12VHPWR has a higher number of reported failures than 6/8 pin does despite being much younger

[harm9963]

New cable is beefy .

 

[Totally]

It would seem so... but I understand the beneficial idea - why have 3 or 4 connectors when they could be replaced with just 1....
As I've said before.... why go with a crappy set of small pins with super tight plastic which can easily break or if bent might compromise it's longevity when there were obviously more robust and simpler options... never mind... I can already see the appeal for a connector that is (almost) guaranteed to grenade itself at some point in terms of new card sales.

The pitch, the manufacturers get to pad their bottom line from the reduced overhead by not having to juggle multiple types of connectors, and the new connector will also be cheaper. They'll be raking in so much money from reduced costs that they won't mind Nvidia and/or whoever else collecting royalties in the background. Win - Win until the problems started.

 

[harm9963]

Pulling the cable out too many will affect the 12v rails over time, when new was a solid 12.045 under stress test.

New cable was free from FSP , thanks MR John Chen

Did stress test for new cable , back to new !

 

[Vayra86]

It really is not that hard to click it into place.. you just push until you hear the click, not until you hear your PCB snap. Maybe it is an issue with the cable itself? Maybe A-Brand PSU maker has a better cable than B brand?

Just grasping at straws here.

Not gonna lie, it is nicer to deal and route 1 cable instead of 2 or 3.

Excuse me, what?

NVIDIA RTX 5090 "Blackwell" Could Feature Two 16-pin Power Connectors

NVIDIA CEO Jensen Huang never misses an opportunity to remind us that Moore's Law is cooked, and that future generations of logic hardware will only get larger and hotter, or hungrier for power. NVIDIA's next generation "Blackwell" graphics architecture promises to bring certain...

www.techpowerup.com

They only need one generation to get you back on track, don't worry

Everything you said is just pure hyperpole that there is no point discussing

As if you are capable of discussion on this subject lol. You bought into this so its the greatest thing since sliced bread.

 

[londiste]

NVIDIA RTX 5090 "Blackwell" Could Feature Two 16-pin Power Connectors

NVIDIA CEO Jensen Huang never misses an opportunity to remind us that Moore's Law is cooked, and that future generations of logic hardware will only get larger and hotter, or hungrier for power. NVIDIA's next generation "Blackwell" graphics architecture promises to bring certain...

www.techpowerup.com

They only need one generation to get you back on track, don't worry

That was a completely unsubstantiated rumor.

 

[_roman_]

It would seem so... but I understand the beneficial idea - why have 3 or 4 connectors when they could be replaced with just 1....

The obvious.

If one cable is loose, the other cables can take the load because the cables and connectors are oversized.

I have one power supply unit with that 12V nvidia cable. that connector looks very small. I would not use it If I had the choice

I dislike the nonsense of those y cables where you have two gpu connectors in "daisy chain" / serial mode on one power supply cable. That should not be done!

--

This personal story should show that sometimes you get a bad product.

My Corsair RM750 PSU died after 18 months. I replaced it with an enermax power supply unit. During wiring the new power supply cables went loose. The CPU connector cables and the 20 or 24 pin ATX connector. I got new cables for free after an email. visually it's impossible to see a difference in the "not ok" and "ok" cables for the mainboard side connectors.

If you have a bad process during injection molding you get bad connectors.

I crosstested with the cables from the corsair psu and saw the retention force was much higher as with the "not ok = NOK" enermax cables.

If you get garbage which gets loose during wiring you are out of luck. Therefore it's better to have 3 connectors on a graphic card for a power supply unit.

 

[Vayra86]

That was a completely unsubstantiated rumor.

We shall see, but you are simply denying inevitability here, surely you realize that? If it isnt now itll happen next gen.

 

[londiste]

We shall see, but you are simply denying inevitability here, surely you realize that? If it isnt now itll happen next gen.

600+W into a GPU? That is A LOT, regardless of what the connectors used are. I honestly cannot see that happening with a consumer GPU, not this and not the next gen.

 

[Arctucas]

Since the older PCI-E cables and the new cables both use Molex 44476-**** contacts,

All about the various PC power supply cables and connectors

All about the various PC power supply cables and connectors

www.playtool.com

the ampacity is the same for each terminal.

 

[AusWolf]

The gold nugget of this video though is this chart illustrating the total number of people who experiences issues with 6/8 pin vs 12VHPWR:

View attachment 366533


I'm surprised GN didn't do more to illustrate why it's extremely bad that 12VHPWR already has a higher total number of failure instances. 6 / 8 pin has been around for 17 years while 12VHPWR has only been around for 1 GPU generation and yet the number of failures already exceeds that of 6/8-pin. If one were to extrapolate this data into a failure rate, 12VHPWR would be several magnitudes worse.

