combo that has two pin sizes, complicated molding, a subpar pin/socket combination, and horrible latch that is not a good fit for the use-case, and no other function in the world?

 

[LabRat 891]

To my understanding (per the video) der8auer attempts to clarify that the 8-pin 150w specification is actually very outdated and this is why that cable has a much higher safety tolerance. It seems plausible to me the technology and manufacturing around the 8-pin improved over the years thus creating this higher safety tolerance and allowing the 8-pin to actually do more than it was originally designed for. As an aside to me it seems this is why (to a certain point) you can get away with using those daisy chained 6+2 cables with 2 connectors instead of actually using 2 x 8-pin cables for GPU's that require two 8-pin connections.

EPS 8-pin was on-path to have addressed this.
(Versus PCI-E 8-Pin) it's spec'd for 1 less gnd and 1 more hot, @ 300+W ea. with extraordinarily-widespread use in the industry, already.
Note: many "MI"/"AI" PCI-E cards for servers/workstations, used EPS, not dual 8PCIE

Elephant in the room on EPS though:
Ug the Barbarian-Technician/Gamer can mix up EPS and PCIE 8-pin, to catastrophic consequences.

 

[iameatingjam]

A single pallet of broken 12VHPWR connectors is vastly more cards than 8-pin has in over a decade. In fact I don't think I've ever seen those video card repair channels have a melted 8-pin. Just goes to show you the difference in frequency. Even if we assume melted 12HPWR connectors is rare, it is magnitudes worse than 8-pin regardless. Just another thing that could potentially nuke your GPU when it shouldn't given we had the connector completely solved with the 8-pin.

Well just to be clear, I do think there are flaws with the connector and do agree with der8auer as much as I can ( I have somewhat limited knowledge on the subject). I just don't think that one source 'Northridge Fix', is a good place to draw that conclusion from, because its missing a vital variable.

Melted 8 pin connectors happen too (I'm not saying it happens as often, but it does happen). Its very likely there are pallets of them somewhere that just didn't get the media attention 12VHPWR did. If you were to see a pallet full of them, would that convince you the 8 pin is bad? I wouldn't hope so. So the opposite shouldn't be the case either.

 

[trparky]

Yeah but without a sample size... I mean... it doesn't really tell us a whole lot.

When I think of a pallet, I think of a wooden skid that's 48x40 inches that's about three or four feet tall. Given those dimensions, I'd have to say there's about a good two hundred or so video cards in some kind of package on that pallet.

 

[iameatingjam]

When I think of a pallet, I think of a wooden skid that's 48x40 inches that's about three or four feet tall. Given those dimensions, I'd have to say there's about a good two hundred or so video cards in some kind of package on that pallet.

Okay and how many cards did it take to produce that pallet? 1000?10,000?100,000? We don't know. So while I agree with you that there's issues with the connector.... I just don't think thats a good way to draw that conclusion. But we can blame nvidia for that. They would most likely want to bury the numbers if they were as bad as it seems like they might be.

 

[trparky]

Okay and how many cards did it take to produce that pallet?

Given the dimensions that I stated, I'd say about two hundred per pallet. And if we're going to go by Northridge Fix's statement in which he gets a pallet a day, that's two hundred or so cards to be fixed on a daily basis. If that doesn't bring a "Yikes!" to your lips, I don't know what will.

 

[iameatingjam]

Given the dimensions that I stated, I'd say about two hundred per pallet. And if we're going to go by Northridge Fix's statement in which he gets a pallet a day, that's two hundred or so cards to be fixed on a daily basis. If that doesn't bring a "Yikes!" to your lips, I don't know what will.

It took 200 cards to produce 200 damaged cards...? So there's 'a 100% failure rate?

 

[trparky]

It took 200 cards to produce 200 damaged cards...? So there's 'a 100% failure rate?

I'm going to assume yes since why would one send a card to him if the card works. Right?

Kind of stupid to send a working card to the guy.

 

[iameatingjam]

I'm going to assume yes since why would one send a card to him if the card works. Right?

Kind of stupid to send a working card to the guy.

When I said 'Okay and how many cards did it take to produce that pallet?' I didn't mean... how many damaged cards are in the pallet since you had already clearly stated that.... I meant, what kind of sample size does it take to produce that many damaged cards, to give an indication of failure rate. I don't even understand why we are having this argument really when we are agreeing on the conclusion, we just came to the conclusion in different ways.

 

[evernessince]

Well just to be clear, I do think there are flaws with the connector and do agree with der8auer as much as I can ( I have somewhat limited knowledge on the subject). I just don't think that one source 'Northridge Fix', is a good place to draw that conclusion from, because its missing a vital variable.

Northbridge fix or Northwestrepair both get a ton of melted 12HPWR in. It's not just one source.

Melted 8 pin connectors happen too (I'm not saying it happens as often, but it does happen). Its very likely there are pallets of them somewhere that just didn't get the media attention 12VHPWR did. If you were to see a pallet full of them, would that convince you the 8 pin is bad? I wouldn't hope so. So the opposite shouldn't be the case either.

