PCIe 12VHPWR and 12V-2x6 power connector issues

[DegustatoR]

power distribution across the 6 12V wires

These wires are connected in the plug. There is no "power distribution" on them because of that, they all should get about the same amount of current. The only way for that to be different is if the connection inside the plug is somehow broken and all the power flows through just one of the pins - which is obviously not an intended behavior.

 

[homerdog]

On the other hand, Andreas Schiller is a legend, and his measurements on the power distribution across the 6 12V wires look solid (and he also has one for the same 5090 Aorus Master I'm using).
I guess it's something of a relief that it does not use the Nvidia reference PCB. I can see at least 2 shunts/power delivery channels on teardowns.

If the current is properly distributed across the wires then there is no problem. The problem is, if the current is not evenly distributed (for whatever reason) you wouldn't have any way to know. Some of the AIB designs may warn you but this is unclear. der8auer had to use thermal imaging and a clamp meter to figure out what was going on. His system would have functioned normally until it melted or caught fire. And even by the end of the video he didn't seem to know why this was happening. Hope he figures it out and lets us know.

 

[Florin]

These wires are connected in the plug. There is no "power distribution" on them because of that, they all should get about the same amount of current. The only way for that to be different is if the connection inside the plug is somehow broken and all the power flows through just one of the pins - which is obviously not an intended behavior.

Yeah so that begs the question, why are people seeing large differences in current flowing across the separate 12V wires?
There isn't any active balancing to complicate the matter. The only thing preventing (close to) equal distribution would be poor connections - or indeed actually broken wires.

Now, quite aside from the fact that ideally you'd want a design that would detect such a case and refuses to work (i.e. something like the 3090Ti), I think we can assume der8auer has made a good faith effort to check that his cable is faultless and is plugged in correctly.

 

[DegustatoR]

Yeah so that begs the question, why are people seeing large differences in current flowing across the separate 12V wires?

So far I've seen just two results with only one of them showing this behavior. It is quite a bit early to say "people" here.

The only thing preventing (close to) equal distribution would be poor connections

Not in case of wires.
Poor connection would put all the current on just some pins in the plug - this is why plugs are melting when not inserted properly (fully).
With wires connection shouldn't matter. The only way they would be overheating is if all the current in the cable between two plugs goes through just some of six (this number could be less or more probably) wires. And this could happen if the connection of all wires inside the plug is broken for some reason.

Honestly I think this is an even less likely failure scenario than the badly inserted plug.

 

[Kaotik]

The only way they would be overheating is if all the current in the cable between two plugs goes through just some of six (this number could be less or more probably) wires. And this could happen if the connection of all wires inside the plug is broken for some reason.

Honestly I think this is an even less likely failure scenario than the badly inserted plug.

Yet this is exactly the scenario happening in Der8auers video

 

[Ext3h]

These wires are connected in the plug. There is no "power distribution" on them because of that, they all should get about the same amount of current. The only way for that to be different is if the connection inside the plug is somehow broken and all the power flows through just one of the pins - which is obviously not an intended behavior.

Quite the opposite. It happened in both cases with exceptionally short, thick cables, and high quality connectors that were still within their rated plug cycle count.

We are talking about 5-6 mOhm of total resistance per wire (20cm, 16AWG) including connectors when it connects perfectly well.

A "mediocre" connection (minor oxidation, less contact area, imperfect press fit etc.) that's still well within tolerances could easily go up to 50 mOhm of resistance. That gives you already a 10:1 ratio in resistance and leads just to that 10:1 ratio in amperage observed.

This wouldn't happen with a longer wire, but it's fatal to run a bundle of short wires when there is no current balancing circuit at either end, and the baseline resistance ends up below the expected tolerances.

 

[DegustatoR]

Yet this is exactly the scenario happening in Der8auers video

We have no idea what exactly is happening in Der8auer's video.
I can think of a hundred ways to burn a GPU. Doesn't mean that any of them are an actual danger to the cards out there.

when there is no current balancing circuit at either end

The ends are fused, there is no need for any current balancing.

 

[homerdog]

BTW it was mentioned earlier there are two shunts on the 5090, and while that is apparently true on some models look at this:



They are in parallel?

At least the FE says YOLO and only has one instead of having 2 for no reason.

 

[iroboto]

I just figured out what a 12V2x6 is.

