8-pin PCIe to ATX 12VHPWR Adapter Included with RTX 40-series Graphics Cards Has a Limited Service-Life of 30 Connect-Disconnect Cycles

[Blue4130]

Der8auer who owns Thermal Grizzly. It's called the pump out effect. The lower the viscosity of a paste the quicker the process is. Lower viscosity paste tends to perform better than that of high viscosity paste at the cost of increased pump out. Intel stock paste for example of a low performing high viscosity paste but doesn't need to be replaced frequently.

This is particularly important for high wattage components like GPUs as the larger heat load means the impact of pump out is more pronounced.

So the guy selling you thermal paste says that you need to change it out yearly..... Hmmmmm. Ok.

 

[Frick]

Can confirm. Same thing happens with PSUs. The TIM used between MOSFETs and the heatsinks. Yet, when a PSU dies, it's always "U mSt HaVe BlowN a CAP" when odds are it's a MOSFET or schottky diode suffers thermal runaway.

The capacitor plauge will never leave the collective consciousness.

 

[Haile Selassie]

News flash everybody, most LGA CPU sockets are rated for 25 insertions, some even for as low as 10.
LOTES PCI-Express slots are rated for 50 insertions as well.

 

[Valantar]

I honestly don't see this as any kind of surprise. They're converting 4*8=32 cables into 12 (though it's really 12+20 into 6+6), meaning there'll either be multiple joins in each cable or you might have some crimp terminals with four wires crimped into them (unless they leave some grounds disconnected, which they probably do). Even with three wires per crimp, there's no way of doing this well, and certainly no way of doing this in a way that isn't very sensitive to bending and pulling - unless they solder the crimps, which would make these cables extremely inflexible. Either way: bad times. I like the 12HPWR cable in some ways, but this is definitely not one of them.

 

[Bomby569]

So the guy selling you thermal paste says that you need to change it out yearly..... Hmmmmm. Ok.

no, you just didn't understood what he said

 

[R-T-B]

They won't. Drama gets hits.

I know you're right, but it's still sad.

 

[konga]

I honestly don't see this as any kind of surprise. They're converting 4*8=32 cables into 12 (though it's really 12+20 into 6+6), meaning there'll either be multiple joins in each cable or you might have some crimp terminals with four wires crimped into them (unless they leave some grounds disconnected, which they probably do). Even with three wires per crimp, there's no way of doing this well, and certainly no way of doing this in a way that isn't very sensitive to bending and pulling - unless they solder the crimps, which would make these cables extremely inflexible. Either way: bad times. I like the 12HPWR cable in some ways, but this is definitely not one of them.

No, this is just the standard durability rating for all connectors of this type.

 

[damric]

To be fair, 30 insertions was the best 2 minutes of her day.

 

[evernessince]

So the guy selling you thermal paste says that you need to change it out yearly..... Hmmmmm. Ok.

Or you can use the internet to see that other sources verify what I just said.

For reference, here is a time stamp in a video where Steve and DB talk about the pump out effect (among other things):

 

[Jism]

I have a 850 Antec PSU, with 2x 8 Pin connectors. Only used one over the years. The one 8 pin connector simply eroded over time of pulling and putting back in. If that happens the voltage that can be provided to your GPU VRM will be lower. to even below 11V if it's really eroded.

 

[jonnyGURU]

I have a 850 Antec PSU, with 2x 8 Pin connectors. Only used one over the years. The one 8 pin connector simply eroded over time of pulling and putting back in. If that happens the voltage that can be provided to your GPU VRM will be lower. to even below 11V if it's really eroded.

Well.. quality of the terminals play a big part. There isn't just one quality of connector/terminal. Just look on Molex website. There's mini-fit jr. connectors/terminals of different thermal ratings, current ratings, insert/remove cycles, etc. and pair that with the gauge wire used and you have a big random pot of whatever the manufacturer decides to use and you end up with different results.

I will say this: everyone here should buy a tube of dielectric grease.

 

[Blue4130]

Or you can use the internet to see that other sources verify what I just said.

For reference, here is a time stamp in a video where Steve and DB talk about the pump out effect (among other things):

It's fine, I know what pump out is, I just don't agree with your yearly time frame.

 

[Mussels]

30 uses aint much when manufacturing tolerances vary, so you might get 50 or you might get 5



It'll get 0 from me

 

[jonnyGURU]

30 uses aint much when manufacturing tolerances vary, so you might get 50 or you might get 5



It'll get 0 from me

LOL... I think it's you might get 30 or you will probably get more.

I did a 100 piece CPK for 30 cycles of the 12VHPWR connector and saw negligible degradation of the connector's capability.

As someone else already point out: The Molex Mini-fit Jr. We all already use today is also rated at 30 cycles (unless you use HCS, which only some companies do).

This isn't news. It's a soap opera.

