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

[Chrispy_]

I think he writes the reviews for igorslab? Or am I wrong? And he always advertises there for cybernetics there. I refer to the german igor website version.

The Story Behind Cybenetics: Past, Present & Future - Hardware Busters

Hardware Busters - The Story Behind Cybenetics: Past, Present & Future - Freestyle

hwbusters.com

Aris is becoming somewhat of a celebrity in PSU circles. For many of us he filled the vacuum that John Gerow of Jonnyguru left when he went to work for Corsair. OklahomaWolf kept Jonnyguru running for a while and did good work, but where John Gerow was instrumental in highlighting good and bad PSUs in the PCs transition from CPUs to GPUs being the dominant load in a system, Aris is aiming to improve the certification system since 80Plus is such a useless standard for defining quality or capability these days. It's hard to make a bad 80 Plus Titanium or Platinum PSU, but 80 Plus Gold is a minefield.

 

[pavle]

It will keep happening as long as companies keep writing 600W instead of 300W on this connector. Same would happen if they wrote 400W or 500W on 8-pin.

 

[Chrispy_]

It will keep happening as long as companies keep writing 600W instead of 300W on this connector. Same would happen if they wrote 400W or 500W on 8-pin.

I think I would place my bets on a 400W 8-pin instead of a 600W 12V6X2 simply because an 8-pin Minifit Jr is rated to 324W by Molex themselves. The 150W PCI-SIG rating is extremely conservative. Meanwhile, Amphenol's Microfit 9.5A per circuit of 12VHPWR/12V6X2 seems very optimistic given that it's more current per pin than the much larger and more robust Molex Minifit Jr.

I'm no electrician but the laws of physics say that a larger connector (MiniFit Jr, rated by Molex at 9A per pin) should be able to cover exponentially more current than a smaller connector (Microfit, rated by Amphenol at 9.5A per pin) given that it's the same metals we're talking about and all that matters to the laws of physics is the contact area which scales with the square of the size.

 

[Endymio]

I think I would place my bets on a 400W 8-pin instead of a 600W 12V6X2 ... I'm no electrician but the laws of physics say that a larger connector (MiniFit Jr, rated by Molex at 9A per pin) should be able to cover exponentially more current than a smaller connector (Microfit, rated by Amphenol at 9.5A per pin) given that it's the same metals we're talking about and all that matters to the laws of physics is the contact area which scales with the square of the size.

With apologies for being a stickler for accuracy -- Microfit Std is rated at 8.5A per pin, not 9.5, larger pins don't handle 'exponentially' more current (the correct term would be "polynomially"), there are many other factors besides contact area, and frankly I don't see the point of comparing a 400W connector to a 600W one. If my power needs are 600 watts, I'll take the 600W standard any day of the week.

 

[Chrispy_]

With apologies for being a stickler for accuracy -- Microfit Std is rated at 8.5A per pin, not 9.5, larger pins don't handle 'exponentially' more current (the correct term would be "polynomially"), there are many other factors besides contact area, and frankly I don't see the point of comparing a 400W connector to a 600W one. If my power needs are 600 watts, I'll take the 600W standard any day of the week.

The thing is, 12VHPWR and 12V6X2 aren't actually made by Molex, so even though Molex rates them at 8.5A per pin, the actual connectors used in GPUs and PSU cables are a Microfit-compatible connector called Minitek Pwr 3.0 manufactured by Amphenol with a datasheet current limit of 9.5A per pin.

The mating surface area of the male and female pins in the connector that carry the current is an area that has two dimensions, so as you scale up the size of the pins, so the area increases with the square of the size, the very definition of an exponential relationship.

 

[Endymio]

... so the area increases with the square of the size, the very definition of an exponential relationship.

I'm sorry, but this isn't correct at all. An exponential relationship is defined f(x) = b^x, whereas you're describing a quadratic (polynominal) relationship f(x) = x^b (with b=2 in this case). Exponential doesn't simply mean "fast".

In the mathematics of growth analysis, the spectrum is: linear, polynomial, exponential, superexponential.

The thing is, 12VHPWR and 12V6X2 aren't actually made by Molex, so even though Molex rates them at 8.5A per pin, the actual connectors used in GPUs and PSU cables are a Microfit-compatible connector called Minitek Pwr 3.0 manufactured by Amphenol

I'm not a power engineer, but it's my understanding "these connectors" are made by a variety of firms. This certainly seems to suggest Molex provides them:

https://www.mouser.co.uk/c/?marcom=185305453

 

[Chrispy_]

In the mathematics of growth analysis, the spectrum is: linear, polynomial, exponential, superexponential.