That's before you even consider that the question is narrowly enough worded so as to exclude 12V2X6 issues. Even if we assume that they reduce the failure rate of the connector by ensuring it has to be properly mated, the failure rate is still not nearly as good as the 6/8-pin due to the lower safety tolerances and the 25cm straight cable requirement of the specs.

IMO this kind of increase in failure rate due to a power connector is absolutely unacceptable. More than 17 years worth of failures now happen in a single year thanks to this new connector. IMO GN did not go nearly far enough in pushing from improvements or replacements to the connector.

I don't know if this has been said, but I can't be asked to read back 3 pages, so...

That diagram doesn't show the total number of people. If 10,000 people have used the old connector, but only 100 the new one, then 3.3% of 10,000 is 330, while 4% of 100 is 4. In that case, 82.5 times more people experienced a failure of the old connector than the new one.

What the diagram shows is that the new connector is slightly more prone to failure (not by much, though), which we already knew.

 

[Vayra86]

600+W into a GPU? That is A LOT, regardless of what the connectors used are. I honestly cannot see that happening with a consumer GPU, not this and not the next gen.

Better hope those node shrinks keep on giving then, otherwise, I don't see any other way. We already know Nvidia won't shy away from cinderblock sized GPUs that have a max draw of that wattage. The next x80 is supposedly already moving to 400W territory. Make it 350W with an undervolt and its still excessive, and a further escalation of TDPs per tier.

Place that in perspective; for decades we've had GPUs cap out at 250-300W every gen, on much larger nodes. We're shrinking, AND increasing TDPs now. The only GPUs doing more were duals, for a long long time. And if you follow that train of thought... its not at all a wild idea anymore to see generations further ahead simply placing several dies side by side, or sliced up in different ways on chiplets, suddenly you have more surface area to pull power through and 600W is suddenly just two measly 300W x70's linked together.

Blackwell for enterprise is testing the waters for us with that idea already. Its not far fetched to put two and two together here.

 

[ty_ger]

I don't know if this has been said, but I can't be asked to read back 3 pages, so...

That diagram doesn't show the total number of people. If 10,000 people have used the old connector, but only 100 the new one, then 3.3% of 10,000 is 330, while 4% of 100 is 4. In that case, 82.5 times more people experienced a failure of the old connector than the new one.

What the diagram shows is that the new connector is slightly more prone to failure (not by much, though), which we already knew.

It was brought up; almost immediately.
You are ignoring the age factor though. Failures are expected to occur more and more over time and one has had a lot of time than the other.

 

[AusWolf]

It was brought up; almost immediately.
You are ignoring the age factor though. Failures are expected to occur more and more over time and one has had a lot of time than the other.

That only matters if you're talking about numbers, not with percentages. If 3.3% of 8-pin and 4% of 12VHPWR connectors have failed in 2024, then one can expect the same in 2025.

 

[ty_ger]

That only matters if you're talking about numbers, not with percentages. If 3.3% of 8-pin and 4% of 12VHPWR connectors have failed in 2024, then one can expect the same in 2025.

Huh? You think percentages won't change year to year?
With 6/8 pin, I would expect that the failure percentage was near 0% the first year, slowly growing to near 3% on year 15. Similarly, I expect the 12vhpwr failure percentage to grow in the coming years as they wear out and oxidize.
The total number in the market grows year to year, but until saturation and maturity, there are always a greater percentage of old ones than new ones, and the old ones fail at higher rate each year.
2025 should have a higher failure percentage than 2024. Only way it wouldn't is if old ones are retired at the same rate that new ones replace them. It will take time for the market to be saturated and for new vs retired to reach stability. Expect a higher 12vhpwr failure percent in 2025.
The 12vhpwr connector has a much lower margin of safety, therefore its failure percentage will be higher and the failure percentage will continue to grow in the coming years. The fact that they are near parity at such a young age is not good.
If you look at it logically, you will see that the rate of new adoption trends downward year to year while the rate of failures trends upwards year to year. The failure percentage will continue to increase until it plateaus at some unknown value in an unknown future year.

 

[eidairaman1]

That only matters if you're talking about numbers, not with percentages. If 3.3% of 8-pin and 4% of 12VHPWR connectors have failed in 2024, then one can expect the same in 2025.

This is certainly keeping northridgefix busy, i shared a vid on page 2 about it

 

[AusWolf]

Huh? You think percentages won't change year to year?
With 6/8 pin, I would expect that the failure percentage was near 0% the first year, slowly growing to near 3% on year 15. Similarly, I expect the 12vhpwr failure percentage to grow in the coming years as they wear out and oxidize.
The total number in the market grows year to year, but until saturation and maturity, there are always a greater percentage of old ones than new ones, and the old ones fail at higher rate each year.
2025 should have a higher failure percentage than 2024. Only way it wouldn't is if old ones are retired at the same rate that new ones replace them. It will take time for the market to be saturated and for new vs retired to reach stability. Expect a higher 12vhpwr failure percent in 2025.
The 12vhpwr connector has a much lower margin of safety, therefore its failure percentage will be higher and the failure percentage will continue to grow in the coming years. The fact that they are near parity at such a young age is not good.
If you look at it logically, you will see that the rate of new adoption trends downward year to year while the rate of failures trends upwards year to year. The failure percentage will continue to increase until it plateaus at some unknown value in an unknown future year.

