Raptor_iHow no one can force Nvidia to recall cards with this connector or use a new standard, we didn't have such issues with 3x 8 pins ... It's simple physics: high load over small pins = overheating :banghead:

It really does seem to be more than just the connector, but the card’s board design does not allow for proper load balancing, which can result in critical amperage levels on some lines. A bad connection is one thing, but it sure sounds like the GPU simply doesn’t manage the voltage-in correctly. The high-end cards are just such power pigs that they expose the design flaw in a catastrophic fashion.

I wonder why they didn't just adopt EPS instead. It's a standard that isn't going to change any time soon, and 2 4-Pin EPS provides more power than 1 8-Pin PCI cable (around 280w per 8-Pin EPS vs 150w per 8-Pin PCI). Further, some modular PSUs already have both EPS and PCI power plugs using the same socket on the PSU, so they could have just tossed in extra EPS cables instead of PCI.

Engineered to fail. End of statement...

A Computer GuySad a $15,000 GPU lacks the necessary protection circuitry in 2026.

There are tons of expensive equipment everywhere in use that have no specific connector related protection circuitry and just presume that the connectors will work properly - that most of the time means maintaining high enough contact force. Problems with corrosion in bad enviroments of course exist, but I do not think enviroments in which you find normal personal computers or workstations deployed are usually very corrosive.

Loose connectors just burn out. That is a very common electrical failure. The reason is mechanical - too low contact force leading to too small contact area leading to too high current density which means too high heat production.

You do not approach a burned out connector hypothetising about complex balancing circuitry missing or other crazy stuff, you just think: aha, this connector was too loose and that it why it burned out. You do not approach a traditional bulb not producing light anymore thinking aliens hit it with a death beam, but thinking it happened because the filament just broke or melted due to mechanical shock, age or too high voltage - because that is how normal bulbs break.

Again, and again: Connectors burn out because they get too loose. PERIOD!

BoggledBeagleThere are tons of expensive equipment everywhere in use that have no specific connector related protection circuitry and just presume that the connectors will work properly - that most of the time means maintaining high enough contact force. Problems with corrosion in bad enviroments of course exist, but I do not think enviroments in which you find normal personal computers or workstations deployed are usually very corrosive.

Loose connectors just burn out. That is a very common electrical failure. The reason is mechanical - too low contact force leading to too small contact area leading to too high current density which means too high heat production.

You do not approach a burned out connector hypothetising about complex balancing circuitry missing or other crazy stuff, you just think: aha, this connector was too loose and that it why it burned out. You do not approach a traditional bulb not producing light anymore thinking aliens hit it with a death beam, but thinking it happened because the filament just broke or melted due to mechanical shock, age or too high voltage - because that is how normal bulbs break.

Again, and again: Connectors burn out because they get too loose. PERIOD!

There is a possibility that a lot of them come "predamaged" from the factory.

These expensive equipment have simple one line power delivery (i.e. 110 V/230 V or 48 V), not multiline with unknown characteristics of current flow. These expensive equipment also does not use connectors which is very susceptible to bad connection or without necessary margin.
Also, power delivery connector chosen is one of points of device electrical design. Choosing wrong makes device design wrong.

Again, and again: Connector design is the problem.

Its not the connector or the standard guys, its user error

Sheesh

We get a case of a burned connector a week. That's 52 cards a year. Let's multiply that by 10. That's 500 cards a year. Let's multiply it by another 10. That's 5000 cards a year. Let's multiply it by another 10. That's 50.000 cards a year. Nvidia is selling 10s of millions of GPUs a year using said connector. There are estimated to be more than 50 million GPUs out there since 2022 with the new connector. That's a failure rate of 0,001%. That's after I multiplied the cases 10,000fold. That's how big of an issue it is...

In the last 5 years I had 3 mobos fail on me, 2 of them where RMAed. In the meanwhile my 12vhp cable has been working flawlessly nonstop, but somehow if I buy "ngreedia with taht melting houseburning connector" im an idiot. Okay

Connector is junk and should have been discarded on design table.

However, electrical design of RTX 5x00 series is also junk.

Also, regardless of design, it is stupid idea to send 600W through one connector in tight space prone to heating.

