PSA: Don't Just Arm-wrestle with 16-pin 12VHPWR for Cable-Management, It Will Burn Up

[Fasola]

The plot thickens: It appears there are at least 2 types of adapter, a lower quality one (Igor's) and a higher quality one (GN's). GN hasn't been able to trigger a failure with their adapters even after cutting the side cables.

 

[Ravenas]

I don't recall a graphics card where my cable bend how to be a certain length or the connector would burn and my graphics card would have damage. What an absolute mess.

 

[dj-electric]

I have ordered parts to assemble my own 12VHPWR cable. I have crimpers and all lab equipment to prepare and test it. Might update if people care

 

[Nater]

Now I'm a Design Engineer, CAD Tech, Machinist, CAM Programmer, etc etc...but I'm no electrical engineer, yet I see a simple SIMPLE solution staring them right in the face.

Update the PCIe slots and mainboard layouts. Quit trying to fit the large square peg in the small round hole. The card is already taking up 4 PCIe slots in most designs, so have it pop into 4 rigid PCIe slots and get yourself 4x 75w of power right off the top without changing much of anything.

And I know some of you will reply "but that will make it too expensive!"

 

[ARF]

but I'm no electrical engineer

Dr. Lisa Su from AMD - AMD's CEO is an Electrical engineer - she knows best.

Radeon RX 7000 Series Won't Use 16-pin 12VHPWR, AMD Confirms | TechPowerUp

 

[thegnome]

This all greatly reminds me of why i tell everyone not to use any adaptors or extensions on modern GPUs

These melted on an undervolted 3090, locked to under 250W
View attachment 267657


Native direct cables or risk a fire. Nvidia was dumb to force this new connector so early, they just didnt want the stigma of 4x 8 pin connectors on these GPUs

All depends on the extensions used, most cheap ones totally aren't good enough for any load, even under the official 150W. My extensions on my 250W GPU from CM (also 2x 8pin) has had zero problems with melting or heat at all.

 

[Denver]

The plot thickens:

lol AMD should do marketing on top of the problem, "Unfortunately, our GPUs still don't have the flamethrower / self-destruct function, guys"

 

[ARF]

lol AMD should do marketing on top of the problem, "Unfortunately, our GPUs still don't have the flamethrower / self-destruct function, guys"

Err, this is evil - just stop the sales already and return all (working and burnt) cards to Huang.

 

[HTC]

I think i figured out the REAL problem: the temperature of Huang's oven wasn't right for some of the cards ...

 

[hat]

Now I'm a Design Engineer, CAD Tech, Machinist, CAM Programmer, etc etc...but I'm no electrical engineer, yet I see a simple SIMPLE solution staring them right in the face.

Update the PCIe slots and mainboard layouts. Quit trying to fit the large square peg in the small round hole. The card is already taking up 4 PCIe slots in most designs, so have it pop into 4 rigid PCIe slots and get yourself 4x 75w of power right off the top without changing much of anything.

And I know some of you will reply "but that will make it too expensive!"

I can see a number of problems with this. First, you are correct that it would make the design more expensive. There may also be some challenges stacking multiple PCI-E ports to a single card, even if they're only there for power. That would make the card a fair bit more prone to breaking somewhere along all those PCI-E pins, and getting power to the main PCB would be a challenge in itself. Beyond that, you would need to ensure that your motherboard actually has 4 PCI-E slots all lined up in a row to fit such a card. And then you run into the problem of supplying up to 300w of power from the slots alone, something that burned up 24-pin ATX connectors in the past. We've had a few mentions of exactly that happening here on these forums when people were jamming many cards into a single machine for Foldind@Home. You would need motherboards to be specifically designed for this. Take a look at the cryptocurrency mining motherboards that had asstons of PCI-E slots for jamming as many cards in a single machine as possible... they all had tons of extra power connectors to feed the slots.

 

[Sisyphus]

Technical transition problem. Power consumption has reached the limit of existing connection structure. nVidia wanted backward compatibility in a highly fragmented, modular system preserved, bad decision. It did not work sufficiently reliable. Adapter are always a bad solution, as every additional mechanical contact brings in new contact resistance, mechanical issues, more than double the failure rate.
New standards must be used, old cables and plugs must be replaced. If old power supplies are made compatible using an adapter, the adapters/connectors must be certified, together with a correct installation without bending/mechanical stress on the plug.
600W is not difficult to connect, as long as you use cables/connectors originally intended for 600W and follow quality control. In the field of household appliances, 500-1500W are normal power ranges. No issue if the manufacturer installs all parts. A problem when customers use hundreds of different plugs and power supplies from different origin with very limited ability to check the quality before installation.

Typical early adopter issue. Presumably one will still improvise with the 4090, with the 5090 new power supply units/connectors will be necessary and no guarantee granted, using old cables/old or no-name adaptors.

 

[Aquinus]

they just didnt want the stigma of 4x 8 pin connectors on these GPUs

Now they have the stigma of melting power connectors instead. Well played, nVidia.

 

[nangu]

The horror has a face - NVIDIA’s hot 12VHPWR adapter for the GeForce RTX 4090 with a built-in breaking point | igor´sLAB

Those who are now beating up on the new 12VHPWR (although I don’t really like the part either) may generate nice traffic with it, but they simply haven’t recognized the actual problem with the…

www.igorslab.de

Igor just confirmed the Built by Nvidia adaptor has such a high quality, that the wires are held (only) by soldering onto a very very thin piece of metal , and can be broken with very little force.
The card doesn't even know one of the soldering is broken, cause all the pins are joint inside the adaptor so the pin didn't "disconnected" , the load just spread to other wires and pump up the Amps (and wire temps) , it just keeps going until it melts.

