NV 12VHPWR issues revisited

[MrTeal]

https://www.msi.com/news/detail/MSI-Unveils-the-World-s-First-ATX-3-1-Ready-Motherboard--Designed-for-Next-Gen-AI-Computing-144544

DISCUSS!

I don't think I've seen this mentioned anywhere. Is only MSI using this? Maybe they figured it IS crazy to rely on a single cable to provide all the power so they took the initiative to show the way to the rest of the industry? Or should we look for "ATX 3.1 ready" motherboard with supplemental PCIe power connector next time we are shopping?

We've had motherboards for decades with supplemental power, it just used to be a 4 pin molex instead of an 8 pin PCIe connector. It probably was more useful in the past than now when people were running a couple GPUs along with other PCIe cards.

I smell crap from start to end. Good luck overclocking your fans to use 100W.

LOL... I'm pretty sure my cooling system can pull 100W running full out though it's not on the MB. A couple D5 and a dozen fans plugged into the MB instead of a controller and I bet you could get up there pretty easily.

Still, the biggest use of something like that would be someone using a slot powered card like an RTX4000 Ada SFF. 70W for the GPU, maybe another 15-20W for a RAID card, all the other 12V the MB needs and then a few system fans and all of a sudden that's a pretty good draw from the two 12V wires on your 24pin.

 

[ToTTenTranz]

If I sell you a power connector, and it's rated for to handle 10 amps, that means I've already done the safety factor analysis, testing etc.

You can use it right up to 10 amps.

Doesn't matter if the power connector is rated for 50 amps. If the female connector or the cables are rated for 2 amps, the connection's rating is 2 amps.

There is absolutely ZERO need for you to underrate it, for safety factor. That's already built in.

No one's underrating anything. The safety factor for the 12VHPWR connector is 1.15 as per PCI-SIG's own documentation. It means that connector can simply pass the certification test if it handles just 15% more than the rated current. It's absurdly low.

I have no idea how something like this was ever accepted by regulators.

We have massive industry of connectors and switch, relays etc, that all have power ratings, that you can use 100% of because they already have engineered in safety factors.

Industrial connectors typically have a safety factor of 2. All the cables, connectors, power circuits etc. support twice the rated current.

 

[Fallen Kell]

No one's underrating anything. The safety factor for the 12VHPWR connector is 1.15 as per PCI-SIG's own documentation. It means that connector can simply pass the certification test if it handles just 15% more than the rated current. It's absurdly low.

I have no idea how something like this was ever accepted by regulators.



Industrial connectors typically have a safety factor of 2. All the cables, connectors, power circuits etc. support twice the rated current.

Exactly! And to add some common sense on this, your home's electricity and circuits all have ratings as well. For example the standard 15amp circuits in USA homes will typically use either 15AWG or 14AWG wire. While that 15AWG is rated for 15amps at 120V, it will easily carry 25+ amps before it really fails (i.e. becomes hot enough to melt the insulation and/or start a fire). This is why you will see many homes that even have 20amp circuits on lines that used 14/2 romex (two 14AWG wires + ground, yes, against code, but you would be surprised how many times you will run into this, as 20amp should have a min of 12AWG wires according to code).

Heck I think there is a youtube vid of an electrician testing the various methods of connecting electrical wires to see what is better with 14/2 romex (i.e. Wago connectors, knock-off Wago connectors, twist connectors (testing both pre-twisting the wires and just jamming them in the twist connector and twisting the twist connector on)). Then he measured how many amps he could put through the cable before it made it to the point that it could no longer keep itself cool and turned into a runaway heat loop of getting hotter over time without increasing any power being sent to the cable (you can test and measure this, as it turns out that copper cable increases in resistance as it gets hotter, and there is a point in time when you can then watch as the resistance continues to increase without increasing the current). Most of these went to over 50amps before failing, for something that's official rating is only 15amps...

 

[IEC]

Another one bites the dust:

While it’s tempting to blame the MODDIY cable, Spanish YouTuber Toro Tocho has experienced the same burnt cable (both at the GPU and PSU ends) with an RTX 5090 Founders Edition while using a cable supplied by PSU manufacturer FSP. Plastic has also melted into the PCIe 5.0 power connector on the power supply.

Sauce: The Verge

 

[Abwx]

Safety factor of 14% is crazy. What the hell were they thinking, using this to a >$2000 consumer electronics product?
Even 25% is low, this should be 50% or higher.

Generaly that s 30% for consumers dedicated products, FI a laptop that pull 50W at most require a 65W PSU.

As to why they used 2x 6 pins that s likely a marketing decision, because such
a small connector instillate the impression that the card is not that power hungry
and is well below its rated TDP, while a 2x 8 pins create the opposite perception.

