Updated 12V-2×6 Connector, Aka Formerly 12VHPWR, Survives 640W Stress Test Even When Partially Inserted

The problem is that folks have trouble fully and securely inserting it, as well as having it back out on them. This is what the new connector is aiming to solve.

This seems to be part of the issue, but does not explain all failures, the repair guy at northridge fixit had had cards in for repair where the connector was dully inserted and still melted.

And while news reporting about the issue seems to have died down somewhat there are still dozens of cards dying from this issue. I'm quit interested to find out what exactly is happening.

GPU prices these days suck but I'm kinda relieved the priced the 4090 out of my range or I would have bought one and possibly ended up with a lot of hassle.
 
This seems to be part of the issue, but does not explain all failures, the repair guy at northridge fixit had had cards in for repair where the connector was dully inserted and still melted.
To be very honest - I don't trust the users here. Given what we know about the connector in operation, I can see users honestly believing that they had done everything possible to ensure that the connector was fully seated, and yet still failed unaware. But perhaps there's another failure mode, or a version of the connector used by OEMs (PSU, GPU, or even custom cables) that produce a new failure mode somehow.

Well, we're getting a new connector that does everything that the original one was designed to do, and more - what surprises me the most is that Nvidia and their board partners didn't run across enough issues to consider it a problem, and neither did reviewers. It's a decent example of something that works in the lab that doesn't work well enough in the field.
 
These are my toughts on this new AI crap.
I'm still waiting for the 'killer app'.

Especially when pushed down to the consumer level - the models that everyone is gaga about today take up whole datacenters. Now, we've seen that level of technology filter down before over the decades, and there's obviously a chicken-and-egg scenario in getting it marketed, but still, about the only thing I've used 'AI' for has been for photo editing, specifically for noise reduction.

And for that, my aging (in terms of modern AI specs) RTX3080 screams right through.
 
Now that I finally seen this infamous connector in person, having bought a 4080, all I can say is WOW! What an actual piece of ****. No wonder it has problems. The HPWR -> 3x8pin adaptor is such a rigid cable made worse by the fact that it connects to the side of the video card, that no matter what you do, it will put pressure on the connector because it pushes at the side of the case. Even the length is the most ill advised possible. It is just the "right" length to put the 8 pin receivers right at the distance where the side of the computer case is. If it was just slightly shorter or longer, the connectors wouldn't be at the worst possible distance.

And when you connect it to the card, there is no audible click to know it is completely seated, I was waiting for some lever or retainer to fall in place, so I kept pushing harder and harder, but there is no such thing. You have to take it on faith that it is seated properly.

It's like they aimed to make this thing the worst it can possibly be. For what? To save 2 connectors worth of PCB space?
Finally someone says it as it is:

 
I'm a huge fan of Roman's videos and highly respect his expertise and opinions. I've posted quite a few stories on things he's done or reported on.

Interesting how his ultimate solution(s) point to redundancy. Either 2x of these types of connectors, which given his logic does make sense but opens doors to other problems with all of the "fun" folks who will try to feed max power through both, or 4x 8-pin which is where we were heading. Granted, he explains how 3x 8-pins is more than enough, for now, and still within a reasonable safety margin, but in either case having 3 or 4 8-pins is getting pretty crazy. I've had a number of cards with 3x 8-pin and cable management becomes an increasingly pia issue even though with 3 of them you were already committed to having to use 2 cables off the PSU and the 4th 8-pin would've been a part of that anyway. I also wonder why we don't have a 24-pin connector similar to motherboards for GPUs now and granted, that kind of goes back to what he's suggesting in having two of these types of connectors.

