It's happening again, melting 12v high pwr connectors

[ShrimpBrime]

My bad. Didn't actually read them. I typed in "3090 power melt" that those came up.

You guys can dogpile all you want, I know damn well there was a problem with them as I personally RMA'd two of different cards, one EVGA and one Gigabyte. I was not alone.. The problem wasn't as widespread as the 4000 series, but it started with the 3000s.

No worries. This was talked about a bit at OCN. Worth a read maybe. But these where 8 pin connectors I mentioned 2 dozen pages back already.

CableMod PCI-E cable connector melted in my RTX3090 :(

www.overclock.net

 

[lexluthermiester]

No worries. This was talked about a bit at OCN. Worth a read maybe. But these where 8 pin connectors I mentioned 2 dozen pages back already.

CableMod PCI-E cable connector melted in my RTX3090 :(

www.overclock.net

Exactly. Saw that post. There were melted connectors & cables for both types.

 

[x4it3n]

My bad. Didn't actually read them. I typed in "3090 power melt" that those came up.

You guys can dogpile all you want, I know damn well there was a problem with them as I personally RMA'd two different cards, one EVGA and one Gigabyte. I was not alone.. The problem wasn't as widespread as the 4000 series, but it started with the 3000s.

There is definitely an issue for sure, but the 3090 Ti didn't have that issue due to the power design as far as I (and even YT channels) are aware of.

When you compare the PCBs of 3090, 3090 Ti, 4090 and 5090 you can notice the difference right away :
The 3090 Ti is the only one having 3 shunts resistors that distributed power over all the cables the other ones had 2 (3040, 4090) or 1 (5090). And 3 was the right move apparently.

The AD102 and GB202 Engineering Samples even had 4x 16-pin connectors apparently :

https://x.com/9550pro/status/1889237955665125502

Why did they not even keep 2x 16-pin on the 5090 knowing that it has a 575W TDP and the cable is rated for 600W with a very limited safety factor compared to 8-pins, go figure...
Sure the PCB of the 5090 FE should have been changed to be slightly bigger but it would have been so much safer !

 

[lexluthermiester]

There is definitely an issue for sure, but the 3090 Ti didn't have that issue due to the power design as far as I (and even YT channels) are aware of.

When you compare the PCBs of 3090, 3090 Ti, 4090 and 5090 you can notice the difference right away :
The 3090 Ti is the only one having 3 shunts resistors that distributed power over all the cables the other ones had 2 (3040, 4090) or 1 (5090). And 3 was the right move apparently.

The AD102 and GB202 Engineering Samples even had 4x 16-pin connectors apparently :

https://x.com/9550pro/status/1889237955665125502

Why did they not even keep 2x 16-pin on the 5090 knowing that it has a 575W TDP and the cable is rated for 600W with a very limited safety factor compared to 8-pins, go figure...
Sure the PCB of the 5090 FE should have been changed to be slightly bigger but it would have been so much safer !

View attachment 387118

Fair enough. You kinda highlighted the very point some of us have been trying to make. The electrical math says it should work, but only in Ideal situations where perfect connections are maintained. Real life is something else. The connector needs to be over-designed, IE leaves room for error and less than ideal conditions. That's not the case. This "new"(latest) connector leaves no room for error and the result is unsafe, melting connectors and cables which are a fire hazard..

 

[jnv11]

The whole reason 12VHPWR exists is to be a single connector. I can't see Nvidia ever allowing dual 12VHPWR. I can't see them ever fixing it though either Melting issues are unlikely to affect anyone except 90 series owners and there's so few of them, they probably don't care.

Personally I don't understand why it's not a violation of some electrical standard.

Nvidia has allowed 2 of these connectors in Galax's GeForce RTX 4090 HOF as seen in https://www.galax.com/en/graphics-card/hof/geforce-rtx-4090-hof.html .

 

[lexluthermiester]

Nvidia has allowed 2 of these connectors

"Allowed" is the wrong word. NVidia has no say in AIB board designs. They can do almost whatever they want.

