It's happening again, melting 12v high pwr connectors

[taka]

Or be a man and tear down your PSU and GPU, remove the connectors with a heat gun and a 100W solder iron. Put some 16AWG-LSZH cables instead. For safety you can add 10Ah fuses for each wire.
Problem solved and warranty terminated, but no worries.

 

[Vayra86]

I'm glad we're at page 25 and we can still just quite simply conclude the cable is underspecced for its job and far too prone to manufacturer variance. Something, again... a child could tell you given the size of the older cable and its rated wattage. Simple game of spot the 10 differences.

Let's go guys, page 50 you can do it. Don't forget to keep buying in the meantime.

 

[Dredi]

far too prone to manufacturer variance.

Or to unbalanced current distribution.

The connector is rated to 10mOhm max. resistance for the connection. Min. is likely super close to 0 Ohm. So assume 0-10mOhm resistance in the connection for a connector that is _within spec_.

The cable has a resistance of 13Ohm per kilometer, so for a 50cm psu cable the resistance is 7.5mOhm.

A total system of two connectors (within spec) and a 50cm cable in between has a resistance ranging from 0-10 + 7.5 + 0-10 Ohm, so 7.5 to 27.5 Ohm.

Now, lets calculate the worst case current distribution for a six lead cable, that is within spec, using https://www.digikey.fi/en/resources...conversion-calculator-circuit-current-divider
So lead 1 with resistance of 7.5 Ohm and the rest at 27.5 Ohm. Total current at 50 A, the max. spec for the cable.

What we get is 21.1 A for lead 1 and 5.8 A for the rest.

FOR A CABLE THAT IS WITHIN SPEC, AND CONNECTED AT EACH END WITH RESISTANCES WHICH ARE ALSO WITHIN SPEC.

The max. current for the individual pins in the amphenol spec is 12 A. It is impossible to guarantee that number without any load balancing, even for cables that are perfectly within spec.

 

[R0H1T]

Or be a man and tear down your PSU and GPU, remove the connectors with a heat gun and a 100W solder iron.

Real men don't need cable

Spoiler

 

[chrcoluk]

Corsair officially reacted to the JayzTwoCents Video on Reddit:
https://www.reddit.com/r/Corsair/comments/1iovz6i/_/mcnjpig

To me that doesnt give confidence.

Someone should have took responsibility on the inconsistent pin height issue on those cables, we know there is issues with pin contact and such and they supplying cables to customers with issues that are visible to the naked eye.

Molex style cables have always had this type of vulnerability, where you need a lot of pressure to connect the cable, and it basically is prone to structural issues.

The proper solution is to ditch these new connectors and stop making 600W gpu's but we know that wont happen, so another option would be to use externally connected power, from back of PSU to back of GPU, or even direct from socket to GPU, main issue being though it has to fit within a gap of a pcie slot bracket.

 

[Bomby569]

In relation to both JohnnyGuru and Aris coming out saying you can't pass 20A though a single wire of a 12VHPWR / 12V2X6 cable, it looks like Aris has changed his stance to you can after Derbauer's latest video:

What I don't get is, why didn't they at least test this before making statements. People were lead to doubt Derbauer and understandably after two experts came out saying 20A+ isn't possible on a single cable.

Now that we know that Derbauer's results are valid, we need someone else to confirm them. I'd also like to see testing done for other modes of failure.

speak first check later, that's why we are in this mess. Shitty experts

 

[R0H1T]

To me that doesnt give confidence.

There's a reason for that, no one wants to touch this radioactive Chernobyl sh!t

Goes without saying that none will take responsibility & thus liability with that.

 

[Degreco]

So here we are - 25 pages later and not any wiser. A week full of uncertainty. YouTubers are now engaged in pathetic wars, fighting over vanity, while their fans wage their wars on the forums.

Can you or can’t you pull 20+ amps with a AWG16 cable? Will it melt instantly, or will it take a couple of minutes?