That brings up the next issue which is the space to install cards that need any of these connectors. It's a reason I've often gone with bigger cases and I'm so glad that all 3 (Carbide, H7 Flow, HAF X) of the ones I have now work out well with the 12VHPWR. I'm now well over a year with my 3090 Ti's and each has been plugged/unplugged and moved around multiple times. Each also usually pulls 385-425W under load. Meanwhile my 4090 runs mostly with a light OC on top of the factory OC which ends up being close to the same wattage but I've seen some instances of it going to 485-515W during heavier OC testing and rare gaming moments. About once every six months or so I've removed its cable just to inspect, and I'm well aware there's a rate limit for that as well. So far, so good, and this card is used the most as it's in my daily runner but the CRG9 3090Ti Hybrid rig gets a lot of use as well and I've got a mild OC on top of the factory that keeps it close to its max clock speeds and power draw. I've occasionally even seen it hit upward of 450W draw.

I think that no matter what we are ways from seeing the end of this since it's clear that the top-tier cards are as power-hungry as ever, and this is true for both AMD and NVIDIA. We've seen the stories of what happens with extreme overclocking for the 7900XTX and it ends up looking like a sibling to the 4090 in that respect. I also agree with Roman that it doesn't make sense to use the 12VHPWR on cards not needing the increased power of the 4090 but we all know how business works in keeping costs down. More volume usually means bigger discounts hence why NV is pushing this on pretty much all its cards.

I do think one solution is that all GPU manufacturers need to stop putting the power connector on any of the edge sides and switch to either the backplate or fan sides. The bottom line is that most cards are now too big to accommodate extra cabling there anymore but those sides are essentially unused real estate where there's plenty of room for safe options.
 
I think that no matter what we are ways from seeing the end of this since it's clear that the top-tier cards are as power-hungry as ever, and this is true for both AMD and NVIDIA. We've seen the stories of what happens with extreme overclocking for the 7900XTX and it ends up looking like a sibling to the 4090 in that respect. I also agree with Roman that it doesn't make sense to use the 12VHPWR on cards not needing the increased power of the 4090 but we all know how business works in keeping costs down. More volume usually means bigger discounts hence why NV is pushing this on pretty much all its cards.

I do think one solution is that all GPU manufacturers need to stop putting the power connector on any of the edge sides and switch to either the backplate or fan sides. The bottom line is that most cards are now too big to accommodate extra cabling there anymore but those sides are essentially unused real estate where there's plenty of room for safe options.
I agree with you here. Part of my decision to down grade to a 4080 was that my 4090's were just getting ridiculous on power draw. I know there are methods to help alleviate this, but honestly in the over twenty-five years I've been into computers I have never had to worry about any hardware being a hazard. I honestly was defending these cards at the beginning and saying most of it was user error (which I still believe has some part in it). The power draw, plus the less than stellar connector shows how little things are tested before going out to the consumer. I never had an issue with my 4090's, but it seems like something may happen at some point regardless how careful you are based on recent findings. If and when I do upgrade my system next time I will be more careful in what I choose.

Good watch
 
Last edited:
I do think one solution is that all GPU manufacturers need to stop putting the power connector on any of the edge sides and switch to either the backplate or fan sides. The bottom line is that most cards are now too big to accommodate extra cabling there anymore but those sides are essentially unused real estate where there's plenty of room for safe options.
Realistically they can't put it anywhere else. The fanside would be a problem if you have any other expansion cards, the backlplate is a problem if you have air cooling on the CPU. In one of my previous systems. I had to put spacers between my graphics card and cpu cooler so it doesn't short out, because they were touching. And the end is also no bueno as you say due to the length of the gpus.

What could work is instead of having a PCB connector, they should have a short extension cord with a wire to wire connector at the end of that. That would reduce the stress on the connector itself.
 
A better plug, thicker wires, better specs. Is not that hard. A good 4 wire plug with thicker gauge wires would be a better choice. Why 4 data cables while my interner comes via one is a mystery to me but, well whatever number of data cables is fine.
The hpwr plug could seat in the middle of the pcb, doesn't have to be right at the edge, just plan accordingly, plenty of flat components and things allowing a clearance to do so. Its just pure laziness. This plug is a triumph, a triumph of demanding aesthetics over traditional safety choices, and of the 'yes men' culture.
The one guy that said no was probably fired.
 
