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

[STSMiner]

Yeah, it's crazy to say the least.

As for the RTX 5090 melting the connector, what I want to know is if these pins in the connector are of the four contact design (NTK), can't see too well inside these pins to say one way or the other here.

 

[Gatt]

4080? For real?

Sure, the V1 CableMod was lightly damaged while the V2 connector literally melted. And the card was running at default settings since, being a basic PNY model, it could not be overclocked. Anyway, I never solved the the 4080 black screen (with fan at 100%) random problem, so I got back to the 4070 Ti Super since I play 1440p and it is enough @ ultra settings. Never had a card or driver problem since then.

 

[R-T-B]

This thread has gone from wire safety margin, to shunt resistor, cooling down the actual cable with a fan then latest how to bend the cable properly. We all need to go on EWIS (Electrical Wiring Interconnect System) course before using this connector

you won't learn much here if that's your objective I fear.

To clarify, there may well be an issue of note. But expecting this mob to understand it is sort of like expecting a monkey at a typewriter to write shakespear. It COULD happen but I highly doubt the monkeys literary prowess (as well as our collective electrical engineering knowledge).

 

[wolf]

this isn't reddit you don't come here with a personal vendetta and allow it to color your opinion we don't do that here period

I support this statement 100% and would love it if everyone could get that memo, but unfortunately there are people here who do this constantly and are spurred on by likeminded people that agree (by virtue of positive reactions).

I'd happily tag a few right now, but I genuinely don't want to incite arguments and completely derail this thread. Boy oh boy do they need to hear it though.

 

[Apocalypsee]

you won't learn much here if that's your objective I fear.

To clarify, there may well be an issue of note. But expecting this mob to understand it is sort of like expecting a monkey at a typewriter to write shakespear. It COULD happen but I highly doubt the monkeys literary prowess (as well as our collective electrical engineering knowledge).

I do learn something, avoid this shitshow till they really fix the problem.

 

[R-T-B]

I do learn something, avoid this shitshow till they really fix the problem.

None of us really have any idea how widespread it is at this point in time, but that's certainly a safe option.

 

[Werwulf]

Those 12V-2x6 shorter sense pins working well I see

 

[Caring1]

Yeah, it's crazy to say the least.

As for the RTX 5090 melting the connector, what I want to know is if these pins in the connector are of the four contact design (NTK), can't see too well inside these pins to say one way or the other here.

View attachment 386369

Two side contacts, not four.

 

[Werwulf]

I think all the talk about pins, cables and failure rates of 5090s is kinda missing the point. The design is bad. We know it's bad. We don't need to wait and see if it's bad. There is no fixing it by finding some magic cable with quadruple dimple pins. If things melt when the cable falls out of spec... it's bad. It's like waiting to see how many people splat into the steering wheel of a car that the manufacturer didn't put an airbag in to determine the extent of the issue.

 

[N/A]

J2c unplugged the cable 100s and of times and it wasn't the dimples, and if anything the distribution only got better, and it's only when he deliberately widened one dual split dual bumps terminal that it went down to 1A. The cable needs some sort of reverse thermostats for the sense pins in the sense that it cuts the grounding when it >60C. And possibly the FE has some sort of weird issue where one pin has so little resistance that it tries to load it with 25A because the connector is vertical or something. And it could really use some per pin load distribution.

 

[Bobaganoosh]

Gotcha. You don't have his answer and his answer isn't in those posts. Why tell him to pound sand? Just for fun?

Maybe you should use your words better.

I was ignoring the previous posts as I hoped you two would figure it out, but they don't seem to want to explain themselves to you so I'll just tell you how I read it...someone claimed "millions and millions" and I asked where they got that number. Pretty sure Degreco was suggesting they got that the original "millions" number from nowhere in particular and that person just wanted to participate. I had taken it as sarcasm, just not directed at me, but the person I asked the question to. let's be done with that tangent lol

I would love to know how many 4090 cards have sold though. I know a huge portion of the repairs were from CableMods adaptors, but there were hundreds of them. So if sales were in the hundreds of thousands and we had hundreds of failures, I figure that puts us at somewhere between 0.1-1% failure rate. Sure, a huge chunk of that is from adaptors one company made, but there are also several failures from other vendors/cables. If a "450W" card (which often games at closer to 375W) running on a cable spec'd for 600W gives us 0.1% failure rate, what will it be on a "600W" card that runs close to that running on cables rated 600-675W? Higher is the only possible answer, but I'm curious what the actual percentages are as these are entirely guesswork without real data.

