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

[ty_ger]

Clamp meter must be very expensive, since Aris doesn't even have one and Igor only uses two for 6 wires.

They are, for accurate DC meters. Hall effect measurements and all that. AC clamp meters are cheap. Cheaper ones are cheap, but not accurate or reliable. A jouranlist won't bet their reputation on them.

 

[BoggledBeagle]

Mine was $140.

 

[Endymio]

You...might not understand this, but a recall is not an incidence of failure. ... For dangerous stuff a recall cynically exists when the cost of lawsuits is too high, and optimistically is done the second there's a practical chance of harming the consumer. There are recalls for airbags, switches, etc... They are not recalling 2.8 million cars for "explosions during regular operation." If they would, then it'd be a huge news story.

Why is logic so difficult for some people? The point of the recall is that an engineering mistake was made and detected, and to correct it BEFORE some potentially catastrophic failure occurs.

As for the "explosions during normal operation" -- you have video of 5090s exploding and killing people? That would indeed be a "huge news story".

Edit-
To clarify, melting power cables is not the "low issue" you seem to pretend.

It's so much fun to put false words in people's mouths, isn't it? I said no such thing. My point -- to reiterate in the simplest possible terms -- is that engineering failures are not nearly as uncommon as you pretend. Engineering a tiny tweak to a long-existing design is one thing. But when you step into uncharted waters -- be it a rocket to Mars or a video card using unprecedented power levels -- mistakes will be made. Sell your Nvidia hate elsewhere. If an actual design flaw with the cards exists -- rather than substandard cables and/or connectors -- it'll be addressed. In the meantime, go find us some video of exploding 5090s.

 

[ty_ger]

Mine was $140.

Good for you. Do you think a journalist will use that model?
They will use a respected brand and a respected model of that brand that has good specs. Then they will pay a calibration lab to verify that it meets the stated specs. Then they will use it. Price after all that? Over $1000, easy. Like I said, not cheap.

 

[Endymio]

They are, for accurate DC meters ... Cheaper ones are cheap, but not accurate or reliable. A jouranlist won't bet their reputation on them.

Yet they'll risk that reputation by taking a far more inaccurate reading on six wires at six different points in time, then claiming the "load is imbalanced"?

 

[Frick]

video card using unprecedented power levels -- mistakes will be made.

Not entirely unprecedented. This is Furmark numbers, but still.

 

[BoggledBeagle]

Yet they'll risk that reputation by taking a far more inaccurate reading on six wires at six different points in time, then claiming the "load is imbalanced"?

The current distribution on the wires is determined by resistance of the wires and contact resistance in the contacts. At the ends everything is connected together. If you run a constant load, as furmark, there is no problem in making the measurements in different times.

 

[Endymio]

Not entirely unprecedented. This is Furmark numbers, but still. [graphic deleted]

The AMD 295X2 had a TDP of 500W; spikes of 600W+ were brief. And even still, there were many reports of melted connectors and cables with that card, no?

The current distribution on the wires is determined by resistance of the wires and contact resistance in the contacts. At the ends everything is connected together. If you run a constant load, as furmark, there is no problem in making the measurements in different times.

There is no such thing as a "constant load" for these cards. Even under a constant software demand (such as Furmark) the card has ms-duration spikes both up and down, varying by several hundred watts.

 

[A Computer Guy]

The current distribution on the wires is determined by resistance of the wires and contact resistance in the contacts. At the ends everything is connected together.

I remember seeing a cable diagram that internally the 12v wires are not necessarily bridged together in the same way at the connector depending on the cable. They might all bridged, only a few at a time might be bridged, and in the case of custom cables they might not be bridged at all and you essentially get 1 wire to 1 pin from the PSU. Feel free to correct me if I'm wrong.

If you run a constant load, as furmark, there is no problem in making the measurements in different times.

 

[JAS_design]

If we're talking about reasons why a cable (particularly a custom cable) might have different resistances on the different wires, here's a potentially obscure reason that would exist from the moment the cable was manufactured:

Were the terminals crimped with a generic crimp tool, or with the (usually absurdly expensive) "official" crimp tool?

An official crimp tool almost always does a better job. It's not a huge difference considering the price of the official tools, and generics are more than adequate for most uses. But in edge cases where every milliohm matters, it could be the difference between success or failure.

 

[wNotyarD]

I remember seeing a cable diagram that internally the 12v wires are not necessarily bridged together in the same way at the connector depending on the cable. They might all bridged, only a few at a time might be bridged, and in the case of custom cables they might not be bridged at all and you essentially get 1 wire to 1 pin from the PSU. Feel free to correct me if I'm wrong.

That is correct, but the issue is how the graphics card have its power connector terminated. If it terminates all 12V input wires into a bus then into the card, or if it monitors each wire like the Astrals do.

 

[ty_ger]

Yet they'll risk that reputation by taking a far more inaccurate reading on six wires at six different points in time, then claiming the "load is imbalanced"?

I cannot confirm or speculate why someone does something in a way that you disprove of. I was only responding to the question asked regarding price of reputable DC clamp meters.

 

[Degreco]

What follows is pure speculation! Just a theory on what is going on with Occam's razor.

