How Nvidia made the 12VHPWR connector even worse.

Der8auer's video: • 12VHPWR on RTX 5090 is...
Patreon: / buildzoid
Twitch(mainly gaming streams): / buildzoid
AHOC Shirts: actually-hardc...
Bandcamp: machineforscre...
My FPS game: buildzoid.itch...
Bluesky: bsky.app/profi...
Facebook: / actuallyhardcoreovercl...
#nvidia #rtx5090 #rtx4090

[@wulfgarpl]

Titanic designers: we don't need so many lifeboats. They take too much space.

[@TheDoomerBlox]

the more things change, the more they stay the same

[@kgjung2310]

Come on. It's a boat. What could possibly go wrong? Besides, the shipbuilder owner said even God couldn't sink it. Full speed ahead.

[@massivepileup]

At least the Titanic designers had the excuse that they met or exceeded the safety standards of the time. This is at a time when people already know better.

[@sirmonkey1985]

@@massivepileup on paper sure if it was actually built that way but they didn't because none of what they actually did met any standard. the titanic was the perfect example of corner cutting to save time/cost.

[@sylviarohge4204]

@@kgjung2310 The disaster has proven that icebergs are more powerful than God. We should start worshiping icebergs.

[@deimosian]

12 Volt High Failure Rate. Love it. Shit needs to be recalled.

[@haldbl]

"Release nothing recall nothing" nvidia

[@randomthot125]

Yes, all five of them!

[@einarcgulbrandsen7177]

Is this a legal way to do it in EU?

[@riba2233]

@@einarcgulbrandsen7177 it is obviously

[@Hriskataaa]

HA! Yea, good luck recalling all scalped 5090s Ah shit, imagine those guys that even agreed to pay up to 10k $ for one how they feel!

[@MyrKnof]

dude it just hit me. EVGA saw this coming from a mile away, and probably: 1. didnt want their reputation tarnished by burning down peoples houses 2. told nv, complained, and was told to stfu (would get sued for defamation or some shit) 3. signed deal/was bullied to not do amd gfx

[@dziadek_gruz]

They've also seen the shit nvidia pulls with the prototypes, could you imagine how you can compete with such 100% flowthrough card and have enough margin to actually sell it?

[@Spreadie]

Did you see his pinned comment on that vid - "Nvidia RTX50 with MFG - Multi Flame Generation" 🤣

[@juhavehmanen8604]

Yeah :D " Oh look the flames the graphics card is generating! That looks so realistic that it actually burns my hand when i touch it!" xD

[@born2rock4life]

If only they had an engineering team dedicated to testing such things.. /s

[@peterpain6625]

I bet they do but management wanted a 2 slot cooler. By any means it seems. As impressive the cooler is the power delivery SUCKS. Obviously.

[@Gadtkaz]

Malcolm will make a heatsink for the cable!

[@eltonmsn]

there is no time for not make money

[@MilenKitsune]

They do not have time to test them since they have deadline for realese. They also getting pressured from investors

[@peterromano1911]

If they HAD an engineering team....

[@EtaCarinaeSC]

I am here before another connector melts

[@hensiepensie]

you barely made it xD

[@EtaCarinaeSC]

@ thankfully the world supply of 5090 is so low, it gave me this unique opportunity.

[@MrSCOTTtheSCOT]

People are here watching while their connecters prepare for melting

[@Mewsashi-cz9fo]

at this point nvidia completely enjoys crapping on their customers, with hatred as a coating

[@geerstyresoil3136]

@@EtaCarinaeSC its low, because nvidia knows and was quiet about it

[@Dudi4PoLFr]

So, when Nvidia said that they fixed the 12VHPWR issues on the 50xx, they actually made it even worse... This is quite an achievement, actually. Genuinely speaking, how much would it cost to go back to the 3090 Ti's 3-phase design, at least for the $2000+++ 5090?

[@RS_Vaarkaiin]

For Nvidia? Too much. Blame the user, they will eat it anyways.

[@xynonners]

I wonder if they removed it for board space

[@marsovac]

it would cost 1$

[@zanfr123]

on the BOM (bill of materials)? a few cents. redesign side? I would say a couple days but need to push the changes to manufacturers for their PCB designs etc...

[@TheStigma]

We are talking about a handful of components maybe totaling up to a dollar in costs, or two at the most. For a card selling at 1500 USD this is just insulting...

[@adink6486]

You can tell a regular PCIE 8-pin connector (150w) uses more plastic and metal than the new NVIDIA 12-pin connector(600w). Guess NVIDIA found a way to counter physics.

[@kazuviking]

And that PCIE 8-pin starts melting at 386W meaning it have a f ton of safety margin built in.

[@SergeiSugaroverdoseShuykov]

Those connectors are designed by molex, assumingly those guys knows what they are doing and it's not connector fault that current is imbalanced to the state where one wire passes 30 amps while another barely hits 5

[@me1134]

Whats funny is the 600w 12pin connector official corsair cable plugs into 2 8pin on their power supply...

[@zanfr123]

all hardware manufacturers are now trying to defy physics tho; by faking it... fake pixels in photos on phones, fake pixels/frames on GPUs, overvoltage CPUs...