 

[Jan Olšan]

It's actually the 65W charger itself that has the extra safety margin, even if it won't let you draw more than 65W from it. The components will be sized as if the wattage target was higher.

 

[Abwx]

Of course since it s the component that is stressed, likewise if the 2x 6 pins connector max power is rated 650W continuous then it should be capable of 850W continuous power delivery.

 

[DrMrLordX]

I'm not sure how anyone can keep defending this connector, or the companies involved, for continuing to use it as is.

As to why they used 2x 6 pins that s likely a marketing decision, because such
a small connector instillate the impression that the card is not that power hungry
and is well below its rated TDP, while a 2x 8 pins create the opposite perception.

From what I'm reading (in the videocardz comment section and elsewhere), it appears as though some PCB designs simply don't leave much room for power connectors, period. Such as what you'll see in the 5090 FE where there is an angled connector. Of course, can you defend the decision to produce such a PCB design? I wouldn't. And the 4090Ti somehow had a design that featured two connectors, so . . . why didn't that become standard for high-end dGPUs?

 

[coercitiv]

Post is now flagged as misinformation, because same user bragged about running at max OC forever:

[Later edit] Folks, watch from the 13 minute mark. der8auer tests with his own card, Corsair PSU + Corsair cable. On one of the wires he gets temps of over 125C near the PSU connector and 22A current. The plot thickens.

der8auer has the card, cable and PSU from one of the burning incidents:

I'll try to watch during launch break.

 

[amenx]

[Later edit] Folks, watch from the 13 minute mark.

der8auer has the card, cable and PSU from one of the burning incidents:

I'll try to watch during launch break.

Something seriously wrong here. One of the cable strands melted. Further testing found that the load was not being distrusted equally in the cable wires and one measured 22a, another wire also high while the others were cool. How can something like this happen?

 

[In2Photos]

I just watched the der8auer video.

The user with the 5090 has many years of experience building computers. He was also very aware of the connector issues and was diligent in checking the connections. It wasn't mentioned if he did any overclocking or not. The PSU and cable are both 12VHPWR, not 2x6. But they were both plugged in securely.

The connector on the GPU side had 1 12V pin damaged. On the PSU side 5 of the 6 12V pins were damaged. One of the wires in the cable was melted almost from one side to the other. This was a cable from moddiy so it's only about 8 inches long.

Der8auer also noticed a problem on his 5090. During testing he noted that a couple of the wires were very hot while the others were cool. Here are some thermal images he took while running furmark for only a few minutes.

PSU side 150C!!!


GPU side reached about 90C.

Here's a shot showing just 2 of the 12V wires heating up.



He then measured the current draw on all of the 12 volt cables. 2, 3, 5, 8, 11, and 22!!! That is some serious imbalance. He did mention that the astral card has some sort of sensing capability that shows the current for each wire that the FE does not have. He also said that the 90 series cards should have two connectors.

 

[IEC]

The Astral cards implement more shunt resistors and sensing circuitry, which most 4000 and 5000 series cards do not. But they're also 60% overpriced, so...

Not like nVidia could have spent a few dollars on some shunt resistors and sensing circuitry to make sure power is evenly distributed to prevent this thermal runaway scenario. Nah, how could a multi-billion dollar company spare the change? /s

 

[MrTeal]

The cable is the same for both 12VHPWR and 12x6, the change was exclusively on the cards. That was the only feasible way to implement it anyway, it's easy enough to make a BOM change to GPUs going forward and much harder to have people replace PSUs they bought 6 months ago.

Per pin sense would be a nice safety feature, though adding more sense generally reduces efficiency as well. It's not going to help for load balancing though, that's a much more complicated challenge. The simplest way would be to split them up into individual 12V domains, so if you had say 15 phases for core, 2 for mem and 1 for others you'd run 3 phases per 12V pin and that gives you load balancing. That doesn't fix everything though and can make other issues worse, like if one pin is poor and has high contact resistance, rather than the high resistance causing it to carry less current you're still forcing 1/6th through it causing more spot heating and further deterioration.

 

[coercitiv]

did somebody ask for a new Buildzoid video? No?! Here it is, enjoy the rant.

TL;DW
3090TI - connector internally treated as 3x 12V sources
4090/5090 - connector internally treated as 1x 12V source

 

[Krteq]

I just watched the der8auer video.
...

Wow, more than 150°C on connector and 22A through a single wire.

Nicely done NV, splendid engineering on premium $2K product

 

[In2Photos]

did somebody ask for a new Buildzoid video? No?! Here it is, enjoy the rant.

