Nvidia 4090 meltdown maybe an answer

That's not the issue tho.
The generalizations and theories that are all inaccurate is what techJesus was complaining about at the end of the video.

Watch the whole thing. They ran over 600W thru just 2 pins/wires and had no issues, no overheating, and no melting.
The real issue is 2 things. 1) the bumps in one of the connectors where metal shavings are rubbed off and basically provide a 'filament' like in an old incandescent bulb, that will heat up, and 2) poorly seated cables.

#2 is where the design could have been better. And very likely, they can just update the male half of those connections (the wire side connector) to provide better seating in the female side, plus a better latch, and those issues will be much rarer. Technically I guess it's the female side, since there are pins in the connector on the card. The metal pieces just need some re-engineering, and the male plastic jacket around that metal female connector can be thinned a few thousanths of an inch, and they will slide in easier. Latch re-design wouldn't hurt either. All of this can be done to the wire side connector. The cards shouldn't need any change, and the adapters can be made more reliable as well as more idiot proof. If you got one right now with the original design, just be careful to fully insert it (always get permission first), and no sharp bends (ouch).

It's only happened to less than 0.1% (estimated between 50 and 100 out of 100,000 unit sold), so pretty uncommon. But that count is still too high. Redesigning the male side should clean it up. I also think nVidia should have included a 90degree adapter with every card.

Oh btw 600W =! metric shit ton, but it is heavy for most nerd types.


So... I'm wrong but you agree with me?

BTW 600W through that connector is a lot for us electrical engineer types. It's not a failure waiting to happen, but the fault tolerance is uncomfortable. They need to up the QC, up the quality of the materials, and improve their usability.

I would fix this problem through a few changes. First, I would make the connector a different colour, red with black pins or some such, that way the end user can easily see if it's seated properly. If the red doesn't touch the red you know the connection isn't complete. I would NEVER bury the connector in the GPU shroud where it's difficult to see. In a perfect world I would make the connector 50% bigger and make most of these issues go away.

Of course, the filthy, smelly, dim witted, nose picking, messy haired, skinny jeans wearing designers would never allow a big, red connector to bruise the aesthetic of the card. Fuckers.

Wait until the real shitters making back alley knock-off connectors start selling their wares.
 
So... I'm wrong but you agree with me?

BTW 600W through that connector is a lot for us electrical engineer types. It's not a failure waiting to happen, but the fault tolerance is uncomfortable. They need to up the QC, up the quality of the materials, and improve their usability.

I would fix this problem through a few changes. First, I would make the connector a different colour, red with black pins or some such, that way the end user can easily see if it's seated properly. If the red doesn't touch the red you know the connection isn't complete. I would NEVER bury the connector in the GPU shroud where it's difficult to see. In a perfect world I would make the connector 50% bigger and make most of these issues go away.

Of course, the filthy, smelly, dim witted, nose picking, messy haired, skinny jeans wearing designers would never allow a big, red connector to bruise the aesthetic of the card. Fuckers.

Wait until the real shitters making back alley knock-off connectors start selling their wares.
I'm sure PCI-SIG is just full of nose pickers. Those guys and all the companies that signed on to this are obviously less intelligent than you. I think you should take over for all of them.
 
I'm sure PCI-SIG is just full of nose pickers. Those guys and all the companies that signed on to this are obviously less intelligent than you. I think you should take over for all of them.
I was using sarcasm. Mostly. I wasn't referring to the PCI standard, I was referring to the design of the (admittedly great looking) 4090.

Design teams are filled with artisy fartsy smooth brains that don't give a rats shiny left keister cheek about the function of their precious aesthetic. The connector on the 4090FE is a pure case of form over function. It's designed to compliment the aesthetic of the card when it's on a shelf or in a box. Once the card is running it will have a bunch of shit hanging out of it. Out of the middle of the damned card. It's hard to get at the latch and it barely fits in even large cases once the card is plugged in. It's carefully hidden in the black region of the shroud making it harder to see and, more importantly, harder to actually use.
 