Vayra86Its not the connector or the standard guys, its user error

Sheesh

Absolutely!
Lets be clear here, the Connector is FINE! IT JUST WORKS!
All these reported failures (most of them are fake and AI-Slop of course) are only happening, because people are lacking the necessary intelligence to use the Connector correctly!
Its either user Error (very likely), or a $5.27 trillion Companies incompetence to design a Connector (practically impossible).

GrimlingAbsolutely!
Lets be clear here, the Connector is FINE! IT JUST WORKS!
All these reported failures (most of them are fake and AI-Slop of course) are only happening, because people are lacking the necessary intelligence to use the Connector correctly!
Its either user Error (very likely), or a $5.27 trillion Companies incompetence to design a Connector (practically impossible).

Cannot tell if joking or serious

This could easily be a 24v 2x4 cable and connector. People buy new 1000w PSUs for this card... make them buy a 24v PSU...

Passive it may be also coupled to the GND pins.

The hot, 12V DC pins are moving the current. WE all know how big pins when you want to desolder, need a lot of head. The reason why the cheap shit 30W Ersa soldering iron did not work and i bought a ERSA RDS 80 wiht lots of soldering heat.

ShrekThe ground pins are just as susceptible to heating as the 12V pins.

nope.

I brought the soldering iron example on purpose. You can verify it yourself. how long does it take to heat

I do not own a hot air station, but it makes sense to use a small nozzle with lots of hot air to solder components.

A infred thermometer could also show you that, how electronics have high heat spots and lower one. No money and no will to buy a thermal camera which may suit this purpose better.

--

edit: Measuring on the GND pins is nonsense. Igor and other already claimed that the current does not flow back 100%. It would be different when the hole pcie slot were only optocouplers.

-- typical asrock fail, i am not surprised.

-- the power supply is not according to specification. When you can draw more power as specified you need to have proper connectors. Limit it to e.g. 600 Watts and use a 600Watt connector.
note: that stuff is such obfuscated that finding a spec paper, and all the stuff is too time consuming. fill in the number which is correct 600 Watts , 650 Watts, etc.

I do not know what the reference of 1300 Watts is in the text. Was 1300 Watts used in this scenario for the 12 Volt "high power" (lol) v1 or v2 "600" Watt connector?

I do not consider the power strip with around 3.6kW as high power. Or my hair dryer which has around afaik 1200 Watts as high power.
High power are 30 Amps @ 400V AC 50 hz. not 600 Watts

My water kettle has for sure more than 600 Watts. ... Is it high power?

JustBenchingThere is a possibility that a lot of them come "predamaged" from the factory.

I believe that a lot of manufacturers do not pay enough attention that this small connector needs when it it pushed pretty hard with 600W running through it. I doubt that the manufacturers check meticulously each batch of supplied sheet metal, if it has good enough mechanical properties, etc. I even suspect that a lot of the manufacturers even do not have labs equipped so well that they are able to check the materials.

KenjiroAgain, and again: Connector design is the problem.

This is a standard connector that exists for a long time. This is not a new design - only the decision to run 600W through it is new.

ShrekSure, one can have a 12V current and a 5V current and these combine at the ground wire, but here we have only 12V,

you condradict yourself in the same post

Igor failed to explain Kirchhoff. It seems you did not grasp yet Kirchhoff, although i tried to explain it to you in personal message several weeks ago.

You may read the kirchhoff laws first. Basics. Igor had a very low quality german article about it. The sum of currents are always zero, the sum of voltages is always zero -> read yourself kirhchoff law - wikipedia has a low quality text article about it.

A PSU has several GND planes and different connectors with GND. You may cut all the other GND wires on the psu than this Asrock cable will work correctly, when only the 12V high power connector has the only gnd wires left.

TLDR: Kirchhoff Laws

Imagine using 24 or 48V with lower Amps, instead of that 12V thing... Just imagine.
They will totally cut the heat out of equation, but I guess you need a PHD in electrotechnics for that, that a 5 trillion company cannot afford.

ShrekSure, one can have a 12V current and a 5V current and these combine at the ground wire, but here we have only 12V,
so the two currents (12V and Ground) must match as is confirmed by igor's measurements.