[@RobBCactive]

@@SergeiSugaroverdoseShuykov that's actually down to the dumb GPU design.. the current isn't managed, it's just hit & hope

[@aggies11]

Derbauer' Headroom comment is spot on. When you are pushing near 600W of power, you have a very small margin of error. Any small mistakes (user error, cable quality control, contact of pins to socket, oxidization and metal fatigue etc etc) can lead to catastrophic failure. Which means you need a lot of safety margin to make up for that. You have to plan for these things to fail and take that into account with counter measures. "Fool proof" and "fail proof" should be the goal. The current connector concept/design is simply not good enough. It will work if "everything goes exactly according to plan" which means little regard to safety.

[@heavyzakucustom4790]

seems like nvidia cheaped out on the power balance on their "$2000" flagship. things should be over engineered not minimally engineered

[@youkofoxy]

Yep, their safety margins are totally insane for a consumer product. Is more akin to Drag race, record breaking stuff or rockets. In that if works is great, if does not, well... it going up in flame is expected and part of the fun. Not exactly what someone expects from a product they invest hundred on to work without worries.

[@sgredsch]

buildzoid ranting about nvidias stupid 12vhpwr is exactly what i needed today. thanks buildzoid, much appreciated!! edit: totally agree with everything in the video, i just wanted to add: the old molex minifit jr (6+2 pin) not only has better safety margins, but the terminals and crimps are considerably bigger and beefier, it supports bigger gauge wire and the connector itself would be a lot more stable. the terminals inside the 12vhpwr are so thin, small and flimsy, they deform just by looking at it and the connector itself is so small and short that closing your side panel touching the cable will absolutely manipulate the seating of the pins inside the connector. it is so dumb. no matter from which angle you look at it, its 360° dumbness. maybe the dumbness even extends into a 4th dimension.

[@dozerroman4325]

Watched the DB video as well. When Nvidia cut costs to remove a larger profile power footprint, but didn’t compensate with bigger gauge copper, tell me… why was that not the topic of immediate discussion, years ago? This is 100% dangerous and it’s absurd for ANYONE to rely on a man made “electronic” sensor. The more components, the greater the propensity for failures. Hard pass on this generation, regardless, as the 50 series are nothing special. Also, Nvidia will be its own undoing with no competition. Bet.

[@kiwibom1]

Exactly what i thought while watching the video. They cut costs to then proceed to up the msrp's of all their cards across the board. There is no way Nvidia engineers thought that was a good idea. It can't be. And the fact that they didn't change it for 50 series is even worse. They just double down on it

[@halflife352011]

At this point I'm hoping intel will start competing. Never thought I'd say that

[@matocarina]

>it’s absurd for ANYONE to rely on a man made “electronic” sensor. I'm not sure I understand what you meant here

[@LampuhkapsChannel]

This whole release was never about good GPU's, its all about the money.

[@Dandan-tg6tj]

@@halflife352011 Intel should have started competing years ago.

[@TheBoothy666]

As a one time electronics engineer (moved into IT years ago), this is just madness! They literally took something that barely worked, and made it worse, much worse!! (Thanks for the content)

[@dkindig]

I'm a tech that slid sideways into IT a few decades ago but I have extensive prototyping/fixturing/build experience. The fact that there isn't at least 50% headroom on the connectors/cables specifications is ridiculous, aside from the design changes.

[@cracklingice]

All Nvidia needed to do in order to 'fix' the '6/8 pin connector issue' was to switch to EPS. Three EPS connectors would be friggin golden on a 5090.

[@Kent.V]

i have been asking myself this for years why can the cpu power connector can supply (almost) 400w while the pcie power connector is still limited to 150w

[@evilcakez]

Wouldn't even just 2 of them supply more than enough power?

[@xynonners]

you literally only need 2, not to mention the 350W PCIE H100s already use 1xEPS

[@Raxiel497]

@@Kent.V EPS12v has 4 12v wires for a start. I believe its also 18 gauge minimum and you can't have two connectors on one wire as you can with PCIE

[@HyperScorpio8688]

They're committed to the connector that they effectively designed, and they ain't returning to PCIE 12v or EPS. They know what they're doing

[@cipherbenchmarks]

This was a really good, educational video. Particularly the part about the shunt resistor diagrams between the different generation of cards! Thanks.

[@ericchecca3084]

I mean technically it’s a series resistor not a shunt resistor. Anything shunt is connected to ground or a small signal ground (like a supply)

[@TheOnlyToblin]

Nvidia's only option to compete seems to be "ADD MORE POWER! SCREW PHYSICS!" Edit: Finished the video finally and my only reaction is a flabberghasted "This is criminally stupid".

[@Osoronnophris]

its the motto of humanity itself

[@TimGBUK]

I wish hardware manufacturers would get around a table and hammer out an ATX standard that uses 48V so we can stop this sort of thing. And have smaller, more manageable and cheaper wires!

[@yugo_]

Something tells me Vince's team would have found this months before launch (for both cards). Sometimes, the added value for a team is not where you think it is. GPU manufacturers: please let overclockers/hackers stress-test/destroy your hardware before launch. It's a high-return investment.