1668850129576.png
 
BTW 600W through that connector is a lot for us electrical engineer types.
I remember reading somewhere that the wires on the adapters are 14Awg not 16? Those can carry up to 20 Amps <- This Ampacity rating has built in safety margin. It will be less for stranded (vs solid core), but still more than enough. At 600W total card power, -75W provided by the PCIe slot = 525W/6 paths = 87.5W per wire /12V = 7.3A per wire. Hell lets just do 100W per wire/12V = 8.33A. 14 AWG Ampacity = 20A, 16 AWG Ampacity = 17A. We got some wiggle room for the wiggle room. The melted connectors (none of which caused 'Fire') requires a loose connector strained in such a way that only a single connection in the adapter is making partial contact. So for 400W/12V = 33.3A in 1 wire (for those not OC'ing, up to 50A for a 675W OC), it is out of spec, but only the small point of contact gets dangerously hot (up to 300C). The wire is still ok, and none of these instances have ever shown (to my knowledge) the wire jacket melting or even getting close to melting.

They just need to make the connector more idiot proof.
...First, I would make the connector a different colour, red with black pins or some such, that way the end user can easily see if it's seated properly...
I think you are onto something.. just reverse the colors. So make the plastic the directly surrounds the female metal parts a Red plastic, but the main housing Black.
So you just look at the plugged in connector, if you can still see Red, keep on pressing.

The latch could probably use some improvements and beefing up too, but I am no mechanical engineer.
Someone else suggested a slight change to the sense pins, where they wouldn't connect until the connector was all of the way in. This alone would go a long way, and if the connector started to come loose, the sense pins would be the first to unplug and tell the card to shut down.

A lot of simple redesign ideas, all of which affect only the wire side of the 12vhpwr connector. Already manufactured cards would need no modifications. This is all easily doable.. the mfr of that connector should be looking into these ideas.
 
I remember reading somewhere that the wires on the adapters are 14Awg not 16? Those can carry up to 20 Amps <- This Ampacity rating has built in safety margin. It will be less for stranded (vs solid core), but still more than enough. At 600W total card power, -75W provided by the PCIe slot = 525W/6 paths = 87.5W per wire /12V = 7.3A per wire. Hell lets just do 100W per wire/12V = 8.33A. 14 AWG Ampacity = 20A, 16 AWG Ampacity = 17A. We got some wiggle room for the wiggle room. The melted connectors (none of which caused 'Fire') requires a loose connector strained in such a way that only a single connection in the adapter is making partial contact. So for 400W/12V = 33.3A in 1 wire (for those not OC'ing, up to 50A for a 675W OC), it is out of spec, but only the small point of contact gets dangerously hot (up to 300C). The wire is still ok, and none of these instances have ever shown (to my knowledge) the wire jacket melting or even getting close to melting.

Snip for brevity.

The wires are practically never the problem in an circuit unless damaged. The conductance of the entire circuit is going to be crippled by its weakest point. In this case it's the contacts in the plug, namely debris or incomplete connections. That's why we always over-engineer the user end of any given piece of equipment. Or at least we should.

I truly believe PCI-SIG needs to go back to the drawing board on this connector, namely the latch and the separation of the pins, like you said. I fear the arrival of the genuinely shit knock-offs that will arrive with bargain basement 6000W 80+Pratnumgoldina PSUs that will come out of the back alleys in China. I mean, there's part of me that thinks the people who buy them probably deserve their house fire, but you know. I'm trying to have empathy for silly people.
 
Wonder how many people here claiming "user error" are the same people that mocked Apple for "you're holding it wrong." Now fragile parts is an expensive paid privilege that separates the "enthusiasts" from the "common users." It's like people I have talked to in the past complaining about how terrible Windows was for the computer industry because it made GUIs mainstream, bringing a flood of new potential customers for which there was a mass "dumbing down" of software/applications that used to be "much better" when they were command-line-only in MS-DOS. Now we are happy that the latest PC components are getting harder to install without burning your house down.
 