Yes, percentages can and do change, but I have no reason to expect any specific change in the future, unless I look at the trend of this change in past years. If I see an increase towards these 3.3 and 4% values, with the latter being a much shaper one (because the connector is newer), then yes, I can expect what you said. Otherwise, these are just numbers. It is also highly possible that these numbers were kind of the same in past years, in which case, I wouldn't expect much change in the future, either.

 

[R0H1T]

600+W into a GPU? That is A LOT, regardless of what the connectors used are. I honestly cannot see that happening with a consumer GPU, not this and not the next gen.

The only reason they're not happening today is because no one's really making money in crypto these days! But with the AI fad, it could be justified on "consumer" chips; we've already seen this in enterprise sector, so I wouldn't be so sure it's not coming

 

[AusWolf]

The only reason they're not happening today is because no one's really making money in crypto these days! But with the AI fad, it could be justified on "consumer" chips; we've already seen this in enterprise sector, so I wouldn't be so sure it's not coming

Except that AI has almost no use in the consumer space, unlike crypto.

 

[R0H1T]

You can run LLM locally on your cards, among other things.

 

[AusWolf]

You can run LLM locally on your cards, among other things.

You can, but 1. Who cares? 2. How much Tensor computing power does a home user need?

 

[R0H1T]

It's not about moar tensor power but the memory capacity! If they can price the 5090 at say $3k and get away with it, then everything else moves up that chart for Nvidia as well ~ it's just inevitable!

 

[AusWolf]

It's not about moar tensor power but the memory capacity! If they can price the 5090 at say $3k and get away with it, then everything else moves up that chart for Nvidia as well ~ it's just inevitable!

Pricing of the 5090 has nothing to do with anything we just talked about. It's a halo card, the so called "enthusiasts" will pay anything to get one. Pricing of the rest of the product stack will be way more interesting, Nvidia will have to be a lot more clever there.

 

[R0H1T]

I talked about why/how Nvidia can sell 600W GPU's not sure what you've refuted. And no once 5090 sells for $3k do you think Nvidia will continue selling 5080 at just a grand or even $1.5k assuming the former comes to pass of course?

 

[AusWolf]

I talked about why/how Nvidia can sell 600W GPU's not sure what you've refuted. And no once 5090 sells for $3k do you think Nvidia will continue selling 5080 at just a grand or even $1.5k assuming the former comes to pass of course?

There seems to be a huge gap in specs between the 5090 and 5080, so what Nvidia will do with the pricing or power consumption of one will have little effect on the other, imo.

 

[ty_ger]

Yes, percentages can and do change, but I have no reason to expect any specific change in the future, unless I look at the trend of this change in past years. If I see an increase towards these 3.3 and 4% values, with the latter being a much shaper one (because the connector is newer), then yes, I can expect what you said. Otherwise, these are just numbers. It is also highly possible that these numbers were kind of the same in past years, in which case, I wouldn't expect much change in the future, either.

First of all, this is a dubious poll so I really don't believe either number. That being the case, what you said is still illogical.
1) The factor of safety of the 12vhpwr connector is way lower than the 8pin connector. Logically the failure rate has to be similarly affected. It is illogical for the two to be so close to the same. The same use case for the same application: the one with a lower engineered safety factor will fail more often. We don't see that in these numbers yet, so we know we haven't reached the peak failure rate yet.
2) In the first year, the connectors which will fail in the second year are not yet represented. In the second year, the connectors which will fail in the third year are not yet represented. So on and so forth. In the first year, the new purchases are the highest. In the second year, there is less new market. In the third year, even less. Thustly, the failure rate trends up as the years progress, while the new adoption rate trends down. If X trends up over Y trending down, the percentage trends up. You are comparing a product that is 15 years old to a product that is ...2? 3? The product's specifications even specifically acknowledge that it wears out and fails over time. If you don't expect the percentage of failed 12vhpwr connectors to increase substantially in the coming years, I don't think you are living in reality. The 8pin failure rate certainly won't change much.

Look at this sequence:
year 1: 1/20
year 2: 2/30
year 3: 4/40
year 4: 7/48
year 5: 9/56
year 6: 8/62
That is the sort of growth in failure rate that we would expect to see as the connector saturates the market and the failure rate grows with age and then stabilizes. Adoption rate slows down while failure rate lags behind, grows faster, and then stabilizes.

Your statement was that 'because these are percentages, age makes no difference'. That is completely illogical to me. Please explain.