[@L3_FR]

Luckily he is alluding to forming a new team heheheh

[@aggies11]

Nvidia can't not have known about this. The people responsible for designing the power delivery would most likely be painfully aware of this. Unless they let the interns do that at Nvidia now (or maybe their "AI"..), at least one, possible more people were aware of this and voiced their concerns. And likely got overrules by a manager that "knows best". That's the only way I can possibly explain how something like this makes it to market.

[@venomtailOG]

I wonder in how many days Nvidia will announce a recall

[@muzallisam5068]

never. nvidia is so high up on the bullshit throne they will never do a recall. maybe thats why there is no foudners edition card this time since they already know melting will be available to the consumers.

[@Simon_Denmark]

They’re not going to recall anything. Just look at the 40 series.

[@alu4626]

in never days

[@heyhoe168]

Never. Intel did not recall actual factory mistake, why would Nvidia recall merely unsafe design?

[@potato_melon7847]

They wont, they don't care, and if so, it cant be that hard to recall all 9 5090s they manufactured and sold xd

[@kaseyboles30]

What gets me is they add 4 pins to indicate 4 different power limits and that's all those pins do. They could do so much more. A simple thing like the can bus or uart or any one of a dozen lightweigth busses that could allow the psu to tell the gpu usefull things like "I can only deliver 425 watts on this connector and by the way pin3 is drawing 850 watts and might melt soon or burst into flames".

[@fral1073]

thats not something the psu would have to tell the gpu, the gpu should just measure that like some aib cards do. What you suggested also makes adapter cables impossible.

[@Teh-Penguin]

The undervolt on my 4090 is never going to be disabled.

[@brode7111]

Fr

[@herbertwalter8693]

it's so dumb it's brilliant

[@jaredrivera2619]

No! It's just dumb! (I got you bro)

[@Kapono5150]

And people are lining up to buy this fire 🔥 hazard ⚠️

[@igottheshaft]

Sad. I wanted to buy one (or two) so I could fine tune LLMs...but that can take days. I wouldn't feel safe leaving the house with these cards running now. Thanks for the video.

[@betag24cn]

lol, no, nvidoa sent worldwide like 5k units, paper launch

[@igottheshaft]

@@betag24cn Tell us that you have a room temperature IQ, without telling us that you have a room temperature IQ, please 😂

[@alexmills1329]

This is classic Ohm’s law at work here. Nvidia made the connector totally parallel via that termination, and the lowest resistance of the 6 wires will carry way out of spec load, until it hits thermal runaway and something melts to stop the connection. Edited because I am a dumbass when I’m medicated and say things wrong sometimes lmao.

[@alexmills1329]

And as an aside, going to 24v or 48vdc would solve all these issues, and with 48v one 8 pin would be able to power a 5090.

[@L3_FR]

Joule effect be like “my time to shine!”

[@Andras889]

Ohm's law works the other way around, the lowest resistance one will carry the most load, so if the contact is not perfect on some of the wires it all goes whack on the other wires. As the temprature rises it should increase the resistance, that's the only way that the wires are balanced. Also people might upgrade their GPU but not their PSU, and I remember from the 4090 release, that they had a spec the cable can be unpugged and plugged back like 10 times, as it weekened the connector.

[@ericchecca3084]

Sorry but thermal runaway sorta the wrong word lol. But also current wants to flow in lowest resistance path so it would be the opposite unless a wire is cut

[@zanfr123]

they forgot what "path of least resistance" means......

[@AshtonCoolman]

The 12VHPWR connector is cooked!

[@fral1073]

12VHFC - 12 Volt High Failure Connector

[@peterpain6625]

@@fral1073 This will be a complete clownshow once the cards are actually available...

[@nexos911]

litterally 🤣 Current monitoring is necessary! PSU or GPU side!

[@graved1gger]

Again!

[@Monsux]

12V-2x6 isn't the issue, the problem comes when the company doesn't use proper safety margins. I was hoping for more than a year ago that the 5090 would come with dual 12V-2x6 connectors. This would make sure there are proper risk margins for everything. This cable melting wouldn't ever happen if there were dual connectors on every 5090. Even my NZXT 1500C have native 2x 12V-2x6 cables, but I can't use two cables with new 5090. What a massive design failure from Nvidia.

[@MEMETIZER]

Oh, I can feel all those early adopters who spent a ton of money on a 5090 get really uncomfortable...

[@user-qq8ek4zs4s]

Watched the farmer video prior to yours and now everything makes sense , thank you BZ !

[@FromPlsNerf]

According to cablemod…12v2x6 was originally gpu connector only But has since been expanded to include cable and cable side connector

[@peterpain6625]

The hero we didn't expect but the one we need

[@OMGJL]

the underlaying reason (of nvidia became careless about current balancing) could very well be they somehow let go/redundant the engineer(s) who know what they are doing, or who really cared. I don't know if PCB design have any form of version control like programming, but if they do, check the "git blame" author of the shunt resistor near the power connector, I wouldn't be surprised that the author now don't work at Nvidia anymore.... in fact I would be surprised if that guy is still working in Nvidia.

[@erkinalp]

5090 FE PCB is probably partially based on GB100 PCB.

[@locinolacolino1302]

It may not be NVidia doing the laying off, alot of the Engineering talent is quitting as their NVidia stocks are worth enough they can comfortably retire.