What a wonderful design !

Nvidia high quality overengineering!!

 

[Sisyphus]

Now they have the stigma of melting power connectors instead. Well played, nVidia.

All companies have the "stigma" of some disappointed users.

Nvidia high quality overengineering!!

Yes, 4090 is constantly sold out.

 

[cyberloner]

don't buy it~!

 

[nangu]

All companies have the "stigma" of some disappointed users.


Yes, 4090 is constantly sold out.

for you too!

 

[arsh666]

I think the main problem here is the card, Yes its fast, but horrible design, its the size of a xbox and draws way to much power which causes other problems like heat, and now failed power connector. As an electrician, you can do the math, and this card will pull 60 amps, normally for this type of connection we use #6 copper and lugs to connect. I would be surprised that this power connection meets UL safety standards.

 

[Sisyphus]

I think the main problem here is the card, Yes its fast, but horrible design, its the size of a xbox and draws way to much power which causes other problems like heat, and now failed power connector. As an electrician, you can do the math, and this card will pull 60 amps, normally for this type of connection we use #6 copper and lugs to connect. I would be surprised that this power connection meets UL safety standards.

Who buys a 4090 should invest 200-400$ more for a new power supply with proper cables/plugs and a midi/big tower with enough space, no adapters. Simple as that.

 

[ARF]

Update Oct 30th: Jon Gerow, aka Jonny Guru, has posted a write-up about the 12VHPWR connector on his website. It's an interesting read with great technical info.

I see the following illustration:

But I don't think the bend is that bad. How does bend exist when the connector has a locking mechanism which makes sure that the male-female parts allign ideally?

 

[Wirko]

which increases resistance on that terminal forcing the current to take the path of least resistance instead

That's how we've always been taught. May I suggest a better version, more clear to noobs and more adequate to professionals: [all] the paths of least resistance.

Technical transition problem. Power consumption has reached the limit of existing connection structure.

Looking a bit farther into the future: It's the technical transition to 24 or 48 volts, which seems inevitable in the longer term. 12VHPWR is already obsolete. 12VO for motherboards can carry 288 W and so it is equally obsolete. But the transition hasn't been done yet, and hasn't even started yet, because it would break compatibility with any existing PSUs.

 

[QUANTUMPHYSICS]

The REAL ISSUE I see is that PC design has changed and people are building computers for looks and show with form over function.

Secondarily, the power demands of the components have increased and there isn't enough communication between GPU makers, PSU makers and consumers/ modders.

This is a CATASTROPHE waiting to happen as many younger builders don't understand the dangers of running this much power through their cutesy PC cases without proper grounding, shielding and wire management.

I think the entire PC layout needs redigning.

 

[ARF]

Looking a bit farther into the future: It's the technical transition to 24 or 48 volts, which seems inevitable in the longer term. 12VHPWR is already obsolete. 12VO for motherboards can carry 288 W and so it is equally obsolete. But the transition hasn't been done yet, and hasn't even started yet, because it would break compatibility with any existing PSUs.

But the GPU die and memory dies run on around 1 volt, which would mean that the requirements on the voltage transformer on the PCB will be much higher.

 

[phanbuey]

This is just a complete garbage card design. Card is unnecessarily bulky; the absolutely unnecessary dongle connector is unnecessarily intrusive and protruding at 90 degrees. And now it's lighting on fire if you bend the cable incorrectly. What a waste of a phenomenal GPU.

 

[Sisyphus]

[...]
Looking a bit farther into the future: It's the technical transition to 24 or 48 volts, which seems inevitable in the longer term. 12VHPWR is already obsolete. 12VO for motherboards can carry 288 W and so it is equally obsolete. But the transition hasn't been done yet, and hasn't even started yet, because it would break compatibility with any existing PSUs.

Higher voltage or thicker cables/plugs. I am fine with both. 2-3 8 pin plugs will soon be gone for high-end GPUs. Even MB and case design will change, if 3-4 pci-slot GPUs become normal. But don't overdo it either, the problem only affects consumers who think they need a high-end GPU, that's at best 1-2% of all PCs. It is therefore quite possible that an isolated solution will be developed here.

This is just a complete garbage card design. Card is unnecessarily bulky; the absolutely unnecessary dongle connector is unnecessarily intrusive and protruding at 90 degrees. And now it's lighting on fire if you bend the cable incorrectly. What a waste of a phenomenal GPU.

Without the statistical data, complete error analysis, such conclusions are not possible. Always these dramatizations. Apart from that: Thanks to all early adopters.

 

[RogueSix]

I think the main problem here is the card, Yes its fast, but horrible design, its the size of a xbox and draws way to much power which causes other problems like heat, and now failed power connector. As an electrician, you can do the math, and this card will pull 60 amps, normally for this type of connection we use #6 copper and lugs to connect. I would be surprised that this power connection meets UL safety standards.

Do you own a RTX 4090? Doesn't sound like it. Because if you did, you would know that the RTX 4090 is a very cool running card. With some exceptions (ASUS ROG STRIX OC) it is also not *that* huge. I own the MSI Suprim X and it's a big card, alright, but by far not as crazy as the stupid clickbait YouTubers would like people to believe. It easily fit in my beQuiet DarkBase 900 (non-Pro) and there was also zero issue putting the side panel back on. I never had to bend the adapter or anything.

The card is a powerhouse but it remains cool exactly *because* of its sizable cooling solution. That is not the issue at all. We will have to wait and see what the official investigations by nVidia, the AIC partners and the maker of the adapter(s) turn up. The YouTubers are poking around in the dark for the clicks. They can not be taken serious. Let's wait and see for the results of the official investigation...