This is the academic level of discussion we’ve reached.

And while everyone including GN (who apparently need more time), Jay, Roman, Aris and everybody else can’t definitively say whether this is a real issue or not, the forums are full of opinions.

Meanwhile, Nvidia stays silent. Practically ghosting their community. They’re worth trillions of dollars, yet can’t assign a small task force or even issue a brief statement to keep the public from going nuts.

Not that it matters, you can't buy the cards now. But still:
What a very badly managed situation.

 

[R0H1T]

Can you or can’t you pull 20+ amps with a AWG16 cable?

You can probably do a 1000A as well, if it's super cooled. This point of contention is beyond stupid, it's unsafe that's the bottom-line!

 

[Degreco]

You can probably do a 1000A as well, if it's super cooled. This point of contention is beyond stupid, it's unsafe that's the bottom-line!

Yeah i know. I tried to keep this YT nonsense short. They fight for their 'expert' reputation. A vain parade. There is no value in this.

For me, there is only one question: Is the 12VHPWR standard safe or not safe at give or take 600w for the DIY market?

 

[R0H1T]

I keep coming back to this marketing blitzkrieg by JHH & the whole media ecosystem, starting with ~

When your whole life flashes before your eyes, how much of it do you want to not have ray tracing?

Just Buy It: Why Nvidia RTX GPUs Are Worth the Money

Some enthusiasts are upset about the price of the Nvidia RTX 2080 / 2070, but the opportunity cost of sticking with old technology is higher.

www.tomshardware.com

And yet they still keep feeding it

 

[Vayra86]

Or to unbalanced current distribution.

The connector is rated to 10mOhm max. resistance for the connection. Min. is likely super close to 0 Ohm. So assume 0-10mOhm resistance in the connection for a connector that is _within spec_.

The cable has a resistance of 13Ohm per kilometer, so for a 50cm psu cable the resistance is 7.5mOhm.

A total system of two connectors (within spec) and a 50cm cable in between has a resistance ranging from 0-10 + 7.5 + 0-10 Ohm, so 7.5 to 27.5 Ohm.

Now, lets calculate the worst case current distribution for a six lead cable, that is within spec, using https://www.digikey.fi/en/resources...conversion-calculator-circuit-current-divider
So lead 1 with resistance of 7.5 Ohm and the rest at 27.5 Ohm. Total current at 50 A, the max. spec for the cable.

What we get is 21.1 A for lead 1 and 5.8 A for the rest.

FOR A CABLE THAT IS WITHIN SPEC, AND CONNECTED AT EACH END WITH RESISTANCES WHICH ARE ALSO WITHIN SPEC.

The max. current for the individual pins in the amphenol spec is 12 A. It is impossible to guarantee that number without any load balancing, even for cables that are perfectly within spec.

Aye so connect a few more, not unlike the past.

Yeah i know. I tried to keep this YT nonsense short. They fight for their 'expert' reputation. A vain parade. There is no value in this.

For me, there is only one question: Is the 12VHPWR standard safe or not safe at give or take 600w for the DIY market?

The simple answer is no and all you need is to compare cold hard numbers between this pos and Pcie connectors/cables standards.

Everything else is cognitive dissonance.

 

[Dredi]

Aye so connect a few more, not unlike the past.

For 1200W cards? 1800W?

Why?

With two in spec cables the problem would still persist. Maybe even with three, but that would maybe be pretty close to being ok.

Madness. Just do some load balancing, not unlike the past, and you are golden.

 

[Vayra86]

For 1200W cards? 1800W?

Why?

Its not 1800W you've got. That's the max draw the trio of cables can have. And we've clearly seen they can't have that.

In my view you have 3x375W at that point. That's just over 1K. And since the 5090 can have transient peaks of 900W, I think that seems about right.
Two? You could probably wing it, too.

One? Well, even the 4090 isn't happy with one apparently.