[@aggies11]

@@locinolacolino1302 Yeah, brain drain maybe. But they definitely have enough bodies, they've been adding thousands of new employees each year (with all their recent "Success"). So could be the other direction actually, rapid growth that becomes unmanagable, leading to an erosion of corporate culture and perhaps internal disorganization and chaos. (It's very hard to grow employee count while still maintain the same culture , standards and commitment to quality. And that's if you take your time. Do it quickly and you can forget about it).

[@googletarded8300]

I'm so shocked this is happening again... That is to say I'm not shocked at all. I could see this happening with a 6090 if we still use these cables two-three years from now.

[@Raxiel497]

Enshitification continues. I know its glib, but 'Ngreedia' really does feel appropriate here. They had to have known.

[@aggies11]

It can't be just for cost savings. There is enough lapse to safety process here that one could make a case for negligence. That's a big liability you are open to. This smacks of someone having the idea that this is a more "elegant" and "simple" design. Can't have been an EE because this should give them nightmares. So some manager somewhere deciding that they know best, probably with a self congratulatory pat on the back. That'd be my guess.

[@farminglol]

I don't think they did. This hurts them too much. I know people love bashing them because they're rich and greedy, but this is bad for them and bad for business. No way this gets waved through if it's clear what the problem is. You don't design a cooler this good, and then fuck up this bad at the PCB level, because you're trying to save money. This was stupidity or at most ignorance, not a financial decision. Engineering fuckup, not management.

[@Raxiel497]

@ I should have been more clear in what I said. I think they probably believed they could 'get away with it' and that it would be 'OK', but they had to have known they were eating into safety margins that someone decided were extravagant. $2k-$3k for a gaming card is in no way 'budget' but compared to their data centre business, I'm sure there are some at Nvidia who consider the gaming market to be the 'poors' who deserve bare minimum for their 'cheap' products.

[@Deuxiit]

6 cables can hold max current 9,5amp. Here we are pushing over 20amps. We need solis copper bridges from psu to gpu and trow these crappy multiwire cables to trash. Electrons will go allways trough the lowest resistance and now we have cable bundle that can have different resistance if connector has even smallest impurity inside. I need to use electornic cleaner and apply copper paste to fix the different ohms on pins.

[@Jeffcrocodile]

i blame GN for saying this was all user error and clearing Nvidia. I hate influencers

[@betag24cn]

it is user error caused by nvidia doing a poorly desogned connector carrying top much amps and top much watts this is caused by nvidia, the user error

[@Jeffcrocodile]

@@betag24cn makes sense. I wonder how much he got paid

[@crnd11]

Copying what I posted on der8auer's video: Based on your current distribution and a 5W loss through the connectors, I estimate a contact resistance variation of between ~4 mOhm (23A conductor) and ~50 mOhm (2A conductor). This is assuming there is no potential difference across the 12V pins from the PSU. If the PSU has split two 12V rails across the connector that could also explain what you are observing. According to the specification, these connectors should have a maximum contact resistance of 5 mOhm and an endurance rating of 50 mating cycles. Clearly 50 mOhm falls far outside of this specification. I would be surprised if either you or the owner of the burnt cable have exceeded the 50 mating cycle rating so either the testing and validation for this was inadequete or both of your cables are 'out of spec'... I wonder if this problem would persist with a brand new cable - perhaps these are effectively single use items and more than a handful of insertions damages the contacts to the point of failure. Given the desire to avoid having to actively balance multiple conductors, one wonders why NVIDIA didnt move towards a design similar to high power battery connectors like the XT90 - it would've been more compact as well...

[@aggies11]

It'd be neat to see someone do the math, and a simulation of how it could happen. Start with the increased uneven resistance, current imbalance, temperature increases, more resistant and watch the cascading failures escalate through the cable. Neat, but also terrifying. It's shocking when you see something where it's obvious that there was "plans for everything to go right" but not plans for when "somethings go wrong".

[@riba2233]

Yep we will see more of this unfortunately

[@betag24cn]

on founders, i dont see how we wont see more molten units the hope here is custom designs will not do this same stupidity

[@riba2233]

@@betag24cn they are all the same :( except for asus astral

[@jesniels]

Most underrated video on the subject... Thumbs up from here!!! I really hope that all the good guessing people out there sees this. I will share it wherever possible! Very nice walkthrough....

[@tkgg]

I just hear the "dumb ways to die" theme while imagining melting connectors.

[@MandrakeFernflower]

10:40 😂

[@NANOTECHYT]

It's worse than I thought. A recall of all 10 cards they've shipped is needed immediately.

[@betag24cn]

a worldwide recall, yes

[@zephyrsimon]

Happy that the rich people are testing all of this out for the rest of us.

[@betag24cn]

nvidia? they clearly did not tested this

[@Tulah]

There was a very good comment (it was comment to Der8auers comment where Der8auer brought something up from comments of his German video about the same) on Der8auers video about resistance of the wires and if you keep disconnecting the piece between GPU and PSU resistance will increase and not at the same rate for each pin basically guaranteeing that enthusiast will eventually encounter situation where all of that power will potentially go only through one wire

[@imbamakaber]

This video explanation makes sense

[@Alvin853]

But why is the majority of the current running down 1 or 2 wires in the first place. Are the connectors that bad, that 4 or 5 out of 6 aren't making good contact? Are the cables not consistent enough? Roman has shown that all 6 wires were drawing current, so they weren't disconnected entirely, yet somehow the current balance was just completely off.