TL;DW
3090TI - connector internally treated as 3x 12V sources
4090/5090 - connector internally treated as 1x 12V source

Great explanation of what is going on! This is clearly a GPU design problem, compounded by a connector that provides very little overhead compared to 8 pin PCIe. There is no load balancing any more so it's possible to draw 600W through a single wire on the FE model! Shame on nVidia!!!

Here's a layout of power delivery on GPUs.

 

[AdamK47]

I was using the same power setup as der8auer in his burnt cable video with my 600W 4090. Same AX1600i (still one of the best years after its release) and the same 12VHPWR cable.

Ran many full load stress tests at 600W in the nearly two years with it.

As is evident in my previous photo, no damage.

The guy with the cute tiny Asus power supply with a 12VHPWR to 12VHPWR connector was running incredible load through only one of the wires. Not a safe setup for 600W.

Maybe it was a good thing I broke the clip on my Corsair 12VHPWR cable. Now I have four wires to spread out the load since the half octopus adapter requires me to use four PCIe Type 4 cables to my AX1600i.

 

[Aapje]

@AdamK47

The guy with the cute tiny Asus power supply with a 12VHPWR to 12VHPWR connector was running incredible load through only one of the wires. Not a safe setup for 600W.

But it's the card's design that allows it to run with the entire load going over one wire! And nowhere does Nvidia specify how you can run it in a safe way and that he was not doing so. This is just blaming the victim.

And frankly, there is no safe way too run it, because safety margin is so low. An invisible defect in the connector (or the card or the wire) can be enough to cause this to happen.

 

[igor_kavinski]

The only damage I have gotten from a GPU in my entire life was almost getting burnt when I touched my Voodoo3 3000's passive heatsink.

What has happened to gamers since???

By the way, I'm so glad that Nvidia is continuing to entertain us with the connector saga

 

[Heartbreaker]

[Later edit] Folks, watch from the 13 minute mark. der8auer tests with his own card, Corsair PSU + Corsair cable. On one of the wires he gets temps of over 125C near the PSU connector and 22A current. The plot thickens.

der8auer has the card, cable and PSU from one of the burning incidents:

I'll try to watch during launch break.

Good Video. I hope he actual investigates the cause. Usually we only see the melted aftermath which makes it hard to determine the actual cause.

With that massive current imbalance, there is obvious a high resistance source somewhere that he could easily find if he investigated.

 

[DrMrLordX]

He also said that the 90 series cards should have two connectors.

Without proper load balancing, adding a second connector won't help that much. You could still have two connectors each with one wire carrying 11a (or worse, potentially), which is still well over spec (which, if I recall correctly, is 7.5a or something close to that). That being said, having two connectors is still far more sane that using one!

Also based on the simplified diagrams provided by Buildzoid, it definitely adds some context to what may have been happening to a lot of 4090 owners with melted cables: improper load balancing across power pins.

 

[Det0x]

Quick testing in OCCT @ 600w with a ASUS 5090 Astral
AMP spread over the pins work as it should 🧐
GPU 12volt drops to ~11.8v average


Glad i went with ASUS time time around which is the sole vendor that have taken these problems with the connector serious
(only card with software monitor for amp draw over the pins)

PSU is a dark power 1000w
Have also bought the later released 12v-2x6 / 12vhpwr 90° cable pci-e for this psu

 

[In2Photos]

Quick testing in OCCT @ 600w with a ASUS 5090 Astral
AMP spread over the pins work as it should 🧐
GPU 12volt drops to ~11.8v average

View attachment 116857
Glad i went with ASUS time time around which is the sole vendor that have taken these problems with the connector serious
(only card with software monitor for amp draw over the pins)

PSU is a dark power 1000w
Have also bought the later released 12v-2x6 / 12vhpwr 90° cable pci-e for this psu

Having to run an additional piece of software to make sure that your GPU works correctly isn't a step in the right direction.

 

[poke01]

Quick testing in OCCT @ 600w with a ASUS 5090 Astral
AMP spread over the pins work as it should 🧐
GPU 12volt drops to ~11.8v average

View attachment 116857
Glad i went with ASUS time time around which is the sole vendor that have taken these problems with the connector serious
(only card with software monitor for amp draw over the pins)

PSU is a dark power 1000w
Have also bought the later released 12v-2x6 / 12vhpwr 90° cable pci-e for this psu

and how much did you pay extra for this projection AIB model which should have come standard on the $2000 FE?

 

[Det0x]

and how much did you pay extra for this projection AIB model which should have come standard on the $2000 FE?

I bought both my 5090's at stock (spesial) launch msrp. (got the TUF for 2k $ + norwegian tax)
Fully agree with everyone that is very slimy to raise the prices right after the first batch of cards got sold..
I'm just happy i manged to get both my orders in, without using any bots or paying scalpers