News Corsair’s PSU Expert Jonny Guru Weighs in on Nvidia Connector-Gate

[watzupken]

Many "experienced reviewers / testers" have COMMENTED on the issue and shared their thoughts on what they think the root cause might be but I am not aware of any of them having personally witnessed or reproduced it themselves yet.

I stand corrected. But what you mentioned further confirmed that nobody knows the root cause of the issue. I recalled watching the testing on Gamers Nexus where they tried the negative possibilities to try and recreate the issue, but was not successful. So it just prove that Johnny Guru's claim may not address the issue. The only thing they highlighted as a potential cause may be the cable. In any case, I feel people were looking for issues with the cables/ connectors/ connection, but with Nvidia keeping so quiet, I do wonder if it has something to do with the connector on the GPU itself.

 

[digitalgriffin]

One thing I will note about these pin connectors is they can and do fail when the barb is not properly deployed during the mfg process.

Let me explain. When manufacturers make these connectors, the pins are attached to the wires and then inserted into the connector housing.

To prevent the pins from slipping out there's a barb that get squished when inserted one way, till is passes the catch, and then pops up. To visualize this imagine how your Ethernet cable has a barb on it that squishes as you push it in. Then the barb pops up as the barb passes the catch. Then you can't remove it unless you squish the barb.

I've had more than a few connectors fail on me in my lifetime. Why? Bad barbs or catches. It's more sensitive on tiny connectors.

So, what happens is you go to push the connector in and the pin pushes out because of a bad barb or catch. Then you have a bad connection. Especially if these connectors are dependent upon full insertion.

The other thing that can happen is the socket/pin is out of spec. When they are made they are folded in a row on a tin sheet. So you have a row of these pins and sockets sitting on the a wire after the press stage. These are manually trimmed off. This might result in some minor length variation that affects offset inside the socket. I've seen this cause minor pin and socket discrepancies on cheap molex connectors. But the pins and sockets are so big here, minor variation doesn't affect them much.

I have a lot of respect for Johnny G., And I do NOT think he's trying to mislead. He's always been about quality and one of the reasons I buy Corsair and Seasonic for 90%+ of my builds. I think he's trying to promote a solution. But the truth is, if these power connectors are failing so easily, and causing so much damage, then they should be recalled. Grease is not the answer.

 

[Zurbo]

First, let me state that I do not have one of these cards and connectors to investigate.

However, as someone who has performed FA to the component level on computer data storage products for a living for the past 20 years for a leading data storage provider, I do see some concern over this design as I have personally observed overheated connectors of this type in our own products over the years where we were running the upper edge of the ratings for the power connectors.

Observation #1:
The connector on the graphics card is shown as an Amphenol Minitek Pwr CEM-5 PCIe.
Website - https://www.amphenol-cs.com/product-series/minitek-pwr-cem-5-pcie.html#
Datasheet - https://cdn.amphenol-cs.com/media/w...ardwiretoboard/bwb_minitek_pwr_cem_5_pcie.pdf

Takeaways from the datasheet:

1. EACH power connector is rated for 9.5A per contact - As there are 12 contacts (6 positives and 6 grounds), this creates 6 current loops.
2. The rated temperature rise @ the rated current is 30C max. Taking 25C as ambient 55C should be what these run at full power draw which is not a stretch from the typically observed 53C and is not hot enough to melt or deform Nylon 66 which is 268.8C.
3. The rated insertion count/durability rating is 30 mating cycles (after that the connector should be considered worn and derated from its maximum ratings as appropriate).
4. Insertion force is 8.0 Newtons = 1.79847 lbf.

Observation #2:

1. Using Kirchoff's junction rule (the sum of all currents entering a junction must equal the sum of all currents leaving the junction) the connector is really only capable of sustained current carrying capability of 57A which at 12V means 684W.
2. In my professional opinion, this is underrated for the application (assuming a constant 600W load).
3. Robust electrical design dictates a DOUBLE derating meaning that if they are going to rate it for 600W, the connector should really be capable of 1200W.
4. They care certainly riding the upper limits of the ratings with this design.