However, that is a better design. This is what I was talking about with @freeagent earlier. Spreading the load is the right way to design any high current power circuit.
This is what we're talking about;

 

[pavle]

"Allowed" is the wrong word. NVidia has no say in AIB board designs. They can do almost whatever they want.

However, that is a better design. This is what I was talking about with @freeagent earlier. Spreading the load is the right way to design any high current power circuit.
This is what we're talking about;
View attachment 387121

If they were honest, they would have written 375W x 2 (750W) max. power and not some fanatic adherence to the number of a man.

 

[Dredi]

My bad. Didn't actually read them. I typed in "3090 power melt" and those came up.

Heh

Are you kidding with that crap?

:cheers:

Did you find any?

 

[Gatt]

Right angled adaptor or right angled cable? its important to make this distinction as so often people are confused over it. The adaptor had the recall with some confirmed issues.

You are right, it was the angled adaptor. I checked mine just when it was beginning the melting. Surprisingly it was the V2 to have the problem, not the V1.

 

[Dr. Dro]

Sapphire Nitro+ 9070 XT confirmed to use the 2x6 connector

NITRO+ AMD Radeon™ RX 9070 XT GPU

Up to 3060 MHz, 16GB/256 bit DDR6. 20 Gbps Effective

www.sapphiretech.com

Hoping to see the PCB shots soon, see how they've implemented it on an AMD design

 

[wNotyarD]

Sapphire Nitro+ 9070 XT confirmed to use the 2x6 connector

View attachment 387231

NITRO+ AMD Radeon™ RX 9070 XT GPU

Up to 3060 MHz, 16GB/256 bit DDR6. 20 Gbps Effective

www.sapphiretech.com

Hoping to see the PCB shots soon, see how they've implemented it on an AMD design

As usual with Sapphire, there'll be fuses. Also, by their render:

Are those shunts between them fuses?

 

[MxPhenom 216]

As usual with Sapphire, there'll be fuses. Also, by their render:
View attachment 387235
Are those shunts between them fuses?

Yes there is shunts there as well.

 

[_roman_]

A fuse will hardly solve the possible problem. Especially on an amd card, when other same chip type amd graphic cards have the older connector.

I hardly remember when I changed or replaced the last fuse which I was the root cause for trigger it. A "car mechanic" ruined the fuse for the amp measurement range with a true rms multimeter in my family. Wrong measurement method in the first place. Current clamp - multimeter would have been the proper method.

 

[x4it3n]

"Allowed" is the wrong word. NVidia has no say in AIB board designs. They can do almost whatever they want.

However, that is a better design. This is what I was talking about with @freeagent earlier. Spreading the load is the right way to design any high current power circuit.
This is what we're talking about;
View attachment 387121

Finally an AIB GPU with 2x 16-pin connectors! I wonder how the other are going to react now... ASUS are going to launch a 5090 MATRIX for $5000 now lol

On the other hand "NVidia has no say in AIB board designs. They can do almost whatever they want." I'm not sure, because why would all the AIBs only use 1x 16-pin when they know that those connectors are still melting is unfathomable... Also it means all their models are limited to 600W, so Overclocking is pretty much useless too! Maybe NVIDIA lied to them too by saying "the 12V-2x6" fixed the melting issues, and all AIBs bought it just like everyone else. But I would have never allowed only 1x 16-pin on 5090s if I had been in their shoes.

 

[lexluthermiester]

Finally an AIB GPU with 2x 16-pin connectors! I wonder how the other are going to react now..

That one was a 4090, so it's been around a little bit.

I'm not sure, because why would all the AIBs only use 1x 16-pin when they know that those connectors are still melting is unfathomable...

Agreed. But if you look at all the various designs, NVidia can't be testing and qualifying them all.

 

[Dr. Dro]

That one was a 4090, so it's been around a little bit.