 

[OneMoar]

this is standard fare for connectors of this type
this is not new this is not different then pcie 8 pin

the adapter melting is happening because you are trying to BOND 9 12V pins and 9 GROUNDS to 6 12V pins and 6 GROUNDS (the 12VHPWR 12 pin connector has 6 power and 6 ground) which results in lots of resistance As bonding wires at the end is quiet terrible for electrical resistance and you add the smaller pin Diameter to that and boom stuff gets melty

this is possible todo safely but would require somebody engineer a pcb adapter with a big power and ground plane to spread the load and dissipate heat at the connector probably something the size of EVGAS power link and would need to contain a fairly heafy pcb with enough copper to handle the thermo-electrical load

 

[saki630]

Why cant they have two PCIE connections since all the new GPU's are about 3-card-slots-thick with the cooler? You can then get over 500W with another slot to go!

 

[Valantar]

Why cant they have two PCIE connections since all the new GPU's are about 3-card-slots-thick with the cooler? You can then get over 500W with another slot to go!

...what? PCIe slots deliver a maximum of 75W (really less, as that 75W is for 3.3+12V). You need power connectors for anything more than that. And, of course, that would make the cards incompatible with the vast majority of motherboards out there.

 

[Divide Overflow]

Remember when cable management meant routing for cooling and aesthetics and not carefully monitoring them to make sure they aren't melting?

 

[AleXXX666]

Right, but PCI-E 6 pin was around for a long time before they introduced the 8 pin connector. Now that you mention it, I do remember the 12VHPWR connector originally being 12 pins, now it's 16... well, that escalated quickly.

NGreedia wants more money so they reinvent power connectors i hope with 5000 series we will see USB-C universal connector for powering gpu right from the wall with some Xiaomi USB-C power adaptor lmfao

Fair point,.....

However, how many times are you going to realistically remove and install an Intel CPU? Especially so when virtually each new processor line requires a new motherboard. In some respects Intel might as well have soldered the CPU to the motherboard,.....

Not so with video cards,...

lmfao you all look one-minded how about selling mobo without cpu so it goes for re-use in other hands and you could sell it like 30 times so what? i doubt the socket will fail lol

 

[OneMoar]

3 8 pin connectors might be just enough power for a 4080 as long as it to doesn't have those nasty transients
there is nothing wrong with the 12 pin connector, the issue is the type of adapter cables used and the amount of conductors

 

[jonnyGURU]

3 8 pin connectors might be just enough power for a 4080 as long as it to doesn't have those nasty transients
there is nothing wrong with the 12 pin connector, the issue is the type of adapter cables used and the amount of conductors

No. That's incorrect.

Mini-fit HCS terminals support more current than the 12VHPWR connector's terminals (10A vs. 9.5A), so you don't need more than 12 conductors (therefore doesn't need to split to three 8-pins). Nvidia and AIBs are providing adapters with more than two 8-pin connectors because they don't know what your PSU comes with in terms of terminal quality and wire gauge, so they're covering all bases.

The "issue" with adapter cables was a story that was made up by wccftech and the Jayz re-reported with additional crying and eye rolls. The fact of the matter is, the connector with the problem that PCI-SIG was telling members about was the 12VHPWR connector. NOT the 8-pins: https://cultists.network/8815/melting-12vhpwr-connectors/

For the record: I work for a member of the consortium and am on that mailing list that leaked email was on and can confirm what's fact and what's fiction. Too bad wccftech, Jayz, etc. didn't bother to reach out to any industry people before spreading fake news.

 

[stimpy88]

I'll just say it... "nGreedia".

 

[OneMoar]

No. That's incorrect.

Mini-fit HCS terminals support more current than the 12VHPWR connector's terminals (10A vs. 9.5A), so you don't need more than 12 conductors (therefore doesn't need to split to three 8-pins). Nvidia and AIBs are providing adapters with more than two 8-pin connectors because they don't know what your PSU comes with in terms of terminal quality and wire gauge, so they're covering all bases.

The "issue" with adapter cables was a story that was made up by wccftech and the Jayz re-reported with additional crying and eye rolls. The fact of the matter is, the connector with the problem that PCI-SIG was telling members about was the 12VHPWR connector. NOT the 8-pins: https://cultists.network/8815/melting-12vhpwr-connectors/

For the record: I work for a member of the consortium and am on that mailing list that leaked email was on and can confirm what's fact and what's fiction. Too bad wccftech, Jayz, etc. didn't bother to reach out to any industry people before spreading fake news.

fun both jay and linus reported that the failure was happening with adapters so I figured it must have been a poor bond or buss connection in the 3 way adapter
do you know if the testing was done at 600W or 450W ?

but each 8 pin can only carry 150W per spec so really you need 4 for the full 600w right

 

[Valantar]

fun both jay and linus reported that the failure was happening with adapters so I figured it must have been a poor bond or buss connection in the 3 way adapter
do you know if the testing was done at 600W or 450W ?