Every GND line in total has the summary of the currents. as stated by kirchhoff.

There are a lot of GND pins on a power supply unit on multiple connectors, not only the 12V high power one

WirkoWe were taught about Kirchhoff's laws here, what's the name that Americans use?

Shrek asked me in personal message weeks ago and i tried to explain
what DELTA means and what kirhchoff means

with references to my school books in german / other german technical electronics books + isbn numbers

I do admit, i did not read the last pm than.

English call it kirchhoff law, as i had to look it up for Shrek at that time. Wikipedia page exists.

I will not post any personal message in public. i can mention it but i will not post it

-- The school i attented was 5 years and not everyone grasped it. I myself had to read it several times as a teenager and apply it. When you do not grasp kirchhoff you assume it wrong. I will not question core principles. I have a True rms current clamp meter multimetrter and i verified it anyway on my car anyway 20 years ago. A Diesel Car is a nice example, alternator, generator, battery, charger, etc..

The current limitations were told already over and over again. Kirchhoff Law were also told over and over gain.

WirkoMOSFETS don't need them, they somehow self-balance (something something temperature coefficient, I forgot)

you usually thermal couple power electronics. It has a reason why power electronics are on the same heatsink. So they are on the same thermal field of the specification characteristics.

mosfet usually it is enough to thermal couple those. that is also the reason why some need an insulator between the power electronic component to the heatsink.

The RDS on value also matters with other characteristics. Do not ask me for specifics, I did not deal with that stuff for 20 years by now. -- ohm law also applies

When you thermal couple those, the component which gets hotter changes characteristics is prevented by thermal coupling those. With its own heatsink you would have a runaway component.

ShrekThose 16 balance resistors are rated for 10W each, so one wonders how much power is lost to balancing in that 300W amplifier.

The inefficency of Bipolar transistors are a given and well known fact

The inefficiency of Class A amplifier is a given

I have a reason why i bought a class D amplifier, a ADAM T8V studio monitor. Another word for an active speaker with flat line input to output frequency speaker.

Those bipolar transistor designs are technology from year 1980 in my point of view and some fancy marketing stuff.

-- we should stick to the core topic, the 12 V high power connector, gen 1 and gen2

Darmok N JaladThose with more electrical knowledge can maybe help me out, but what's the difference between this situation and say, a 600+W amplifier that's powered by a 12V line in an automobile? Those typically are fed off one 12V line and a ground. Are they not comparable, or is it just a matter of a larger gauge of wire doing the job?

I only watch from far away.

Usually you have 1 Farad 12V DC capacitor in the car audio system. 1 Farad is very big capacitor to buffer high demand spikes.

The wires are bigger and use a different connector.

I think around 1996 someone explained me in school about the "skin effect". Power electronics sometimes behave similar as the high frequency field which is more my field of expertise.
You need to look into physics, solid core, vs not solid core, how many small wires in a cable, etc.
physics / power electronics / high frequency electronics will answer that. That is a lot of different materials to read.
I am sure there are books with tables, spec sheets etc, which explain it.

I do not know any car audio system which usees the intel / nvidia / whatever else 12V high power connector.

-- Those car amps, i am not sure, are peak watts. That is also a rabbit hole to read and explain. Someone else should elaborate and correct my statement aobut peak power in speaker, vs amplifier, vs efficiency of amplifiers in audio equipment etc. That will be a lengthy text where most likely most will skip reading frustrated at some point

When we talk about efficiency, i think i remember 25% for class a amplifiers efficiency. What wirko showed with teh year 1980 bipolar transistor design.

-- Do not be early adopter. I bought on purpose AM5 in 2023 and not earlier. Let other tests and taste the green bananas which slowly turn yellow or tasty, or which are turning bad.

Vayra86Its not the connector or the standard guys, its user error

Sheesh

Hey in the end everything is user error, you shouldn't buy broken garbage, that's on you.

_roman_Or my hair dryer which has around afaik 1200 Watts as high power.
High power are 30 Amps @ 400V AC 50 hz. not 600 Watts

My water kettle has for sure more than 600 Watts. ... Is it high power?