The funniest is the RTX 4080 version lol

GALAX RTX 4080 HOF LAB PLUS-X Specs

NVIDIA AD103, 2640 MHz, 9728 Cores, 304 TMUs, 112 ROPs, 16384 MB GDDR6X, 1400 MHz, 256 bit

www.techpowerup.com

 

[x4it3n]

That one was a 4090, so it's been around a little bit.

Oh okay. I wish it was a 5090 with Dual 16-pin lol.

Agreed. But if you look at all the various designs, NVidia can't be testing and qualifying them all.

Don't they send AIBs the Chips + whole PCB and then AIBs do whatever they want with it? AIBs should definitely make all their new GPUs with Dual 16-pin to avoid melting. Most of their High-End models have 2x 16-pin spots but only 1 is used...

 

[MxPhenom 216]

Oh okay. I wish it was a 5090 with Dual 16-pin lol.


Don't they send AIBs the Chips + whole PCB and then AIBs do whatever they want with it? AIBs should definitely make all their new GPUs with Dual 16-pin to avoid melting. Most of their High-End models have 2x 16-pin spots but only 1 is used...

They send GPU core on package. Its up to the AIBS for their custom cards to make the PCBA, etc. Nvidia still provides a reference design for PCB if AIBS want to use that for their "MSRP" models, but they dont have too.

 

[oxrufiioxo]

Rtx 50 series theme song

 

[Dr. Dro]

Rtx 50 series theme song

Unsure if "We are the Champions" or "Bohemian Rhapsody" applies to the RTX 5090, though

provided your order has been fulfilled (it likely hasn't), your card didn't catch fire, and has all working ROPs, you lucky lottery winner you

 

[R-T-B]

Those 12V2X6 cables with only 2 small wires are the only ones that I have seen burnt.

Mine has the full 4 wires. Have any of those gone up?

Those are sense pins indicating cable capacity, the "small wires." I think we may have some users reusing inappropriate cabling for a 5090.

 

[lexluthermiester]

The funniest is the RTX 4080 version lol

GALAX RTX 4080 HOF LAB PLUS-X Specs

NVIDIA AD103, 2640 MHz, 9728 Cores, 304 TMUs, 112 ROPs, 16384 MB GDDR6X, 1400 MHz, 256 bit

www.techpowerup.com

Yeah, that one is a bit goofy. In fairness though, for vertical mount it would be a decent show piece for a buyer that wants that kind of thing.

Unsure if "We are the Champions" or "Bohemian Rhapsody" applies to the RTX 5090, though

It really doesn't. I was optimistic about this launch, but it's turned out to be a very bad showing from NVidia.

Don't they send AIBs the Chips + whole PCB and then AIBs do whatever they want with it?

NVidia sends the dies and make reference PCB designs and specs available. AIB's are free to make them however they wish. Thus the wide variety of styling, one brand to another and one model to another.

 

[R-T-B]

NVidia sends the dies and make reference PCB designs and specs available. AIB's are free to make them however they wish. Thus the wide variety of styling, one brand to another and one model to another.

This is true but they are also free to (and do) set stipulations to getting said dies.

It's one reason why you don't see any cards with old style PCI-e connectors, even where they could be used.

 

[EsaT]

However, that is a better design. This is what I was talking about with @freeagent earlier. Spreading the load is the right way to design any high current power circuit.

In short halve the current per contact, and power loss per contact drops to quarter.

Because:
P = V x I
And
V = I x R
Getting rid of that not usefull voltage part:
P = I² x R

Though that still helps only assuming current is balanced between individual contacts and doesn't make this connector any less of Charlie Foxtrot.
That would happen only by de-rating connector by like one third.

 

[Werwulf]

Sapphire card has the Meltex connector on the top under a removable cover. They apparently use fuses to protect over-current on the cable... why?? Who wants to RMA their card if the cable becomes unbalanced? Why not just monitor with shunts and act on an imbalance with software?? Maybe it does that too? Weird.