AFAIK the failures are unrelated to the 30 connect-disconnect cycle rating though, but rather due to poor pin connections in the adapters, whether this is caused by low quality wiring, poor crimps, the inherent weakness of having multiple wires crimped into a single terminal, or something else.

but each 8 pin can only carry 150W per spec so really you need 4 for the full 600w right

The PCIe power spec for the 8-pin is 150W/header, but the spec for the terminal block used and its crimps is 10A/pin - which translates to 10Ax3x12V or 360W/connector on paper (an 8-pin has five grounds and three 12V pins). So, the headers can technically deliver 600W (though they'd need to be paired with some beefy wiring!), but to do so they'd have to exceed the PCIe power connector spec.

 

[jonnyGURU]

fun both jay and linus reported that the failure was happening with adapters so I figured it must have been a poor bond or buss connection in the 3 way adapter
do you know if the testing was done at 600W or 450W ?

Linus copied JayZ. JayZ coped wccftech. If everyone just copies the same incorrect information, that doesn't make it correct.

Honestly, what the "press" has become lately makes me really, really sad.

do you know if the testing was done at 600W or 450W ?

The testing was done after securing a < 30mm bend radius at the PSU. Room temp was 26°C. Temperatures were above spec after 2.5 hours at continuous 55A load and exhibited melting at 10 to 30 hours (multiple results of multiple samples tested). Root cause is suspected to be the terminals farthest from the inward bend are deflected away from the outer wall of the receptacle surface causing high resistance and therefore current demand increasing on pins of lower resistance.

but each 8 pin can only carry 150W per spec so really you need 4 for the full 600w right

I don't know why everyone keeps saying that. Connector and terminal specification documents are publically available from every manufacturer.

The 150W "spec" comes from PCI-SIG, driven by Nvidia, as well. The had already defined that the reason for the 6-pin on a GPU was because the GPU needed 75 more watts of power than what the slot could deliver. Needs 150W? Put in a second 6-pin. But when cards started needing 225W additional, Nvidia thought it would be silly to put a third 6-pin on the card. So they asked PCI-SIG if they could put two sense pins on an 8-pin connector so, if grounded, the card would "know" that it was safe to demand 150W from a single connector.

Think about it logically. Surely you would expect 6 conductors on a mini-fit jr. connector to deliver more than the equal amount of power as 4 tiny pins in a PCIe slot. And you don't magically double the capacity of a 6 conductor connector by simply adding two sense pins. If you want to double the the capacity, you MORE THAN double the conductors. But that wasn't the goal here. The goal here was to specify the power demands of the card: 75W, 150W, 225W, 300W, so on.

The cheaper terminals, of reputable brands, use the brass w/ tin terminals with 18g wire. In a 2x4 configuration with 2x3 terminations, those terminals support 8A per conductor. So, your "typical" 6-pin PCIe, or even 8-pin PCIe (since it's still technically only 6 power conductors) is capable of 24A. So, at 11.4V (because we always work +/-5%), you're talking about 273.6W per connector. Fine for a 6-pin and 8-pin on the same cable, but not good for two 8-pins which is when they tell you not to daisy chain. It's also why Nvidia's squid adapter has three 8-pins and not just two.

Some manufacturers use mini-fit HCS terminals. These are rated at 10A per terminal in a 2x3 configuration. So using the same math, you have 342W per connector assuming voltages drop to 11.4V.

The cable Corsair made here: https://www.corsair.com/us/en/Categories/Products/Accessories-|-Parts/PC-Components/Power-Supplies/600W-PCIe-5-0-12VHPWR-Type-4-PSU-Power-Cable/p/CP-8920284 uses 6x mini-fit HCS terminals per 8-pin connector. So two of those type 4 connectors are capable of a total of 60A, which is, obviously, GREATER than the capability of the 12VHPWR connector.

AFAIK the failures are unrelated to the 30 connect-disconnect cycle rating though, but rather due to poor pin connections in the adapters, whether this is caused by low quality wiring, poor crimps, the inherent weakness of having multiple wires crimped into a single terminal, or something else.

Nope. Not even poor pin connections in the adapters. Pin connections in the 12VHPWR connector itself.

Which is funny that JayZ noticed it... touched on it.... showed a picture of the connector melting... .but still went on about how it's the adapters fault. Yes... There's A LOT of tape wrapped around the 12VHPWR connector on the GPU side to limit the bend radius. Correct. To limit the bend radius at the 12VHPWR connector. Come on Jay... use your head... what does that have to do with the 8-pin or the fact that the cable you're holding is an adapter?????

He's telling people "you better buy a PSU that's ATX 3.0 (which to him means has the 12VHPWR connector on the modular interface)" which is LITERALLY the part that is melting!!!!!!!!