[@ActuallyHardcoreOverclocking]

the low current wires didn't have proper contact at either the GPU or PSU end.

[@Alvin853]

@ but I didn't hear a bunch of reports that people had trouble with the 3090 Ti plugging the connector in properly, back when the card was sensing each pair individually. What suddenly caused the same connector and cables to have that many bad contacts which the 3090 Ti didn't?

[@ActuallyHardcoreOverclocking]

@@Alvin853 the 3090Ti will brute force the current balance in 3 groups of 2 pins. The 4090 and 5090 just lets the current take the path of least resistance and if that path is just 1 wire then that wire ends up melting.

[@Alvin853]

@@ActuallyHardcoreOverclocking but that doesn't answer the question why it is so common to have the current only down one wire on 4090 and 5090, when it wasn't on the 3090 Ti

[@DCMamvcivmEvony]

​@@ActuallyHardcoreOverclocking Pedantic interjection from a sparky. Parralell paths of varying resistance, not the path of least resistance.

[@punktkomma9489]

Not using EPS connectors from the start was the real mistake

[@E9Project]

I appreciate your honesty about these connectors. Many other channels are not willing to place the blame where it should be.

[@fabiancojman4524]

The real reason the 3090 had 3 (or 2 depending on model) shunt resisitor is because the 12v power plane was not continuous. The phases for vcore (possibly vmem) where connected such that half used one 12v plane and the other half used the other. With the ASUS per pin monitoring it only makes a difference if the power planes are split such that an equal number of phases are connected to each plane, if they they are joined to form one 12V plane, then it would only be able to prevent melting, however the card would need to be placed in a "limp mode". The Der8auer video showed that if you take the lump sum model of the cable, those specific pins for what ever reason have a lower resistance (impedance is the more correct term) than the others and are therefore carring more current.

[@backinbusiness95]

DerBauer made a video where he checked a seconic 18awg 8pin and estimated that that cable can supply up to 230w max. An 16awg will be able to do even more. Here is the video kzbin.info/www/bejne/pmHJiGiJgaujnrc&pp=ygURVGhlIEJhdWVyIDEydmhwd3I%3D

[@aggies11]

The scary thing is, if he is measuring 120C or more on the outside plastic connector, what temperature must the actual wire inside be at? (The plastic is going to be a crappy thermal conductor and so it will take time for the heat to escape from the wire). Could be rather terrifying.

[@AsthmaQueen]

This is pretty concerning, this means the sense pins weren't an issue and aren't even a factor in this type of failure. Clearly there is some resistance on a few of the 12 volt pins in Romans case, and that's enough even with a very small amount of resistance increase on the pins from plugging and unplugging a few times to make the current that's going through the two wall connected pins that are low resistance very high

[@jamesalexw]

Damn lazy electrons, blowing up our GPU's cables...

[@L3_FR]

Urrgh those damn electrons trying to tell me about the ohms law and Joule effect. Like just “flow through” dude and stop bumping in your bretherin!

[@ppeez]

Its more than just a little resistance, because as temp rises, resistance does too and they still dont balance out

[@AsthmaQueen]

@ppeez very minimally, it certainly doesn't help it but current squared and resistance is ultimately what's occurring at the level of the terminal. So when you have multiple connectors, if you have a couple connectors total like The wire and all the pins are higher resistance and ones low resistance the low resistance one is going to have the highest current in the current distribution. Basic KCL. And so that's the situation where you end up with the most heat at a connector and or specifically at the PIN. Resistance does rise with heat however it's also a pretty small amount before we're talking about plastic melting percentage wise it's not that much. It certainly doesn't help though it's not going to make it thermally run away if anything it would help it balance out the power distribution. Because it gets in this situation to begin with because the other pins are higher resistance outside of the spec that they're supposed to be

[@AsthmaQueen]

@@L3_FR yeah this is a pretty good way to visualize KCL and ohms law

[@chapstickbomber]

<a href="#" class="seekto" data-time="680">11:20</a> "this is gonna catch fire" bro my sides

[@ElysaraCh]

Funny observation: based on the spec and headroom, using the pigtailed end of a typical 8pin to connect 2x 8pins on a GPU using only 1 cable is actually safer and more tolerant than using 12VHPWR on a 4090 or 5090. And thats something that i saw stressed over and over back with the 20 and 30 series. "Dont use the pigtail, use individual cables for each connector." Recently i had to run a 1080Ti FE with a single pigtailed connector out of necessity when my main PC was down. Worked fine for the week i had to use it. Lmao.

[@HansRamoray]

wonderful video thank you

[@igottheshaft]

Class action lawsuits please! This is a major fire hazard sold into consumer households. Unacceptable.

[@Flaszeczka1]

It's gam3rs fault, they dont like to plug 2/3/4 cables, they want 1 cable, possibly thin, so they can actually see their rainbow GPU and shiny ASUS logo for 2x MSRP😂

[@me1134]

The 600w corsair official cable plugs into 2x8pin on their power supply...all they had to do was make a 8pin 2.0 to the corsair specs and we could be using 2x8pin instead of this stupid cable.