So, something that comes to mind is that all HDMI connectors can work into another. But not all HDMI cables are equal. Some can handle more bandwidth than others. Some with only 4K and HDR but no DV. Some 8K. Some 4K120.

Cables matter.
And I’m wondering if 3rd party cables designed in the past are not compatible with the 2025 cards. Which is Effed up, but worth considering. The PSU with an12V2X6 may have had enough juice to support a higher power draw, but if the cable was the older spec and not the newer spec we may be seeing that issues on burning.

Choosing to use the cables that come with the card should keep you safe from this. It’s just going to be ugly until you are told specifically that these 3rd party cables are up to 2025 spec.

 

[DegustatoR]

The ends are only fused on the PCB side of the connectors. Not on the wire side.

The wires are connecting to the PCB so why does that matter?
The only way this would be an issue is if the wires would come off or be torn/cut off, and this has been the first theory which was debunked back when the plugs started melting at 40 series launch.
IMO this whole test is incomplete and tells nothing of value without testing other cables with the same card.

 

[arandomguy]

The connector is from what I understand it was pushed as part of the spec by major OEMs as well. I believe OEM RTX 40 series cards (we don't know about 50 series yet) that 4070 ti and above (that would require more than 1x8pin) all use the 12VHPWR connector. These are mass assembled under time pressure but by much more experienced/skilled workers then enthuasists (sorry to break to you guys but it's true, they'll likely build more systems in a week then most enthuasists in a lifetime). They're also all going to be shipped after assembly with the connector in multiple times (from assembly plant -> distribution in between -> user).

These OEMs are going to have a way larger sample size than any individual user and even prebuilt SIs in which they have an conforming assembly standard (due to not relying on individual users and self styled experts). I would wonder what the their failure rate data they have on this is in terms of warranty claims. They face higher liabilities and warranty claims would affect the entire system. But on the surface I'm not aware of them changing course on this issue and it seems like they at least prefer the newer standard. I'm also no aware of ancedotal failures reported for these cards.

Just in terms of ancedotal reports it seems like they are all DIY builds? I know enthuasists let's just say often have a higher opinion of themselves but if the failure reports skews to just DIY builds and not OEMs or SI what does that say about the issue?

 

[iroboto]

The connector is from what I understand it was pushed as part of the spec by major OEMs as well. I believe OEM RTX 40 series cards (we don't know about 50 series yet) that 4070 ti and above (that would require more than 1x8pin) all use the 12VHPWR connector. These are mass assembled under time pressure but by much more experienced/skilled workers then enthuasists (sorry to break to you guys but it's true, they'll likely build more systems in a week then most enthuasists in a lifetime). They're also all going to be shipped after assembly with the connector in multiple times (from assembly plant -> distribution in between -> user).

These OEMs are going to have a way larger sample size than any individual user and even prebuilt SIs in which they have an conforming assembly standard (due to not relying on individual users and self styled experts). I would wonder what the their failure rate data they have on this is in terms of warranty claims. They face higher liabilities and warranty claims would affect the entire system. But on the surface I'm not aware of them changing course on this issue and it seems like they at least prefer the newer standard. I'm also no aware of ancedotal failures reported for these cards.

Just in terms of ancedotal reports it seems like they are all DIY builds? I know enthuasists let's just say often have a higher opinion of themselves but if the failure reports skews to just DIY builds and not OEMs or SI what does that say about the issue?

If this is true it comes down to cable.
DIY aren’t any worse at assembly, that shouldn’t be the case.

But OEM operates by spec not style. They would use only the cables in spec. That’s something DIY folks wouldn’t do because most enthusiasts want to use their enthusiast gear.

 

[Kaotik]

Astire at TechBBS did quick calculations how small resistance differences between cables are enough to cause what Der8auer had on his video, 12V 48A, which is pretty much spot on RTX 5090 TDP.

 

[homerdog]

These OEMs are going to have a way larger sample size than any individual user and even prebuilt SIs in which they have an conforming assembly standard (due to not relying on individual users and self styled experts). I would wonder what the their failure rate data they have on this is in terms of warranty claims. They face higher liabilities and warranty claims would affect the entire system. But on the surface I'm not aware of them changing course on this issue and it seems like they at least prefer the newer standard. I'm also no aware of ancedotal failures reported for these cards.