 

[Dredi]

Its not 1800W you've got. That's the max draw the trio of cables can have. And we've clearly seen they can't have that.

In my view you have 3x375W at that point. That's just over 1K. And since the 5090 can have transient peaks of 900W, I think that seems about right.
Two? You could probably wing it, too.

One? Well, even the 4090 isn't happy with one apparently.

Just do load balancing, not unlike the past, and forget wasteful excess cables.

I’m pretty sure not even three cables would be enough to keep amps below 12 without load balancing.

 

[Vayra86]

Just do load balancing, not unlike the past, and forget wasteful excess cables.

I’m pretty sure not even three cables would be enough to keep amps below 12 without load balancing.

If only we had a way to test that.

 

[Endymio]

In my view you have 3x375W at that point. That's just over 1K. And since the 5090 can have transient peaks of 900W, I think that seems about right.

The specifications refer to sustained load -- they can handle far more for ms-duration transient peaks.

 

[_roman_]

The connector is rated to 10mOhm max. resistance for the connection. Min. is likely super close to 0 Ohm. So assume 0-10mOhm resistance in the connection for a connector that is _within spec_.

Just don't

the proper unit for that should be anyway

Siemens (unit) - Wikipedia

en.wikipedia.org

You may read this first
https://en.wikipedia.org/wiki/Metric_prefix
the line with milli - like in millimeters if you are from central europe / millliliter

No offense, but it seems you never had any physics, electronics lessons. And held a true rms multimeter in the hands for several years and used it.

there is also a wire modell which is completely ignored. Just because it is ignored does not mean it does not count or is important

Common Knowledge for the "0 Ohms - Wire"

Superconductivity - Wikipedia

en.wikipedia.org

Let's go guys, page 50 you can do it.

As long a moderator does not lock up the topic. Or deletes messages.

--

while (true)
{
print: "Random User 8124 claims - I do not have any issues. So everything fine. No problem.";
delay 5 posts;
}

This is the academic level of discussion we’ve reached.

Physics / Electronic basics

 

[Endymio]

Just don't

the proper unit for that should be anyway

Siemens (unit) - Wikipedia

en.wikipedia.org

You may read this first
https://en.wikipedia.org/wiki/Metric_prefix
the line with milli - like in millimeters if you are from central europe / millliliter

No offense, but it seems you never had any physics, electronics lessons. And held a true rms multimeter in the hands for several years and used it.

If you're going to be pretentious, at least be accurate. The correct unit is ohms -- Siemens is merely the reciprocal of that, and is far less commonly used in this context. Nor is mOhm inaccurate either. The only mistake he made was in assuming any connector can be truly 0 resistance...there's always going to be a few uOhms at least.

 

[harm9963]

Got my Seasonic Prime TX 1600 WATT ATX3.1 a month ago , has two 12v 2x6 connectors' and cables , but never had issues even with a 1200watt ATX 3.0 , since two years ago , but wasn't about to use it for a 5090 , and I have WireView to check how much amps are uses, plus a clamp meter to check for distribution per wire ,small price compare to spending over 2K a GPU , mitigation , not foolproof !

Seasonic Support Desk<supportdesk@seasonic.com>






To:You
Mon 2/3/2025 7:09 PM

Dear Harmon,

Thank you for your reply.

The PSU is ATX 3.1 and PCIe 5.1. Intel simplified the naming of ATX 3.1 to ATX 3 and PCIe 5 to avoid confusion. Therefore, our PRIME TX-1600, which follows ATX 3.1 and PCIe 5.1 standards, will include the 12V-2x6 cable and connector on the PSU side. You can check the connector on the PSU; it should have shorter sense wires and longer power pins. The packaging will also feature a 12V-2x6 logo to differentiate it from previous ATX 3.0 and PCIe 5.0 (12VHPWR) versions.

If you have any other questions, please let us know.
Thank you.