Your hair dryer and water kettle work for limited amount of time. PC is meant to work for 24/7 very often - that creates some issues.

I do not know how many hair dryers I saw melted and destroyed bevause overheating whem pushed to work hard.

Simply, it is lot of watts, so it generates lot of heat, so lifespan becomes shorter... Bad choice.

And if it is unavoidable, then very strict QC, lare safety overheads and seriously overengineered connectors should be mandated.

Prima.VeraImagine using 24 or 48V with lower Amps, instead of that 12V thing... Just imagine.
They will totally cut the heat out of equation, but I guess you need a PHD in electrotechnics for that, that a 5 trillion company cannot afford.

Imagine how much more expensive it would be, how much more stress to MoFSETs would be, how much more tracing it would need...

Also, nVidia, AMD and Intel could forget about improving their efficiency, instead they could keep adding moar watts until we are back at 1 kW again... We do not want that...

Finally, backward compatibility would be gone... We can not move on from ATX standard for years, let alone switch to completely different architecture...

WirkoAnother case, certainly familiar to DIY electronics fans, are power transistors in audio amplifiers. Bipolar transistors need these resistors (balancing is one of their roles but not the only one). MOSFETS don't need them, they somehow self-balance (something something temperature coefficient, I forgot). Here's an extremely parallel example:

BJTs are negative temperature coefficient. Meaning their resistance lowers as they heat up. This makes the hottest transistor takes more than its share, which in turn makes it even hotter ... and take even more ... so it always ends up uneven.
MOSFETs are positive temperature coefficient. Meaning their resistance rises as they heat up. Here we get the opposite effect where the hotter parts take less of the load and thereby naturally balances.
Copper wire (all metals, including the connector contacts) is positive temperature coefficient, just like MOSFETs. Therefore adding resistors is pointless. The reason for adding any resistors is as a detector, not a balancer.

PS: The reason for the rather high value of 10 ohms for that audio amplifier is because that makes the transistors exceptionally good current control then. Very good for output linearity.

Oh noes! Another day, another RTX 5090 with melted connector due to faulty design user error.

And this is only a tip of a huge pyramid made of RTX GPUs that have:
- partially melted but are not causing black screens, thus users are unaware of the situation,
- melted but cases were not published or disclosed,
- melted but sold to other owner, as the price of the card multiplied (repairing connector is pretty cheap compared to getting new card).

Unfortunately, TPU does not cover all the published cases.

This connector and badly designed power delivery rail in RTX 4080/4090/5000 series is golden eggs laying duck for NorthridgeFix.

_roman_edit: Measuring on the GND pins is nonsense. Igor and other already claimed that the current does not flow back 100%.

Charge is conserved, as verified by igor

I am not sure what are you arguing about but I myself measured multiple times, that the ground wires on the 12+4 cable carry less current than the live wires, because part of the current flows into a PCIE ground and does not flow back into the PSU directly through the cable.

I measured it with direct current clamp meter measuring separatelly live and ground wires and the result was clear - it is not a coincidence that most of the time live connector part melts.

Some of the photos from the measurement are in my survey thread (not a comparison between the in and out current), in which I attempted to collect some useful information about melted cable incidents, but which unfortunately did not cause much traction.

BoggledBeagle in 12 pin GPU connector surveyI have a suspition, that not all cables and connectors are made with the care they deserve and there are some substandard and even dangerous cables out there. Some cables might have been damaged by us…

This is most interesting and a very valid point that the PCIe grounds may be carrying a lot of the return current.

BoggledBeagleI am not sure what are you arguing about but I myself measured multiple times, that the ground wires on the 12+4 cable carry less current than the live wires, because part of the current flows into a PCIE ground and does not flow back into the PSU directly through the cable.

Yup. That's why it's always the 12V pins that fail.

At this stage of things, since manufacturers are doing nothing serious, the best solution is cut the plugs off the cables and solder the cables at both ends.

Well, there is also supposed to be the Segotep TitanLoad but I've not yet seen anyone review it. www.techpowerup.com/343694/segotep-claims-titanload-12v-2x6-cable-lowers-temperatures