Observation #3:
There are obviously quite a few reports of the power pins overheating on both the male and female ends of this Amphenol Minitek Pwr CEM-5 PCIe connector.

Items I would investigate:

1. I am not going to disagree with the seating theory as I have personally observed the same root cause in our applications of Molex and Amphenol power connectors.
2. I would love to see a metallurgical analysis of the pins and barrels (both male and female via cross-sectional SEM analysis to prove the material used is meeting the requirements set forth by Amphenol. (to eliminate the possibility of out-of-spec material but also counterfeit components).
3. Also would audit adherence to the following barrel crimp specifications as set forth by Amphenol for the female side of the connectors - https://cdn.amphenol-cs.com/media/wysiwyg/files/documentation/gs-20-0704.pdf as failure to fully comply with these specifications could easily lead to excessive temperature rise.
4. Possible improper retention of the barrels in the connectors preventing full wipe insertion of the barrels to the pins?
5. Possible casting flash on the connector bodies preventing full insertion and locking of the solution to prevent loosening?
6. Is this also occurring at the POWER SUPPLY end of the cable?

 

[PEnns]

The nGreedia shills, paid or otherwise, are still hard at work blaming the users!!

The users should use lubrication for their $1600 + badly designed GPU, otherwise it's their problem?? Some shills would sell their soul and reputation for an extra buck.

And of course nGreedia is silent while those shills are hard at work passing the blame to the users.

Pitiful.

 

[edzieba]

With the total inability for anyone to recreate the other proposed failure modes (internal pin pullout, solder cracking, overbending, etc) the common failure mode of incomplete insertion does seem like the most viable possibility. No connector can be made completely proof of user error, as a connector with a low insertion force will also by definition have a low retraction force, and the (already present!) locking latch can and will be ignored by installers.

OEM testing has also seen heating from incorrectly inmserted connectors:

https://twitter.com/x/status/1585257428291325958

View: https://twitter.com/hms1193/status/1585257428291325958

You don't need dielectric grease. Just touch your finger to some vaseline or the end of a chapstick and rub it on the connectors, and wipe off any residual as it will liquify when it gets warm. Just enough to barely stick to the plug. Push it in, take it back out, wipe the connector a second time to remove any squished up extra and plug it back in.

And no, it doesn't damage the plastic at all. I've done this with 24 pin mobo connectors and other large/tight connectors for decades. Chap stick is what I usually use, because if you're in a room with 3 people, one of them has some.

Of course, "don't put too much vaseline on it before you stick it in" is advice that will serve you well on a universal basis?

Perfectly reasonable explanation right here, I've even had the 8 pin connectors get difficult on me, especially when they're really close together with other 8 pins.

Don't use flammable petroleum grease for high power connectors! That's an absolutely terrible idea, particularly when dielectric grease is dirt cheap and the correct tool for the job.

1. EACH power connector is rated for 9.5A per contact - As there are 12 contacts (6 positives and 6 grounds), this creates 6 current loops.

[...]

1. Using Kirchoff's junction rule (the sum of all currents entering a junction must equal the sum of all currents leaving the junction) the connector is really only capable of sustained current carrying capability of 57A which at 12V means 684W.
2. In my professional opinion, this is underrated for the application (assuming a constant 600W load).
3. Robust electrical design dictates a DOUBLE derating meaning that if they are going to rate it for 600W, the connector should really be capable of 1200W.
4. They care certainly riding the upper limits of the ratings with this design.

The rated current capacity per pin already includes the derating for safety factor. Double-derating is just overspeccing.

 

[RandomWan]

Wouldn’t it be fairly straightforward to replicate the problem to prove this theory?

It should be pretty straight forward to replicate any of the theories put forward so far, but no one has managed to melt one down in a test environment that I'm aware of. If the connector not being fully seated was a cause, then they should have already replicated this. The connector not being fully seated would only cause damage not otherwise seen only if there was arcing involved, I would think.