Isn't that how we got here though? Instead of dismissing stupidity we constantly reward mediocrity, and so this type of thing will continue to happen?
Give them a trophy or ribbon for participating and call it a day is the world we live in now unfortunately.
 
I doubt any members here experienced this. I'd like to know if there are or were any [H]ardForum members who are having this happen to them.
 
There have been products recalled for less failures and less hazardous failures than that.
 
50 out of 125,000 cards. A textbook case in how the internet always makes mountains out of mole hills.
This one was extra mountainous because a lot of people already had an axe to grind, a lot of people upset with pricing and availability, a lot of people "outraged" that didnt even own a 4090, and nevermind the youtubers and blogs that depend on controversies for traffic and ad revenue - "when none exists, will find it financially necessary to create one". So this one was a gift for the "Nvidia just needs to be taken down a notch" group.

But mindshare inertia + continuing to top FPS charts and NV will probably continue to be impervious to most internet flareups. But they don't get off scotfree - adaptergate made the rounds on investment news and made casual observers wonder.
 
Last edited:
This one was extra mountainous because a lot of people had an axe to grind anyway, a lot of people upset with pricing and availability, a lot of people "outraged" that didnt even own a 4090, and nevermind all the youtubers and blogs that depend on controversies for traffic and ad revenue. So this one was a gift for the "Nvidia needs to be brought down a notch" group.

But with mindshare inertia + continuing to top FPS charts, NV will probably continue to be impervious to most internet flareups.
I always feel like internet rage isn't even considered a threat by a lot of companies. Usually a small group of loud people that dont actually affect sales as much as they think they are. xbox red ring of death still sold a ton of xbox 360s. I wouldnt be surprised if people raging about this are still buying the card.
 
"when none exists, will find it financially necessary to create one"
This. All it takes is a few possible problems to get everyone in an uproar and go head hunting. I lost some respect (not that I had a lot mind you) for a few Youtubers that just went along with everything and did nothing but add fuel to the fire.
 
I noticed that the vocal commenters have now moved on to raising the alarms that the 16-pin connector design is so flawed that it requires way too much force to seat properly in order to deflect the issue of user error.

In reality, the whole process inserting and double-checking for a flush connection wasn't any harder than how it was with 6-pin and 8-pin PCIE connectors in my experience, with only difference being was the audibility of the click of the latch. Nowhere near as difficult to insert fully as with those old-school 4-pin molex power connectors for HDDs and add-in cards. :p
 
Last edited:
There have been products recalled for less failures and less hazardous failures than that.
If the recall just involves swapping out the adapter with one that has an updated design that reduces user error then I can see a voluntary recall.

If the socket on the actual cards would need to be updated/replaced to reduce the chance of user error then I highly doubt there will be any recall.
 
If the recall just involves swapping out the adapter with one that has an updated design that reduces user error then I can see a voluntary recall.

If the socket on the actual cards would need to be updated/replaced to reduce the chance of user error then I highly doubt there will be any recall.
I feel like a recall isn't going to happen. I believe the "Hey, check and make sure your connectors are fully seated!" has pretty much caused the problem to go away. Nothing new on Reddit since Gamers Nexus video.

Newer connectors will probably eventually have their signal cables shortened to avoid powering on the cards if not fully inserted.
 
I feel like a recall isn't going to happen. I believe the "Hey, check and make sure your connectors are fully seated!" has pretty much caused the problem to go away. Nothing new on Reddit since Gamers Nexus video.

The connector issue definitely needed a discussion - even if it ended up a slow moving PSA unfolding as a drama - and I believe most of the reports were probably legit, however the idea that one video pulls the battery out of the viral/bandwagon aspect to it is kind of bananas.

I don't know for a fact Jayz2Cents sent a meangirl "nobody likes you" tweet to GN Steve with a burner account after the video aired, I just know that it's true.
 
Back
Top