TLDR: Debris can act like 12V-2x6 was never invented.

What makes me wondering are the reports that reseating the cable connections can improve on (re-)balancing the amps on all lines. Some guy posted this recently on Reddit. I know, never trust reddit posts, but still...

This made me thinking.

Simple logic: If reseating and reconnecting the cable can improve the connection, it should also be able to decrease the quality of the connection either. That said, one can probably rule out improper connections because I think most Blackwell users (currently a small group of enthusiasts and creators own those cards at the moment) are probably aware of the old 12vhpwr connection probs. I'm sure most of them will double check the connectors, despite being 12V-2x6 cables or not, ensuring proper connections to GPU, Nvidia adapter and the PSU. I know, I would (and did check like a paranoid monkey multiple times with my 4090, e.g. when moving the rig - because I don't trust the 12vhpwr standard at all).

So let's rule out improper connections, will we?

Too make things simple let's also rule out cable quality differences (again, before you object, Occam's razor).

Let's further take into account that lab testing at Nvidia most probaly did not reveal a problem. Also, that most users (well, we don't know yet, I just try to get a statistical best guess by appearance of posts and known numbers like the microcenter lists) don't have a problem and lines are usually amp-balanced.

Also, that Nvidia stays silent at the moment. Speculation why: 1. The issue is not instantly deniable ('user error').
2. They need more time to understand the issue.
3. They need more time to understand the number of melting event and statistics (that's what you get for paperlaunching - to little statistical information).

Then only ONE probable thing is left in the equation (based on some people having a problem and most others having none) and that is what the experts call foreign debris. What if microscopic clumps of dust or tiniest lints in the cable or connector are weakening the connections to the pins from within? Then reseating and reconnections sometimes helps and sometimes worsens the connection. Obviously, the debris issue is always a problem with any cable and GPU of any Gen, but could get uglier with 12VHPWR driven at the limit. The less headroom you have, the more massive the problem. BuildZoid and some Intel engineer recently spoke about the general fragility of the 12VHPWR standard. Nvidia skipping the shunt resistors on 4090/5090 certainly did not help. And then comes the debris at 600w.

Just a theory. But testable. Someone bold (and wealthy, because you can brick things) could test this by putting a tiny speck of dust into a 12v line on a connector on the GPU or Cable or on the PSU before connecting everything. Then fire up Furmark and measure how Amps (and temps) are going up or not. Most likely, debris will produce an amps problem on a pin somewhere and thus the resulting heat will end up melting something somewhere. Because PSUs, despite having loads of safety triggers usually have no load line balance or load intelligence (exactly like the 5090 on the other end of the string) they will try to push harder and harder with debris - they will eventually shut off but maybe too late. It's all still a dumb design that gets dumber if the asking end (the GPU) is also load line dumb. Debris will then result in undetected resistance, then comes the heat, over time comes the melting, if you are unlucky and your PSU does not shut off. If this theory is true, then the effect goes likes this:
1. The more the right amount of debris(not too little, to make any difference) is somewhere in the line the more risky is the connection at full power to become imbalanced. But there are scenarios thinkable, where debris will not result in a perceivable problem.
2. Reconnecting is not a problem (even if you only have an finite amount of reconnects on any standard and a tiny amount on the 12VHPWR variants especially). But the more you reconnect the cable to the connectors, the more likely you will introduce debris into the chain.
3. Debris can act like 12V-2X6 was never invented, because it can fully block the contact at pin level, thus the system is behaving like an improperly seated 12VHPWR connector, even when you have 12V-2X6-only hardware and 100% correct connections.
4. The stronger the PSU (1500w or more should be best for trying out), the more imbalance will be there and the faster is heat generated. The 5090 asking for large PSUs does not help either. After a Furmark-while enough heat will be accumulated to start the melting somewhere in the power path between PSU, cable connection points and GPU connector.

You don't need a infrared thermometer or current clamps to do this test - you can simply test with the finger method - the cable (one or a couple of lines in it) should feel hot (if it feels just hot, it's too hot in reality, because it's probably sleeved etc., the copper itself should be way too hot then, hence resistance somewhere).

Speculation! I may be completely wrong.

 

[SmookinJoe]

Jay just posted something you guys can chew on...

Hint...some of the pins on some connectors are receding
8:45+minute point...
Oh..Oh..Oh...Jay is using a device that he shows at the end of the video that protects the card...IMPORTANT info..
PMD2? saving the card...
PMD2

 

[_roman_]

Yet they'll risk that reputation by taking a far more inaccurate reading on six wires at six different points in time, then claiming the "load is imbalanced"?

Thanks to the guy who provided pictures and measurements on his card. I'll just reuse the picture.

Please add additional informations.

In my point of view, I want to proof if there is one wire with too high current. Therefore a clamp meter works awesome.

You can use a clampmeter when you use more than one wire to measure all 12V DC lines or only all GND lines.
I do that e.g. on my car. only negative side or positive side wire from the car battery. Or only one wire. You should not mix wires were the current flows in different directions. (bad explanation with my words)

I also have a Voltcraft (VC-607) clampmeter. I bought it with my own money, not as an adult, in education.