Until someone starts reliably melting them down in tests to confirm an outside or user induced cause, they should just start replacing them with 90 degree connectors manufactured to higher quality standards to avoid the perception of it being a potential issue and eliminate their bending problem all together.

 

[F-minus]

We have 4 slot cards, wide that they don't fit 50% of cases on the market and long over 380mm, and they are trying to tell us that the new connector is for saving space. Just design a proper connector, with fat prongs, there's plenty on the market already, they also take marginally more space than 3x 8pin PCIe connectors. This trend to downsize everything while increasing the load is asinine. The 16 pin ATX connector could fit most of those cards.
But no, oh no, my beautiful 4 slot monster of a card that's almost as big as a mini-itx case has a too big power connector and the PCB designers are clueless where to put it, regardless that PCBs are 1/3 of the size of those new cards. Just make a damn bigger PCB and put a proper plug on it, not some tiny <Mod Edit> that is melting.

 

[palladin9479]

The problem isn't hard to understand if we all stop pretending the world is perfect and that bad things happen without our knowledge. In this case the likely culprit is that poor quality on some adapters solder joints creates a condition where stress causes micro-fractures that then increases the resistance on that joint. All the incoming power lines are soldered to the exact same power plane, might as well just have one giant two pin power plug, this means power is always evenly distributed across all contacts. If a contact breaks completely, then the power is balanced across the remaining contacts, but if it breaks partially then it's still an open circuit and pulling power across the damaged joint. It's incredibly hard to reproduce something like that on demand as someone would need both a defective cable and to reproduce the exact same solder damage.

In any case, no amount of media suppression or shade is going to change that fact that NVidia is now rethinking it's solution. Honestly, that's way too many amps across such a small space on a user serviceable component. Anytime I've seen power like this, it's been with locking connectors that regular users are told not to touch.

 

[Tony1337]

Nvidia paid Corsair and Jonny Guru to blame user error and take heat off Nvidia?

 

[InvalidError]

The problem isn't hard to understand if we all stop pretending the world is perfect and that bad things happen without our knowledge. In this case the likely culprit is that poor quality on some adapters solder joints creates a condition where stress causes micro-fractures that then increases the resistance on that joint.

If the problem was that simple, GN cutting one wire off altogether and a second wire breaking off during testing of their "sacrificed to science" adapter should have caused a failure. It didn't, so double wire/solder failure on 12V on Nvidia's adapter isn't the root cause, the remaining two #14 wires and the connector's internal bridge still get the job done, barely reaching 70C after 4+h of heavy load. This shows that the wiring and soldering themselves have a 100+% safety margin.

 

[RodroX]

Maybe 10% of the melting issues we all know about, could be do to the connector not fully inserted, I hardly doubt many will spend +$1600 on a GPU and not pay attention when inserting the power cable to it, many of this user most likely already had a high performance GPU so they have experience inserting a power cable.
We all get it, it is hard to connect and the latch is not easy to see (the cards are huge), but thats not users fault, thats bad design from the ground.
If you make a refrigerator that you need to press a hidden button to secure the door when closed, then most people will complain the thing is not cooling they beer, thats not user fault, and it makes no sense unless the refrigerator is traveling the wild oceans, or going deep space exploration.

 

[palladin9479]

If the problem was that simple, GN cutting one wire off altogether and a second wire breaking off during testing of their "sacrificed to science" adapter should have caused a failure. It didn't, so double wire/solder failure on 12V on Nvidia's adapter isn't the root cause, the remaining two #14 wires and the connector's internal bridge still get the job done, barely reaching 70C after 4+h of heavy load. This shows that the wiring and soldering themselves have a 100+% safety margin.

A broken connection is actually preferable over a damage connection as there is zero current going through the broken connection. A connection that has had it's solder damaged but not broken just transformed into a poor quality resistor. The microscopic air bubbles that would get into the metal would cause it's resistance to heat up faster then it's supposed to. Plus this would only show up on defective adapter plugs which those individuals wouldn't get unless they are purchasing cards by the pallet and testing all involved. Of course I mentioned this in my statement, but who has time to read anymore.