Just in terms of ancedotal reports it seems like they are all DIY builds? I know enthuasists let's just say often have a higher opinion of themselves but if the failure reports skews to just DIY builds and not OEMs or SI what does that say about the issue?

That wouldn't surprise me but it is possible that DIY users are simply more likely to post about their experiences on PC hardware forums. If a typical user's prebuilt malfunctioned they might just send it back.

 

[arandomguy]

That wouldn't surprise me but it is possible that DIY users are simply more likely to post about their experiences on PC hardware forums. If a typical user's prebuilt malfunctioned they might just send it back.

The ancedotal reporting side is just one aspect. More importantly as I mentioned they would have internal data and the ability, due to volume, as well to go with a different connector if they chose. Now we can't know their thought process for certain but at least on the surface it seems like if failure rates with the new connector were significant they would want to just switch.

Have their been comments from SIs regarding failures? SIs crossover more with tech media and the reporting. Have any of the tech media reached out to SI contacts and asked if they had any failures from the connector?

Puget Systems for example is an SI and they often report on reliability related information (eg. they release hardware reliability information based on their systems, and for example commented on the Intel CPU issue - https://www.pugetsystems.com/blog/2...-perspective-on-intel-cpu-instability-issues/ or switching from Samsung SSDs - https://www.pugetsystems.com/blog/2023/02/02/update-on-samsung-ssd-reliability/ ) and controversies for instance and I don't recall them ever doing something with regards to the 12VHPWR connector. Their public release about hardware reliability for 2024 for graphics cards didn't mention the issue.

 

[arandomguy]

If this is true it comes down to cable.
DIY aren’t any worse at assembly, that shouldn’t be the case.

But OEM operates by spec not style. They would use only the cables in spec. That’s something DIY folks wouldn’t do because most enthusiasts want to use their enthusiast gear.

It wouldn't be that DIY is worse at assembly as a whole (practically speaking assembly isn't all that hard either and the barrier to acceptability is pretty low) but the group consistency is likely lower and less controlled than OEMs and SI. User error related issues can be because the user is not aware of it or their own limitations, and reports from their perspective may not be telling the whole story.

 

[arandomguy]

This isn't directly on 12VHPWR/12V-2x6 reliability but Falcon Northwest (SI) did comment with a larger sample size regarding to I guess the current/temperature measurement issue that was brought up -

https://twitter.com/x/status/1889428378769564121

 

[iroboto]

My before and after. The little power cable is branded nvidia btw. lol. Didn’t expect that.

 

[DegustatoR]

https://videocardz.com/newz/rtx-cablegate-update-geforce-rtx-5090-claims-another-3rd-party-cable-moddiy-now-recommends-to-upgrade-to-newer-12v-2x6-cables

Der8auer’s video has sparked a lot of discussion today. Many reviewers and companies, such as Falcon Northwest, have confirmed testing multiple systems in various configurations and reported no such issues. They also provided high-resolution images from their thermal cameras and their own clamp power meters. This system integrator tested multiple PSUs, cables, and different system builds.

It looks like early RTX 5090 testing has shown some inconsistencies in power distribution. According to hardware reviewer Andreas Schilling, initial tests did reveal uneven power distribution on NVIDIA’s RTX 5090, but a second test showed more balanced results. Meanwhile, some custom variants, like the RTX 5090 Master, seem to have a more even power distribution from the start. This suggests that power management may vary between different models and revisions.

Mods: this probably shouldn't be a part of availability thread?

 

[Ext3h]

Huh, apparently the design guides for 12VHPWR connectors have been silently changed when relabeling to 12V-2x6, and you are now expected to have a metal bridge on both ends of the wire before going into the connector on either end...

@DegustatoR you were right, the wires are supposed to be bridged/fused. But it's seriously dangerous if they are bridged only on the PCB side and the bridge on the wire side of the connector is missing. Then you end up with the wires burning.

Too bad that those 12V-2x6 to 3x PCIe 8-pin connectors are yet another fire hazard though when bridging, as they yet again put multiple paths with ever so slightly different resistance in parallel. Likely case, one of the 8-pin connectors is going to carry most of the load when using such an adapter. And once again, if the 8-pin connector didn't bridge on the wire end, then bridging in the adapter is likewise a hazard as not even the wires of the same 8-pin will end up balanced...

12VHPWR to 3x PCIe 8-pin was safe, but it's not safe to use in a 12V-2x6 system...