 

[chrcoluk]

I was almost frying my 4080 with the damned CableMod angled connectors. Both V1 and V2 versions melted. Be VERY careful.

Right angled adaptor or right angled cable? its important to make this distinction as so often people are confused over it. The adaptor had the recall with some confirmed issues.

 

[lilhasselhoffer]

Where are you getting that Nvidia sold "millions and millions" of 4090s? The number is probably a few hundred thousand globally. The issue with this connection is only when pushing it hard, for the most part, so it's mostly even just the people running their 4090 over 450W, which is likely not even a huge percentage of those owners. It seems as though they've only sold about 100 5090s so far, so we'll see much higher percentage of fallout there. The issue is more likely to occur than most people would find acceptable due to very low safety margins and almost no precautions taken on the design of the cables or cards, but that doesn't mean every single cable/card is a ticking time bomb. It's not always "all or nothing" lol.

Let's do some basic math, and have some fun answering the question. I say this, because about 20 pages ago I proposed a 1% failure rate selling about 30 millions GPUs a year would represent a preventable homicide every few years...but that was just big round math to create a tangible story as to why "a couple percent" failure rates was not something to overlook.

Nvidia's 2023 and 2024 production of consumer GPUs is pathetic. 33.6% and 17.1% kind of shows us why the rarity of these GPUs is high...no? Note that my source on this is here: Nvidia GPU breakdown by year
2023 represented 3.76 million datacenter GPUs...so 55.6% = 3.76 million -> 100% = 6,762,590 ~= 6.76 million GPUs.
Let's assume that the 2023 number is accurate, and they sold 33.6% as consumer GPUs. That's 2.27 million GPUs. Let's pretend that this is doubled, to represent the lifetime of the 4090, given the late 2022 introduction and early 2025 phase out. That leaves us with a consumer GPU pool of 4.54 million GPUs.

Now...how to get the 4090s separated from the pool? It's not ideal...but I'm going to cheap out here and use UserBenchmark. It's going to be weighted to people who benchmark and want to show off their expensive toys...so it's a little weighted to the kind of people who would buy these expensive monsters. That shows a maybe 1.5% maximum...so let's go with that. 0.015*4.54 = 68,100. So...under 70k units sold worldwide. That's across the entire two years...so if just 700 people on the planet ran into issue with these things it'd be a 1% failure rate.


Now...how many 5090s are there? I'd hazard to say less than a few thousand given the absolute madness of launch. With that kind of a shallow pool it takes almost no failures to rather rapidly reach 1%. It's especially galling whenever you have such a small pool that 100% sampling for your $2000 card absolutely would make financial sense...but the greed with these people is so great that every scalper in the chain gets to pad their pockets and still won't help the consumers. That's...well, that's what you get for writing a bad interconnection specification and I'm glad Nvidia is getting kicked in the teeth for this. It's just a shame that AMD and Intel are both sitting around with their thumbs up their respective rears, making this the world's lamest one horse race.

 

[Degreco]

GN uploaded a new video, basically on RTX 50 being the worst GPU launch in recent history with the sum of the different issues. He calls it MFG as in "Multi-failure-generation".
Very entertaining. Also a little bit saddening.

Steve did concentrate on the latest issue with the missing ROPs, but at one point at the start he eventually says:

Including Nvidia flying way too close to the sun with 12V-2x6

and includes some stills about heated connectors and cables. As well as this one with an interesting reading:

How is this possible? Clamp defect? - Wondering what the other pins were like with this wire being negative 7.9A
I guess, this particular wire is very cold

Does somebody know the source of this screenshot?

 

[Dredi]

How is this possible?

DC current is directional. Flip the meter to flip the polarity.

 

[Werwulf]

One user showing problems on a 5090 Astral. It's so ridiculous to see your GPU be like "Hey, look at me, I'm running way out of spec, gonna burn your cables sucker, better be quick!"

Another one, this time with a brand new cable. I feel Asus are going to regret putting this tool in their software. I mean, if you're going to do something dumb like pulling 600w over 12VHPWR, then maybe don't draw attention to it

 

[Degreco]

Another one, this time with a brand new cable.

This is ridiculous. Up to this point I thought one could at least mitigate the risk of amp imbalance with brand new cable(s).

They should probably bundle all those GPUs with a current clamp if it's not the Astral /Matrix so buyers can sleep better.