 

[digitalgriffin]

A broken connection is actually preferable over a damage connection as there is zero current going through the broken connection. A connection that has had it's solder damaged but not broken just transformed into a poor quality resistor. The microscopic air bubbles that would get into the metal would cause it's resistance to heat up faster then it's supposed to. Plus this would only show up on defective adapter plugs which those individuals wouldn't get unless they are purchasing cards by the pallet and testing all involved. Of course I mentioned this in my statement, but who has time to read anymore.

A lot of that depends on the power circuitry on the board. If the power is on a common plane, or they go to separate power circuits.

 

[palladin9479]

A lot of that depends on the power circuitry on the board. If the power is on a common plane, or they go to separate power circuits.

It's a common plane. All four grounds are soldered to the same metal bar, all four hots are soldered to their own common bar. Electrically it's a two wire power plug carrying 600W of power that just uses eight and twelve pins at the junction.

Wiring diagram

 

[InvalidError]

We all get it, it is hard to connect and the latch is not easy to see (the cards are huge), but thats not users fault, thats bad design from the ground.

Bigger, higher-current connectors and cables are harder to work with, there is no magical work-around for that other than raising the power distribution voltage to reduce the current needed to deliver a given amount of power.

Imagine the fun people would have if GPUs switched to bare wires into caged nut terminals!

A broken connection is actually preferable over a damage connection as there is zero current going through the broken connection. A connection that has had it's solder damaged but not completely broken just transformed into a poor quality resistor.

The Nvidia connector has an internal bar connecting all pins together, which means the pins near the broken wire are at fundamentally the same potential as all of the other pins that still have a full connection to their feed wires. Any increase in one wire's connection due to a broken solder joint will steer practically all of its current to other pins.

1m of #14 is ~8mOhm + 6mOhm from the connector = 14mOhm neglecting solder resistance and there are nominally four of those in parallel. Since they are bussed together internally within the connector, lets say the nominal wire-to-bus resistance excluding the PSU-side connector and including the HPWR-side solder is 12mOhm.

How much is your hypothetical solder break's resistance? 100mOhm? So you have 112mOhm in parallel with 3x 12mOhm (the three other wires, which is 3.9mOhm overall, which means ~96% of all current will flow through the three remaining wires, leaving only 2A for the damaged wire. 2A through 0.1ohm is 0.4W which is a non-issue. If you increase the break resistance more, the power goes down even further and becomes even less of an issue. If the break is 20mOhm, that would be 32m // 3x12m = 3.6mOhm, which means you still have almost 90% of the current going through the three good wires and 5A into 20mOhm is 500mW, still not going to melt connectors without some help from rock wool or other additional insulation.

BTW, all melted connectors shown so far have been most heavily damaged on the terminal side, which I take as a strong indication that the issue has nothing to do with the soldering itself.

 

[palladin9479]

Bigger, higher-current connectors and cables are harder to work with, there is no magical work-around for that other than raising the power distribution voltage to reduce the current needed to deliver a given amount of power.

Imagine the fun people would have if GPUs switched to bare wires into caged nut terminals!


The Nvidia connector has an internal bar connecting all pins together, which means the pins near the broken wire are at fundamentally the same potential as all of the other pins that still have a full connection to their feed wires. Any increase in one wire's connection due to a broken solder joint will steer practically all of its current to other pins.

1m of #14 is ~8mOhm + 6mOhm from the connector = 14mOhm neglecting solder resistance and there are nominally four of those in parallel. Since they are bussed together internally within the connector, lets say the nominal wire-to-bus resistance excluding the PSU-side connector and including the HPWR-side solder is 12mOhm.

