PC Enthusiast's Next Stop is... 12VHPWR Power Connector with Active Fan Cooling?

[eidairaman1]

The linked article from 2023, demonstrated a 12V-2x6 having no problems running above max rating while at the same time not fully mated. Here it is again - https://www.techpowerup.com/314066/...-to-handle-full-load-while-partially-inserted
I'm struggling to see why that isn't being paid attention to.

It happened over time is why. Point is nv aint going to do shit about it until a regulatory function gets involved and hits them with a harsh fine

 

[user556]

What happened over time?

So even the dubious 12VHPWR is fine if all 12 pins are making contact and all cables are physically undamaged. At 100W per pair that's 8.3A per pin and the pins are rated to 9.5A. Even if the connector is not fully mated, there should at least be contact.

Should is the operative word. There is at least some 12VHPWR flawed plugs that don't. Those need recalled. If they are not identifiable then a general recall of all 12VHPWR is in order. It's that simple.

Where connectors are melting and thermal imaging of hot cables (der8auer, for example) is happening, that's because both the PSU and Load are 'dumb' designs that aren't balanced, and you can get the vicious circle I mentioned in my last past. All it takes to get a wire too hot is some invisible damage to the internal wire strands under the sheathing, or a bad crimp job on one of the Amphenol connector pins in the cable's plug. Once you have a hot wire with higher resistance, the problem is self-feeding because it's a positive feedback loop that requires shunt resistors and a monitoring to counter, not a single dumb rail with zero monitoring.

Nice bonus, but it doesn't make a working plug. The cables still need replaced to fix the problem plugs. Best way to do that is give everyone the newer 12V-2x6 upgrade.

 

[Chrispy_]

The cables still need replaced to fix the problem plugs. Best way to do that is give everyone the newer 12V-2x6 upgrade.

But the newer 12V-6X2 doesn't fix anything, and it's a connector upgrade on the GPU end, the cables are still the same as 12VHPWR.

It's literally just the exact same dumb design with ever so slightly longer pins, so it might prevent a small fraction of problems but fundamentally there's still no anti-melt protection on the GPUs without multiple shunt resistors to monitor and balance the current.

 

[user556]

That's a myth. No one seems to want to admit a recall is needed.
Moddiy are all but admitting it here - https://www.techpowerup.com/forums/...ables-for-geforce-rtx-50-series-cards.332507/
If the cable plug has a H++ marking then you should be all good.

Yes, I am speculating. But no one has proved me wrong either. I haven't yet seen a H++ marked cable get hot.

If a full 12V-2x6 (H++ on the cable plug) example of a hot running (within spec) connector can be demonstrated. I'll happily change my mind and admit the connector design is simply too weak to support 600 W.

PS: If you read between the lines in Moddiy's statements. They are saying that the spec doesn't cover female pin designs inside the plugs. It's just whatever fits the male pins. So, an upgraded spring contact can happen without a change to the spec.

Which also means it could be applied to the older plugs too. But who'd do that when you can just de-list them and offer the newer model instead.

 

[Vayra86]

The linked article from 2023, demonstrated a 12V-2x6 having no problems running above max rating while at the same time not fully mated. Here it is again - https://www.techpowerup.com/314066/...-to-handle-full-load-while-partially-inserted
I'm struggling to see why that isn't being paid attention to.

You can have 99 examples of things working fine. You only need one of it not working fine. The 5090 didn't exist in 2023 and we're also a year further ahead in time, wear and tear on 12VHPWR cabling.

 

[Chrispy_]

That's a myth. No one seems to want to admit a recall is needed.
Moddiy are all but admitting it here - https://www.techpowerup.com/forums/...ables-for-geforce-rtx-50-series-cards.332507/
If the cable plug has a H++ marking then you should be all good.

Yes, I am speculating. But no one has proved me wrong either. I haven't yet seen a H++ marked cable get hot.

If a full 12V-2x6 (H++ on the cable plug) example of a hot running (within spec) connector can be demonstrated. I'll happily change my mind and admit the connector design is simply too weak to support 600 W.

PS: If you read between the lines in Moddiy's statements. They are saying that the spec doesn't cover female pin designs inside the plugs. It's just whatever fits the male pins. So, an upgraded spring contact can happen without a change to the spec.

Which also means it could be applied to the older plugs too. But who'd do that when you can just de-list them and offer the newer model instead.

I mean, there are two distinctly different issues here.

1. Increased current through a smaller connector. Revisions to 12VHPWR will help avoid/reduce incorrectly-seated connectors from sending far too much current down only some of the wires. It doesn't change the fact that 8.3A are still being pushed through a connector that's only rated to 9.5A, but it at least improves the odds that all six pairs of 12V are making decent contact.
   
   
2. Some of the 40-series and 50-series GPUs (most notably FE cards and AIB cards based on Nvidia's reference layout) do not have per-wire current monitoring/balancing - which makes them unsuitable for a multi-wire connector that doesn't have enough safety margin to run all of the needed current over a single wire pair in the event of a connection fault or cable damage.