[@CarlMartRod]

Yeah, I love Corsair approach. I have the RM1000E power supply (1000w) and a PNY 4090, using the included cable which is 2x8pin straight to 12VHPWR. Absolutely 0 issues so far

[@Jayram2K]

When I first saw board shots of the 5090FE and the way the connector was mounted I wondered if they just combined the rails into one. Sure enough, that's what they did, what's even more interesting is the pin that burned is the one closest to the GPU board, so I wonder if the current was all being shoved down that one rail because it was the path of least resistance. At this point the only hope is per rail current balancing on the PSU side, because nvidia doesn't seem to care about it anymore 😂

[@L1vv4n]

Yes, this is amusingly dumb decision. First, remove current balancing and than make connecting length different, so it will have different resistance. It not even accidental difference it solder quality or micrometeorites length. All that for 500+ watts. It is possible to cook food on less power.

[@fral1073]

before we had maximum of 3x of "default" in 1 cable or with the 3090, maximum of 2x in 1 cable, now with 4090 we got to 6x

[@willruss]

Sad world where a $2000+ ultra-premium product has failures that wouldn't be acceptable for a cheap toaster. I hope these companies get absolutely railroaded one day for selling a firehazard to consumers for years in a row

[@BobHannent]

I mentioned it on the Der8 video, but they really need to reconsider those flimsy connectors. They need to look at XT60 and Anderson connectors. Splitting the phases/load like the earlier cards looks interesting, perhaps the solution is per wire shunts but with ideal diodes to combine the power.

[@CatheteriZedEYE]

i agree with Derbauer, they should have put 2 12VHPWR sockets on the high end cards but i guess they wanted the card to "look" sleek 2X 12VHPWR with 3X 8pin on each connector so 6X 8pin to run the card, would be nice head room for the cables to share load

[@fral1073]

I don't get why it needs multiple cables and 12 connections. Just put in two connections and two cables. Then there'd be no need to have spacing for isolation of the different cables, and that opens up space for mechnical stuff ensuring a good connection. Or as others point out repeatedly, there are already working connectors in smaller form factor which are designed to deliver such an amount of power & amps...

[@betag24cn]

no, if all the load will go to one site instead pf being split between all 6 cables, it will not matter on the der8auer vodeo he is usong the converter from 12vo cable to 4 8 pin molex cables and still got the problem connector does not split load evenly over all wires and load will take the path of less resistance and you saw what will always happen, molten wires, connectors

[@RelakS__]

Time to make active 12VHPWR cables :D Active power monitoring and balancing on both side inside the connector. For $$$$$

[@Dyonivan]

@<a href="#" class="seekto" data-time="885">14:45</a> From what I understand, Nvidia goes through great legal efforts to lock down board partners from making adjustments to the card outside of Nvidia's specs. So ASUS probably wasn't allowed to change the load balancing, and I'm honestly surprised they added the shunt mods that they did. This is why EVGA got out of the GPU market right before the 40-series came out, because I'm almost certain EVGA said, "This card design is stupid and we won't be a part of it."

[@cppctek]

NVIDIA’s RTX 4090 and 5090 power adapters are using 14-18 AWG wire. Recent testing from Der8auer has shown up to 25 amps through the 12v wire. Test show it heating up to 150c on the psu side. This is a serious fire hazard as it exceeds the safe ampacity limits of 14 and 16 AWG wire, causing extreme overheating and even melting connectors. Key Issues: * 16 AWG wire is only rated for 13A - 25A is nearly double its safe limit. * 14 AWG wire is rated for 15A, making it unsafe for 25A loads due to overheating and fire risks. * Third-party adapters using 18 AWG wire are even more dangerous, as they have an even lower ampacity. * This issue is not user error-the current pinout and wire gauge are inadequate for the power draw. What Needs to Change: * NVIDIA must upgrade to thicker (lower gauge) wiring to prevent overheating. * A revised power pinout should be implemented to safely distribute current and reduce failure risks. * Ignoring this issue could lead to more melted connectors and potential fire hazards. This is a critical safety issue that NVIDIA must address immediately to protect users and hardware.

[@kazuviking]

Better option is just to use an XT90 or XT120 connector with 8-6 gauge wire. Small connector, can handle 750W continous and doesnt burn down.

[@kaseyboles30]

The max safe amps for a gauge of wire does go up for shorter wires. The limit you site is for tens of feet in household wiring. At less than 1 foot (.3 meters) It might be as high as 20-25 amps on a 14 gauge wire. Of course that's just the wire and assumeds a FIXED connection, not a socketed one like this. So TLDR they are trying to pull more than they likely should, but perhaps within what they theoretically can, so still a bad idea.

[@SergeiSugaroverdoseShuykov]

16AWG is commonly rated for 15A in case of silicone multi-wire, 14AWG has almost double diameter it can't be rated for additional 13% of current at least it has to be +50% for continues usage so if 16AWG has 13A, 14AWG of exact the same cable will give 19-20A

[@Monsux]

The 600W 12v cables are designed to use 16 AWG wires. Should they use thicker cables? Yes, but it won't fix the issue for all the millions of users. What would fix it, dual connectors on every 5090 cards. This would instantly add the proper safety margins. The average user doesn't have to know anything else than the cable type, and it's designed for 600W usage. The company designing a GPU should add the safety margins, so that it's always safe to use the product. With 8 pin cables, companies always added more slots when needed to raise those safety margins.