How much is your hypothetical solder break's resistance? 100mOhm? So you have 112mOhm in parallel with 3x 12mOhm (the three other wires, which is 3.9mOhm overall, which means ~96% of all current will flow through the three remaining wires, leaving only 2A for the damaged wire. 2A through 0.1ohm is 0.4W which is a non-issue. If you increase the break resistance more, the power goes down even further and becomes even less of an issue. If the break is 20mOhm, that would be 32m // 3x12m = 3.6mOhm, which means you still have almost 90% of the current going through the three good wires and 5A into 20mOhm is 500mW, still not going to melt connectors without some help from rock wool or other additional insulation.

BTW, all melted connectors shown so far have been most heavily damaged on the terminal side, which I take as a strong indication that the issue has nothing to do with the soldering itself.

Can you please read my statement in full before responding. A broken connection is actually preferred, and likely what was tested against. A broken connection has zero current going through it, that is a good thing. Injecting microscopic air bubbles into solder, which happens when it's stressed, will cause it to start acting like a poor resistor. It still conducts electricity, only now it's acting like a light-bulb filament. Normally this isn't a big issue as cables in this amp range tend to be massively over-designed to account for this happening. I believe that we have a case of poor manufacturer QA having defective adapters being shipped out randomly that when placed under the aforementioned stress cause enough heat to melt the plastic.

Like I said before, no amount of media cover or shade is going to shift this to "customers are holding it wrong". Johnny Guru inadvertently admitted that adapters were melting, and that they just replaced the adapter. US consumer safety folks don't take kindly to the words "consumer" and "potential fire hazard" being used in the same sentence. nVidia is already looking for a solution.

 

[-Fran-]

This is the first time I see the diagram and I have a very plausible, yet hard to prove, theory with that way of wiring it. If I'm right, the engineers that approved that "daisy chain" of power need to be executed via electrocution.

So, when you solder or even just pressure-attach cables, there's always potential for expanding and contracting the solder, which can cause micro-fractures and micro-pockets of air/dust creating resistance. If you have DC going up and down, then it means you'll be thermally cycling the weakest link in that stupid daisy chain.

So, in short the key aspects are:

• solder quality or attach pressure
• temp deltas (via power deltas)
• time of the deltas (ramp up / down vs temps)
• micro-fractures

I've seen this at a higher level via material fracture for big mining operations (don't ask how I know xD) and when they crack, it's super visible and dangerous, yet they can still operate to their "rated capacity", but you obviously eat all your safety margins immediately.

That's my theory and if I'm right, it'll cost nVidia a pretty penny to fix it as it needs a full re-design.

Regards.

 

[RodroX]

@InvalidError
"Bigger, higher-current connectors and cables are harder to work with, there is no magical work-around for that other than raising the power distribution voltage to reduce the current needed to deliver a given amount of power.

Imagine the fun people would have if GPUs switched to bare wires into caged nut terminals! "

lol, yeah I get that. Perhaps instead of trying to feed 600watts over 6 high current, yet tiny pins (half of those 12 pins for the 4x 8-pin old school ones), Nvidia should have tried a middle ground and go with at least 8 pins for the 12 volts power?

I already wrote this on other posts, it does not seem, so far, that the cables or the soldering are the issue here. The problem for me is located on the "tip" of the female pin of the adapter or the native psu cable.

 

[palladin9479]

This is the first time I see the diagram and I have a very plausible, yet hard to prove, theory with that way of wiring it. If I'm right, the engineers that approved that "daisy chain" of power need to be executed via electrocution.

So, when you solder or even just pressure-attach cables, there's always potential for expanding and contracting the solder, which can cause micro-fractures and micro-pockets of air/dust creating resistance. If you have DC going up and down, then it means you'll be thermally cycling the weakest link in that stupid daisy chain.

So, in short the key aspects are:

• solder quality or attach pressure
• temp deltas (via power deltas)
• time of the deltas (ramp up / down vs temps)
• micro-fractures

I've seen this at a higher level via material fracture for big mining operations (don't ask how I know xD) and when they crack, it's super visible and dangerous, yet they can still operate to their "rated capacity", but you obviously eat all your safety margins immediately.