Almost certainly a pallet of melted 8-pins would get reported. You are making the assumption that there's some kind of double standard or bias where 8-pin issues are not being reported but logically speaking the conclusion with the least assumptions is more likely correct. In otherwords, it's unlikely that such a specific bias exists, especially in light that there's no argument for such bias to exist in the first place. You'd first need to provide a factual basis for the argument that 8-pin issues are under-reported, otherwise it's just supposition. You also have to consider that the 8-pin has vastly more market saturation, meaning there should be more incident reports involving the connector. Unless you are implying that 8-pin incidents are extremely under-reported (which again you'd have to explain), the logical conclusion is more that the 8-pin is just vastly more reliable.

The PCIe 8 pin has been around now for 17 years, the fact that we don't have a single report of such a pallet certainly qualifies as evidence of absence (not to be confused with absence of evidence). If there was a pallet of melted 8 pins I wouldn't hold in at the same weight as a pallet of melted 12HPWR connectors because the period that each has existed and the frequency of reported issues is vastly different. 12VHPWR has had vastly more reported incidents in a single year than PCIe 8 pin has in it's entire lifetime (17 years now). The 8-pin has a proven track record from years in service, it's demonstrated it's reliable under many different conditions and wear levels. The 12VHPWR has no such track record and hasn't even demonstrated reliability in the short term, let alone after years of wear. Based on the assumption of unreported pallet of melted 8-pins you are further trying to imply that the two should be treated the same when logically that makes no sense given the vastly different track record of the two as stated above. What I gathered from your comment is that you are stacking two logically shaky arguments. 1, that 8-pin issues are underreported and 2, a theoretical pallet of metled 8 pin connectors should be treated the same as an actual pallet of melted 12PWR connectors despite the vast differences as explained above.

 

[taka]

There are some other shops from EU like https://www.youtube.com/@KrisFixGermany/videos reporting this.

 

[iameatingjam]

Northbridge fix or Northwestrepair both get a ton of melted 12HPWR in. It's not just one source.



Almost certainly a pallet of melted 8-pins would get reported. You are making the assumption that there's some kind of double standard or bias where 8-pin issues are not being reported but logically speaking the conclusion with the least assumptions is more likely correct. In otherwords, it's unlikely that such a specific bias exists, especially in light that there's no argument for such bias to exist in the first place. You'd first need to provide a factual basis for the argument that 8-pin issues are under-reported, otherwise it's just supposition. You also have to consider that the 8-pin has vastly more market saturation, meaning there should be more incident reports involving the connector. Unless you are implying that 8-pin incidents are extremely under-reported (which again you'd have to explain), the logical conclusion is more that the 8-pin is just vastly more reliable.

The PCIe 8 pin has been around now for 17 years, the fact that we don't have a single report of such a pallet certainly qualifies as evidence of absence (not to be confused with absence of evidence). If there was a pallet of melted 8 pins I wouldn't hold in at the same weight as a pallet of melted 12HPWR connectors because the period that each has existed and the frequency of reported issues is vastly different. 12VHPWR has had vastly more reported incidents in a single year than PCIe 8 pin has in it's entire lifetime (17 years now). The 8-pin has a proven track record from years in service, it's demonstrated it's reliable under many different conditions and wear levels. The 12VHPWR has no such track record and hasn't even demonstrated reliability in the short term, let alone after years of wear. Based on the assumption of unreported pallet of melted 8-pins you are further trying to imply that the two should be treated the same when logically that makes no sense given the vastly different track record of the two as stated above. What I gathered from your comment is that you are stacking two logically shaky arguments. 1, that 8-pin issues are underreported and 2, a theoretical pallet of metled 8 pin connectors should be treated the same as an actual pallet of melted 12PWR connectors despite the vast differences as explained above.

We came to the same destination... is it really that important to you, that we got here via the same route?

 

[Panther_Seraphin]

16-pin 12VHPWR connector - Wikipedia

en.wikipedia.org

Just have a look at that. Old 8 pin basically has a safety factor of 2. 12VHPWR is barely 1.1 same as 12V2X6

So once you get heat etc involved good bye safety margin, then add in less than 100% contact and now your asking for trouble. The 12VHPWR was purely to make it possible for nVidia to minimise board sizes........yet have coolers nearly 50% larger than 2 generations ago.

Form Over Function in a safety critical point.

 

[chrcoluk]

Wonder how much Nvidia paid PCI-SIG to prevent the official rating from getting dropped to something like 300w.

 

[Edwired]

There's nothing but trouble since NVIDIA changed the design of the connector and yet again alot of fingers pointing at people saying they fully connected the plug and it melted regardless if it had the cable mod angled connector or not since they are recalling the adaptors back due to a defect found in 1.0 and 1.1 variation. The 4080 and 4090 are prone to this problem nothing on 4070 same with 30 series. If anyone was clever they would have used three pin kettle connector like the one on the back of the pc power supply this connector can withstand alot of abuse before melting unlike the flimty 12VHPWR and it broke by breathing on it.
What annoys me the most is where they located the connector on top corner of the pcb not taking into account that it going to bind against side door of the pc case at extreme radius bend running the risk of breaking the connector in the first place. If they moved the connector to the rear centre of the pcb opposite to the display ports it would have been more ideal for routing the cable and everyone would be happy