[@fral1073]

A german viewer of der8auer said he informed the german association for Electrical, Electronic & Information Technologies.

[@cldpt]

at <a href="#" class="seekto" data-time="570">9:30</a>, yes it is a theoretically larger overload than the 3x8pin, but for that scenario to happen on 3x8 you would need to cut 2 out of the 3wires balanced by one shunt. Which is a much less likely failure scenario. Edit: in the event of one wire cut, you get 75W per 16awg, which is less than 100w like in 12VHPW

[@opensourcedev22]

12 Volt High Failure Connector soon to be replaced by DLSS for electrical current. When enabled, electrons appear to be moving, but are more gentle on conductors. Only on Nvidia RTX 6000 series

[@makoveliprod]

Cable current capability upscale, new electrons generator

[@ElDiablo223]

Good video, very informative. Thanks man.

[@brianmiller1574]

Thank you buildzoid! This simple video, yet, I wouldn't think of it, and it is an eye opening. I'm skipping 50 series cards all together, even the low wattage upcoming cards. Nvidia needs to get their act together.

[@neiliewheeliebin]

This design regression is extremely bizarre it has to be intentional sabotage surely

[@auxityne]

"Pay us at least $2,000 for the honor of wondering if your card even works and the terror gacha of wondering if it'll burst into flames."

[@jannegrey]

Yay! I just commented on you previous video about that. NVIDIA at it's finest I guess /s

[@willruss]

it is almost as if a child revised their parent's product without knowledge of why it was built that way. i'd love to see the engineering staff that worked on these generations

[@mapscorp]

Hi Buildzoid, about the 3090ti don't having this issue, our Paulo Gomes gpu repair and mod guy did make a video about this at almost 1year ago, i think you would like to see it

[@progammler]

The cables are used in parallel but are connected separately. So even when assuming the resistances of the cables are equal (which could be wrong when a cable breaks), there could be uneven resistances across the connectors (bad connection, surface oxidation, etc.). The current will distribute over the cables inversly proportional to their individual resistances. This is just bad design from the beginning. Either use separate lines in parallel with load balancing or short the parallel lines together within the connector. But they just combined the negatives of both ways...

[@RadarLeon]

8pin pci-e 150w(288-16awg, 396-14awg, 432-12awg) rated for 50w per 12v with pin connectors that are 1/3 larger and 2x thicker, came in split and solid 12vhpwr 100w per wire anemic split pin head connector that will turn into a heating element, should not be loaded more than 386w based on the terminal connections.

[@WayStedYou]

<a href="#" class="seekto" data-time="640">10:40</a> It was at this moment, they failed to realize they F'd up

[@Server0750]

The problem is not the WATTS its the AMPS. 2000 Watt space heater on 240 volt only puts a load on the wire of 8.3 AMPS. But 600 Watt on 12 Volt puts a load of 50 AMPS on the wires. Using watts is kinda misleading.

[@eroticimp]

Watts is amps for a given voltage, so its basically the same thing. But yes its as they say, the amps kill you

[@p_serdiuk]

ATX standard specifies 3.3, 5 and 12 volts, so you can't have anything but 12 volts

[@Real28]

This is the kind of detail I've been looking for.

[@downwiththesneaker9933]

I called this when Steve at Gamers Nexus first did his testing wth 40 series. For some reason the GPU is pulling more power on individual pins and not equally between them all and this proves it. Not something wrong with the connector or the pins or user error but a issue with the power design of the connector and how it draws power from the PSU.

[@ilionsd]

On 4090/5090 connector wires are in parallel, which means they have same voltage, so the current will be balanced according to resistance of each wire + connectors 1/R. In ideal world all wires would get same current, but if connection to PSU or GPU is bad, it increases resistance and shifts current to other wires. As for heat, wires with good pin contact will get more current and generate more heat across the wire, while bad pin contact will generate more heat in connector, but less on wire. Overall, its the same old connector issue - the pin contact in uneven, which may depend on connectors quality or user error, but it doesn't mean that 12VHPWR standard is ok - it should incorporate safety measures and resiliency to perform as expected in real life scenarios

[@ploed]

Its like using a plug socket with 6 sockets, on every socket you plug a water kettle.

[@Dave-ct1jk]

Thank you for uploading this. I watched derbaurs video right after work when it was uploaded and I was like why are they not running balanced in parallel. So you've answered my begging of him lol Also, solution. Make your own cables using 8 gauge so even if they all but 1fails that single conductor can handle the 50 amps

[@fral1073]

The problem ain't the cable sizes alone. They don't change bad contact causing excessive heat in the sockets.

[@pkuutn]

No help. Pins themselves will have different resistance and thus different current and temperature

[@Dave-ct1jk]

@fral1073 your right, the socket will still be fucked, but the conductor will be fine 😉

[@LBXZero]

I have an Asrock Radeon RX 6900 XT OC Formula. It has 3 PCIe 8-pin connectors, and each one is as Buildzoid has up here. When I disabled the power limit setting using MPT, the card reported 560W ~ 570W power draw in Furmark, running close to thermal limit. It can definitely handle more power, but the GPU will never use it because this GPU is not stable above 2724 MHz without overvoltage techniques. PSU is an EVGA 1200W. Just to say, this card has the power delivery structure to be safe at RTX 5090 levels, but Nvidia's $2,000 card can't afford it for some reason. As for the 4 sense pins, 2 pins are used for Watt selection (Amps selection). 2 pins are reserved for future use. My assumption for those 2 pins is phasing into the PSU market higher voltage PSUs, and those 2 pins tell the GPU the voltage possible.