That's my theory and if I'm right, it'll cost nVidia a pretty penny to fix it as it needs a full re-design.

Regards.

That was my thought the moment I saw that it was all wired to a common bar. If we add this design to the penchant for cheap third party manufacturers to skimp on QA, along with the adapter being a user serviceable component in a tight enclosed space, we get the potential for a bad adapter getting into a users hands that then causes stress while trying to make it fit, and immediately jacks up the power to max and an hour or two later plastic starts to melt. It would be totally random and hard to reproduce unless someone was testing thousands of adapters.

 

[InvalidError]

Can you please read my statement in full before responding. A broken connection is actually preferred, and likely what was tested against.

My point was that there is almost no possibility that a bad connection would cause the connector to melt when multiple parallel connections exist as long as the other connections are still good. Most of he current will shift to the remaining good connections the instant you introduce even 10mOhm of added resistance on one wire since that almost doubles the path's total resistance.

Like I said before, no amount of media cover or shade is going to shift this to "customers are holding it wrong".

You may want to ponder the consequences of raising a stink about it only to ultimately get told the relevant authorities confirmed it was user error. That could spell the beginning of legislation to require licensed electricians for installing PC hardware.

So far, connectors not being plugged all the way in appears to be the only constant across all known incidents we have before-removal photos of.

 

[palladin9479]

You may want to ponder the consequences of raising a stink about it only to ultimately get told the relevant authorities confirmed it was user error. That could spell the beginning of legislation to require licensed electricians for installing PC hardware.

So far, connectors not being plugged all the way in appears to be the only constant across all known incidents we have before-removal photos of.

This is silliness in the extreme. This is likely bad adapter design pure and simple, and thinking that Igor can't plug in a cable is ... well no reasonable discussion is possible after that.

 

[InvalidError]

This is silliness in the extreme. This is likely bad adapter design pure and simple, and thinking that Igor can't plug in a cable is ... well no reasonable discussion is possible after that.

Has Igor commented on his first-hand experience plugging in the HPWR connector? Not that I know of, he merely commented on public accounts he was aware of. Until that changes, any discussion of his (in)ability to plug a connector in all the way is irrelevant.

 

[TJ Hooker]

This is the first time I see the diagram and I have a very plausible, yet hard to prove, theory with that way of wiring it. If I'm right, the engineers that approved that "daisy chain" of power need to be executed via electrocution.

So, when you solder or even just pressure-attach cables, there's always potential for expanding and contracting the solder, which can cause micro-fractures and micro-pockets of air/dust creating resistance. If you have DC going up and down, then it means you'll be thermally cycling the weakest link in that stupid daisy chain.

So, in short the key aspects are:

• solder quality or attach pressure
• temp deltas (via power deltas)
• time of the deltas (ramp up / down vs temps)
• micro-fractures

I've seen this at a higher level via material fracture for big mining operations (don't ask how I know xD) and when they crack, it's super visible and dangerous, yet they can still operate to their "rated capacity", but you obviously eat all your safety margins immediately.

That's my theory and if I'm right, it'll cost nVidia a pretty penny to fix it as it needs a full re-design.

Regards.

What daisy chain are you referring to?

 

[newtechldtech]

The power plug should be designed by default not to function at all when not fully inserted. and such plugs exist!

Nvidia Fault AGAIN ! $1600 and using simple stupid plug design.

 

[InvalidError]

The power plug should be designed by default not to function at all when not fully inserted. and such plugs exist!

Nvidia Fault AGAIN ! $1600 and using simple stupid plug design.

It is the PCI-SIG CEM 5.0 plug and Intel's ATX 3.0 HPWR plug too.

Unless the PCI-SIG, Intel and Nvidia decide to recall the HPWR connector, this thing will eventually end up on $200 GPUs, so the plugs need to be kept low-cost. While I can imagine many ways of enforcing (close-enough-to-)fully inserted before allowing power, I cannot think of any example in daily life of a connector that actually does.