I learnt in the past how to calculate the accuracy. With the paper handbook which any proper clampmeter / multimeter has you can calculate the accuracy. The important question is how unbalanced the current is. Therefore valid, like post #451 did. the measurement error is most likely around +/- 2% + some digits for a TRUE RMS clampmeter

It depends on what you want to proof or show. When you want to see how unbalanced the current is, post #451 is valid and awesome.

if you want to be igorslab you need those current probes for oscilloscope and an expensive oscilloscope. Most likely with computer read out software and such. I'm out of that expensive field for years. Than you can write awesome igorslab articles and show those current spikes pictures from e.g. a graphic card

edit: picture was not mine - I just took it for reference

 

[agent_x007]

Jay just posted something you guys can chew on...

 

[SmookinJoe]

Please add additional informations.

In my point of view, I want to proof if there is one wire with too high current. Therefore a clamp meter works awesome.

Thanks to the guy who provided pictures and measurements on his card. I'll just reuse the picture.

You can use a clampmeter when you use more than one wire to measure all 12V DC lines or only all GND lines.
I do that e.g. on my car. only negative side or positive side wire from the car battery. Or only one wire. You should not mix wires were the current flows in different directions. (bad explanation with my words)

I also have a Voltcraft (VC-607) clampmeter. I bought it with my own money, not as an adult, in education.

I learnt in the past how to calculate the accuracy. With the paper handbook which any proper clampmeter / multimeter has you can calculate the accuracy. The important question is how unbalanced the current is. Therefore valid, like post #451 did. the measurement error is most likely around +/- 2% + some digits for a TRUE RMS clampmeter

It depends on what you want to proof or show. When you want to see how unbalanced the current is, post #451 is valid and awesome.

if you want to be igorslab you need those current probes for oscilloscope and an expensive oscilloscope. Most likely with computer read out software and such. I'm out of that expensive field for years. Than you can write awesome igorslab articles and show those current spikes pictures from e.g. a graphic card

Watch the Jayz2cents video
Your cable looks like one of the problem childs in the group......

 

[JAS_design]

Jay just posted something you guys can chew on...

Hint...some of the pins on some connectors are receding
8:45+minute point...

In my previous post I mentioned that using a generic crimp tool vs the "official" crimp tool might make a difference.

This movement of the terminal in the housing is the sort of thing you might see if the terminal has been crimped badly, and has been somewhat distorted. If the tangs on the terminal aren't gripping the housing properly, or if the housing itself is damaged during insertion of a malformed terminal... this movement can happen.

Plus, if for some reason the pin doesn't fit into the socket properly, again because of some kind of misalignment or malformation, the socket can be pushed back into the housing by the pin.

This strikes me as a possible QC issue by the manufacturer.

 

[ShrimpBrime]

Jay just posted something you guys can chew on...

Hint...some of the pins on some connectors are receding
8:45+minute point...
Oh..Oh..Oh...Jay is using a device that he shows at the end of the video that protects the card...IMPORTANT info..
PMD2? saving the card...
PMD2

Once the connector melts, the card is junk. RMA time. Doesnt matter if the gpu was saved, nobody should give a shit about that if you have a puddle of plastic at the bottom of the case. XD

 

[freeagent]

This movement of the terminal in the housing is the sort of thing you might see if the terminal has been crimped badly,

I have seen workplace videos of people assembling cables, very tired, kind of sloppy.. what you are saying makes sense to me.

 

[ty_ger]

We should go back to 8-pin cables so we can see if they start failing. Certainly the defects and tired workers only work at the 12-pin cable factories. Or, is it that the cable is not robust, pushed to the limit, and ANY mistake ANYWHERE will cause them to fail?

 

[ShrimpBrime]

5090 GPU owner
"You don't care for me, I don't care about that."

Jacket man
"Got not food?, I like it like that!"

Guy without 5090
"Come'on and let me stand next to your fire!"

 

[SmookinJoe]

Jay was pointing out that some 3rd party sellers connectors were the best of the bunch. The supplied ones from some power supply manufacturers were bad.
AND the adapters from Nvidea (and MSI) were one of the better ones...hmmm
I guess I should feel pretty safe, cause that is all I use...
Once they are plugged into my cards that is where they stay..
Jay also mentioned that Nvidea went to a common rail with the 4090 and newer which in his opinion was a step backward from previous cards where they split the power, amongst the pins ..
You guys know what I am trying to say...
Remember "I Know Nothing" and I am just relaying what I remember in the video..
Good luck everybody...

 

[eidairaman1]

Fight Club taught us:

If A x B x C = X and if X is less than the cost of a recall, we don't do one.

Nvidia knows what it is doing.

Nv=deny the truth

 

[Endymio]

Nv=deny the truth

To coin a phrase entirely from scratch: "you can't handle the truth". If you have a significant imbalance in individual wire current, then you have a significant imbalance in resistance -- which means the cable pin and/or connector mating is the problem. If current load actually was on a per-pin basis, this issue would be happening on most or all cards.

Now you can blame NVidia if you want for not designing in a larger safety factor, but the fact remains that if you have an in-spec cable properly mated, you won't have issues.