So the worst-case scenario where stuff still melts is a combination of both issues. You can't argue one without acknowledging the other. If the GPU had shunts that could detect a cable fault and load-balance or shut down safely, nothing bad would have happened. If a more robust connector like the old 8-pin with enough safety margin to handle a total failure of wiring/connector such that all the current was forced over a single pair of pins, nothing bad would have happened. The problem with 12V-6X2 is that it doesn't fix point 1. The revised connector msure is more robust and connects more reliably now, but poor connections haven't been the problem for all of the 50-series melting connector incidents so far. The issue has been too much current, despite a good connection - and that's why the sleeving of the cables melted and why the PSU end melted too.

 

[chrcoluk]

Pass. Why would anyone except the few hundred 5090 owners even care? Nvidia includes adapters in the box so I don't think anything would change if the connector was changed. New adapter(s) would be included in the new boxes if there was ever a change in connector standard.

600W for a single GPU is horrifyingly ridiculous. This is a world that's trending towards laptops and tablets. Even in the datacenter, performance/Watt is the key motive for purchases and 99.9x% of people do not want a 600W GPU.

Using adaptors with 2 12vhpwr connectors would also be a problem, would need an amount of 8 pin cables no PSU has. Nvidia have gone all in on this ridiculous idea and its coming home to roost.

 

[Chrispy_]

Using adaptors with 2 12vhpwr connectors would also be a problem, would need an amount of 8 pin cables no PSU has. Nvidia have gone all in on this ridiculous idea and its coming home to roost.

If you want two 600W GPU connectors, you're using a 1500W-2000W PSU, and those come with loads of 8-pin capability:

(Corsair 1600W)

It still wouldn't solve the problem, just pointing out that high-wattage PSUs exist almost exclusively for lots of graphics cards so they have a whole bunch of PCIe 8-pin connectivity.

 

[user556]

I mean, there are two distinctly different issues here.

1. Increased current through a smaller connector. Revisions to 12VHPWR will help avoid/reduce incorrectly-seated connectors from sending far too much current down only some of the wires. It doesn't change the fact that 8.3A are still being pushed through a connector that's only rated to 9.5A, but it at least improves the odds that all six pairs of 12V are making decent contact.
   
   
2. Some of the 40-series and 50-series GPUs (most notably FE cards and AIB cards based on Nvidia's reference layout) do not have per-wire current monitoring/balancing - which makes them unsuitable for a multi-wire connector that doesn't have enough safety margin to run all of the needed current over a single wire pair in the event of a connection fault or cable damage.

So the worst-case scenario where stuff still melts is a combination of both issues. You can't argue one without acknowledging the other. If the GPU had shunts that could detect a cable fault and load-balance or shut down safely, nothing bad would have happened. If a more robust connector like the old 8-pin with enough safety margin to handle a total failure of wiring/connector such that all the current was forced over a single pair of pins, nothing bad would have happened. The problem with 12V-6X2 is that it doesn't fix point 1. The revised connector msure is more robust and connects more reliably now, but poor connections haven't been the problem for all of the 50-series melting connector incidents so far. The issue has been too much current, despite a good connection - and that's why the sleeving of the cables melted and why the PSU end melted too.

The flawed cables are flawed, it's that simple. Point one is a bunch of assumptions, I'm going with the assumption that the design/spec is rated correctly (And so far the newer H++ cable plugs are looking good). Point two is only protection, it doesn't fix the problem. If the design/spec is not good enough then point two is moot. It won't keep the cards running.

Just need to clear out all the bad cables now.

 

[Chrispy_]

The flawed cables are flawed, it's that simple. Point one is a bunch of assumptions, I'm going with the assumption that the design/spec is rated correctly (And so far the newer H++ cable plugs are looking good). Point two is only protection, it doesn't fix the problem. If the design/spec is not good enough then point two is moot. It won't keep the cards running.

Just need to clear out all the bad cables now.

I made no assumptions, I'm only stating facts that are verified by the spec, the manufacturers' official posts on why the spec was changed.

You seem to think H++ cable plugs are somehow improved to fix a flaw in the old H+ cable plugs. They are not. H+ plugs are identical to H++ plugs, it is simply a label change to match the labelling of new 12V-6X2 connectors.

H+ vs H++ on 12V-2x6 / 12VHPWR cables: What does it mean?

Does H++ or H+ have any meaning when it comes to your 12V-2x6 cable?

www.corsair.com

Please stop harping on about cables, there's a 700+ post discussion in the other thread that already put the paranoia about cables to rest already.

 

[user556]

Yet Middiy have stated they are improved. And the fact is it is the 12VHPWR cable plugs that are getting hot. So far there hasn't been a single H++ marked cable failure.

 

[KLMR]

So, after all, the connector was to save on shunts and DC/DC conversion the board side?
I mean that power connector is not even used in workstations/servers.

Before, you probably had to have one shunt for each connector, so 8pin + 8pin + 6pin made 3 (to know if the connector "was" connected). Or probably the 8pin required 2. What was the design trend before "the meltdown" of the *12V-single-connector*?