 

[The Plague]

It's sad to watch Aris destroying his and Cybenetics reputation just because of ego... not being able to admit he was wrong and Roman was right.

 

[Dredi]

If only we had a way to test that.

No need to test. Just read the specs for the parts and calculate worst case conditions for within spec operation.

Just don't

You do it then. I'd very much like to hear where I went off the rails.

You may read this first
https://en.wikipedia.org/wiki/Metric_prefix
the line with milli - like in millimeters if you are from central europe / millliliter

And? I fail to see the problem.

No offense, but it seems you never had any physics, electronics lessons. And held a true rms multimeter in the hands for several years and used it.

I've designed and built test harnesses for satellite instruments. I'm pretty sure I understand basic resistance calculations.

Common Knowledge for the "0 Ohms - Wire"

Where did someone assume a 0 Ohm wire? I do not understand you. Maybe you've had a beer too many, it would explain the sassy attitude and typos too.

The only mistake he made was in assuming any connector can be truly 0 resistance...there's always going to be a few uOhms at least.

Low enough for it not to matter in this case. I did round the other values too.

 

[evernessince]

Just do load balancing, not unlike the past, and forget wasteful excess cables.

I’m pretty sure not even three cables would be enough to keep amps below 12 without load balancing.

Load balancing would only cover up the other issues the connector has. You still have thin wires (for the 600w rating), a low mate count, low durability, and a low safety margin.

100% the entire connector needs to be redesigned. I don't see the point in half-assing it by just fixing the load balancing on the GPU end when we know all these issues exist.

It's sad to watch Aris destroying his and Cybenetics reputation just because of ego... not being able to admit he was wrong and Roman was right.

It seems to be a misunderstanding but yeah he should not have posted that video. It's still not too late for him to issue an apology.

Speaking of Aris, he released a video stating that the point of shunt resistors on the GPU aren't there to balance load but measure voltage drops:

Double checking this claim, it appears that shunt resistors measure electrical current by measuring the voltage drop across the resistor: https://eepower.com/resistor-guide/resistor-applications/shunt-resistor/#

According to the diagrams Buildzoid has shown, those shunts would then be integrated in a parallel circuit (in older designs) to give multiple groupings of pins / wires the ability to measure the current and then adjust as needed. Of course, not so much for the 4000 and 5000 series as the circuit combines all the pins / wires into one before and after the shunts. In the later scenario the shunts are not balancing the load between the pins / wires of the cable.

Yeah the shunts themselves aren't load balancing but they are enabling it along with a change in circuit design by running a separate dedicated wire to each wire/pin. To me Aris is trying to be very technical but is missing the forest though the trees and potentially causing drama as a result.

This is past my limit of knowledge, would be great is someone with experience in GPU Design / Modding could chime in.

 

[taka]

Let's take this EVGA 3090Ti as example. This PCB uses 24+3 phases design. This specific card uses only 21+3 phases.

The 12V is split in three circuits monitored by a voltmeter connected to the shunts on the upper right side.
From there the circuits are going in the copper layers inside PCB.

All three circuits are connected to a block of 8 down step MOSFET convertors.
The 3 separate ones are powered from the PCIE bus and are powering the RAM.

The circuits in the bottom right are the shunt voltage monitors and PWM controllers for the fans.

This circuit is balancing the 12V line into 3 separate VRM's controlled by the same IC. The control IC is the one making the balance based on shunt voltage drop and input + output voltage on the MOSFET's.

 

[Dredi]

Speaking of Aris, he released a video stating that the point of shunt resistors on the GPU aren't there to balance load but measure voltage drops:

Depends on the gpu. If this is about the one asus 5090 with shunts, it can only provide diagnostics and some passive load balancing due to the increased resistance of the circuit. So the ”best case” resistance is up by 2mOhm, which helps a little bit.

Some older gpus had proper balancing going on, necessitated by having multiple power cables connected with differing power ratings.