 

[londiste]

One user showing problems on a 5090 Astral. It's so ridiculous to see your GPU be like "Hey, look at me, I'm running way out of spec, gonna burn your cables sucker, better be quick!"

View attachment 386646

Spec is 9.3A. 9.7A is about 5% over spec.
It is a problem, just the magnitude isn't exactly "way out of spec".

 

[Werwulf]

Spec is 9.3A. 9.7A is about 5% over spec.
It is a problem, just the magnitude isn't exactly "way out of spec".

That top user, I think in their other pics the highest was in the 11s. My point was more that it could be 25A and the GPU is going to show you but isn't going to do diddly squat about it.

 

[N/A]

Eventually they'll get around to doing more. Now it's per spoon per hour and unfinished products for exorbitant prices. But it's better to have this reporting and it could tell you that reseating the cable is in order, just don't go too crazy about it.

 

[Degreco]

Spec is 9.3A. 9.7A is about 5% over spec.
It is a problem, just the magnitude isn't exactly "way out of spec".

I see your point.

3-5% over spec, 9.7A on a wire... sure, nothing burns. Probably. No big deal.

But honestly, it doesn’t matter if it’s a tiny bump over spec or a massive one - spec limits exist for a reason or else we don't need them.

Brand-new hardware, stock settings, $2,000, $3,000, $4,000 price tag - this shouldn’t be happening. It's not acceptable!

If a $550 mid-tier GPU from AMD or Intel had these issues, we’d riot! So why does Nvidia get a pass? They won’t even deny there’s a problem. No "This is overblown" - just total radio silence. Gaslighting at it's finest.

Yeah, they dominate the high-end market. But does that mean we just accept this? I’d love to hear anything from them - like, “Relax, it’s just a few bad cables,” or “0.2A over spec? You're totally fine.” Instead, they’re just ignoring it and hoping we do too.

At this point, does anyone really believe they care about their gamer community anymore? The very community that kept their business alive all those years from 1999 until AI came along?

This is all you need to understand about where Nvidia’s priorities really are right now:

 

[ShrimpBrime]

GN uploaded a new video, basically on RTX 50 being the worst GPU launch in recent history with the sum of the different issues. He calls it MFG as in "Multi-failure-generation".
Very entertaining. Also a little bit saddening.

Steve did concentrate on the latest issue with the missing ROPs, but at one point at the start he eventually says:

and includes some stills about heated connectors and cables. As well as this one with an interesting reading:

View attachment 386578

How is this possible? Clamp defect? - Wondering what the other pins were like with this wire being negative 7.9A
I guess, this particular wire is very cold

Does somebody know the source of this screenshot?

He must not had been old enough to remember the 8800 Ultra at launch. Was a horribly hot GPU, high prices, big power consumption and heat. And if memory serves today, even a couple melted 8 pin connectors.

 

[Werwulf]

I'm not really surprised Asus didn't implement a more robust safety mechanism for launch. As mentioned before, they only have a handful of days to do any actual testing once they get chips from Nvidia. Would Nvidia even allow AIBs to implement some kind of safety feature? You'd think it would simply be a case of dropping the power limit when the card detects an imbalance but how then do you notify the user their GPU is in 'limp mode'? A flashing LED? What if they don't see it? And what amperage would you trigger at? These cards run so close to the wire and I wouldn't be surprised if they exceed the 9.5A spec fairly regularly.

Some people say that you only need to do a quick check if you change or move the cable but even that isn't true. The user from one of those screenshots said he didn't move a thing, cable just went out of spec. What is the official advice from Asus, re-seat the cable and if that doesn't work replace it?? Maybe there are a whole bunch of 5090s sucking down 600w across just 3 cables, humming away, user completely oblivious.

 

[JAS_design]

View attachment 386588

One user showing problems on a 5090 Astral. It's so ridiculous to see your GPU be like "Hey, look at me, I'm running way out of spec, gonna burn your cables sucker, better be quick!"

View attachment 386646

Another one, this time with a brand new cable. I feel Asus are going to regret putting this tool in their software. I mean, if you're going to do something dumb like pulling 600w over 12VHPWR, then maybe don't draw attention to it

It's curious that in both images, pins 5 and 6 have lower readings. Maybe something to do with the orientation of the GPU, or how the cable is putting pressure on the connector, affects which pins have good contact and which don't.

 

[freeagent]

Those 12V2X6 cables with only 2 small wires are the only ones that I have seen burnt.

Mine has the full 4 wires. Have any of those gone up?