[@wanderingsandmusings9255]

this is straight up dangerous.

[@afre3398]

I know much of reason for the 12VHPWR, is to take up as little PCB real estate as possible for the connector. For NVIDIA board PCB real estate is commodity in short supply. For board partners traditionally they have created bigger boards

[@jaredrivera2619]

Let's see if we can get someone internally to leak the email where the engineers tell management that "this is going to result in overloading the connector" and then management saying they don't give a shit and do it anyway to save cost.

[@nexos911]

lovely! Thank you for explaining! EDIT: Food for thought. Can't the power supply be "smart" and do this part to load balance and monitor its 12V pins ?!

[@DahakaCL]

Yes, it's possible, but increasing a lot the price for the PSU, also being "smart" it's not enough, you will need to be monitoring "how smart" is the PSU, also saving logs somewhere in case there's something weird and is properly documented. All of this is more price and still doesn't fix the main issue which is Nvidia.

[@gaborkeresztes1739]

Why not give use the wall cable, like to one needed for the PSUs? It would be better.

[@ActuallyHardcoreOverclocking]

those are high voltage not high current(relatively speaking)

[@ppeez]

Are you serious? @gaborkeresztes1739

[@p_serdiuk]

That requires designing a 600W 12V PSU into the card itself.

[@willruss]

It really was genius, you can't have a class-action lawsuit if you never get cards into the hands of customers.

[@gambiting]

I posted this on the other video but I'll post it here too: I just checked my 5090FE with an infrared thermometer - after running FurMark for 10 minutes at full(575W load) the connector on the GPU side was around 45C, the connector on the PSU side was at around 32C. It's using Corsair RM1000x and its original bundled 12V-2x6 cable. I have no idea what's going on in this video, 140C on the connector side is insane. Mine is barely hot to the touch.

[@kazuviking]

I have a feeling its gonna be heavily psu dependent.

[@ActuallyHardcoreOverclocking]

Probably manufacturing tolerances. If connectors always made a perfect connection then manually current balancing them wouldn't be necessary.

[@kodato92]

My 4080 with the original PSU cables hit 80°c easy in that conector

[@kodato92]

Also the original melted in one pin and more 4 discolored I switched to 3rd party and it's working well

[@arminth]

Given all cables and connectors should be pretty similar in resistance, a big 12V blob would not explain only two wires going bezerk! There must be something on the pcb that creates significant different resistances!

[@Skrubmeister]

They should really split these currents through at least two cables, this is just insane lol.

[@detlaslegacy]

Given my past experience with Asus softeware, getting the user to install their special software already is a challenge, for both the consumers' trust and Windows' will.

[@BobDevV]

Planned obsolescence. Repair channels always said that anything other than 4090, 4080 is trash. Seeing this video makes me think that those boards probably have better safety measures than lower end cards from same series. The high failure rate of those cards is probably due to how power is delivered, it increases the chance of the power reaching the core, which is better protected on 4080 and 90.

[@grandtoasty66298]

so we need a cable with all 12v connected together at both plug ends, with a single 12V wire between them. That moves the problem back to the connecter being well seated on both ends.

[@AshtonCoolman]

I was hoping that the stock Nvidia 4xPCIe to 12VHPWR power cable combiner(splitter?) would help but maybe it doesn't. I'd like to see some thermal testing with it.

[@betag24cn]

that video is the de8auer one mentioned at the begining

[@bsv01]

<a href="#" class="seekto" data-time="747">12:27</a> maybe it's becouse those old cards don't break and still work, and new cards need to break earlier?

[@andromeda8418]

All this makes me want to buy 5090 and use the melting connector to make popcorn.

[@WayStedYou]

<a href="#" class="seekto" data-time="768">12:48</a> They went from caring about current to currency.

[@The_Paya]

In other words, 8/6 connectors before had their wires in parallel intended/used _more_ for "redundancy" rather than to "supply more power". 12VHFR goes for 0 redundancy and expects all wires to be perfect conductors with 0 resistance. Just fucking great. Now I'm wondering what's the "typical" area resistance of a nickel plated contact against another nickel plated contact versus a gold plated contact against a nickel plated contact (from the Molex article DB referenced). Materials aren't specified in the 12VHFR space, right?

[@_MintArcade]

Nvidia becoming Apple as Jensen intended

[@TheStigma]

No load balancing at all for 600watts seems completely insane. Borderline illegal in terms of electrical safety?... This is what you get on a 1500 USD card? Are you fucking serious?

[@steaksoldier]

Maybe its a good thing supply is almost nonexistent. I can already see the headlines "LA start-up creates new wildfire after 12 different nvidia gpus catch fire at once!"

[@PhantomBlank]

at this point we are gonna need to run 10 gauge 2-4wire cables to make sure these dont catch fire...