Sure doesn't seem that hard. I mean:


$10 -- retail! would be dirt cheap for a AIB to purchase in bulk. Standard connector - they get used in UPSes and inverters and the like all the time. Would fit on any 2-slot GPU without too much issue. Supports up to 80A - which at 12V is almost 1,000W. And hey, special bonus, this particular one is "flame retardant", which nVidia could use a bit of right now.

It is bigger than the 12VHPWR - but hey, this one won't melt. It's still a good deal smaller than 3x8 pin would be.
 
The molex ones cost probably a few cents for them in bulk. These high quality robust connectors are rarely if ever used in consumer grade high volume hardware. That $10, or even $5 with volume discount adds up to millions very quickly.
 
The molex ones cost probably a few cents for them in bulk. These high quality robust connectors are rarely if ever used in consumer grade high volume hardware. That $10, or even $5 with volume discount adds up to millions very quickly.
Even a Molex is about $3 retail. While pennies count when mass producing, when you are starting at 70%+ margins, you have room to make quality improvements. It's not like you need a high-amperage connector on a budget level GPU anyway - these things are only needed on the highest tier cards commanding a premium price and the highest margins.

Does it need to be this specific connector? No. My point was that there are existing connectors, of proven reliability, of high quality - that have been available for years, and at reasonable costs. Yet we go about reinventing the wheel for ... reasons (vendor lock-in? Licensing revenue? Marketing?) and it's to no one's benefit.

A rebuttal about "but it would add to much cost" to a $1000+ premium product wtih extraordinarily high margins is, quite frankly, not very well thought out.
 
Even a Molex is about $3 retail. While pennies count when mass producing, when you are starting at 70%+ margins, you have room to make quality improvements. It's not like you need a high-amperage connector on a budget level GPU anyway

Might as well add those to more budget oriented cards aswel as the volume needed would drive down the cost. Ofc not on cards that don't need additional power
 
Even a Molex is about $3 retail. While pennies count when mass producing, when you are starting at 70%+ margins, you have room to make quality improvements. It's not like you need a high-amperage connector on a budget level GPU anyway - these things are only needed on the highest tier cards commanding a premium price and the highest margins.
I often buy electronics components for work and hobby projects, these things literally cost cents if you go to the suppliers and not through amazon or some other high margin marketplace.
Does it need to be this specific connector? No. My point was that there are existing connectors, of proven reliability, of high quality - that have been available for years, and at reasonable costs. Yet we go about reinventing the wheel for ... reasons (vendor lock-in? Licensing revenue? Marketing?) and it's to no one's benefit.
I know my connectors, there is a huge leap in quality between the cheap ones and these.
A rebuttal about "but it would add to much cost" to a $1000+ premium product wtih extraordinarily high margins is, quite frankly, not very well thought out.
No need to get angry, but the just use a better connector argument is actually what is not well thought out. High margins for flagship products only applies to nvidia, they sell the 4090 chip at a high margin, but the AIBs barely make a profit, this is why EVGA quit the business recently. And regardless of the margins companies are greedy, if they can use a connector that costs them 30 cents, they'll not upgrade you to one costing even just $3. I'm not saying it is right, but it is how it is.
 
I often buy electronics components for work and hobby projects, these things literally cost cents if you go to the suppliers and not through amazon or some other high margin marketplace.
Yeah - that was exactly my point. A $10 connector for me isn't going to cost $10 to an AIB.

And you are just shifting your claim from poor nVIdia to the poor AIBs -- same exact argument though, still poorly thought through.
 
Yep, so far, knock on wood, so good on my end too and at this point I've easily got many hundreds of hours on the 4090, not to mention the time I had with my 3090 Ti's, and now 4080 SUPER. However, I'm not discounting what others have encountered.
 
Become a Patron!
Back
Top