Okay, happy to be corrected. I've always described an non-linear relationship as exponential whether it's polynomial, exponential, or superexponential. In this case I meant polynomial, which is the relevant relationship for current-handling properties vs size

I'm not a power engineer, but it's my understanding "these connectors" are made by a variety of firms. This certainly seems to suggest Molex provides them:

They may do now, but I've already been corrected here and cited the official PCI SIG spec (probably the first, original thread here on the topic of 12VHPWR) for previously linking Molex MicroFit as the original 12VHPWR connectors that were melting were exclusively Amphenol with different per-pin ratings.

Edit:
Maybe I'm missing something but if Molex are rating their own MicroFit connectors with 8.5A per pin, explicitly calling it a 12V PCIe 5.0 connector, their own power rating doesn't match their own spec. 8.5A*12V*6 pins is 612W, not the 675W they describe it as.

 

[eidairaman1]

For safety 600W/12VDC=50A
50A/6 PEGPins=8.33A per pin, Round it up to 9 or 9.5A, per wire diameter gauge it would be 14 AWG at the largest per conductor, 16 AWG Minimum

The square contact (female conductor) would need its surface area and thickness figured out to support 9.5A safely, same with the square pin (which might just be 14 or 16AWG) for 24/7 366 use.

Ofc there needs to be current monitoring and regulation on cards and psus so the cards cannot draw more than 50A in total or 8.33A on each pin.

 

[qxp]

Okay, happy to be corrected. I've always described an non-linear relationship as exponential whether it's polynomial, exponential, or superexponential. In this case I meant polynomial, which is the relevant relationship for current-handling properties vs size

Usually when something is exponential or faster there is an explosion (as in 2 rabbits make 4 rabbits, 4 rabbits make 8 rabbits, and so on). Quadratic is pretty fast on its own and as an adjective.

And logarithmic is nonlinear but very slow, you can pretty much assume it is constant.

 

[londiste]

Ofc there needs to be current monitoring and regulation on cards and psus so the cards cannot draw more than 50A in total or 8.33A on each pin.

This is probably the key to this entire problem. 50A in total part is there. But 8.33A on each pin? Should it be done by card, PSU, both? The answer to that problem is definitely not apparent and there are different opinions.

 

[TheDeeGee]

Current regulation means cards would throttle if one or more pins arn't properly connected.

This tiny connector is simply a piece of shit and has to go.

 

[N/A]

660W on 12VHPWR to 2x8-pin cable.

My PSU exactly. I hope it holds up. The GPU itself can also increase heat and create additional buildup.

 

[Chrispy_]

Usually when something is exponential or faster there is an explosion (as in 2 rabbits make 4 rabbits, 4 rabbits make 8 rabbits, and so on)

Yeah, that was the mistake I made. I've been calling any non-linear behaviour "exponential" my entire life because I thought any function with an exponent >1 was "exponential". It's opposite, in my mind, was "diminishing"

Any non-linear relationship with an exponent >1 will explode (tend to infinity) at some point on the scale. f(x)=y^2 is a curve with increasing gradient, and that gradient will never stop getting steeper. That is not an exponential curve though.

 

[qxp]

Yeah, that was the mistake I made. I've been calling any non-linear behaviour "exponential" my entire life because I thought any function with an exponent >1 was "exponential". It's opposite, in my mind, was "diminishing"

Any non-linear relationship with an exponent >1 will explode (tend to infinity) at some point on the scale. f(x)=y^2 is a curve with increasing gradient, and that gradient will never stop getting steeper. That is not an exponential curve though.

Oh, I was just taking the opportunity to talk about growth curves The other one I did not mention is "sigmoid" - it has a shape as gaussian distribution. You see them in contexts when something switches levels. You get a flat line, an exponential rise, linear growth and then exponential convergence (with negative exponent) to the next level.

Arctan() also has sigmoid shape, but the growth is closer to polynomial, not exponential.

 

[Ruru]

It will keep happening as long as companies keep writing 600W instead of 300W on this connector. Same would happen if they wrote 400W or 500W on 8-pin.

I just can't understand why it's so hard to use two lower rated connectors instead of the 600W one. That would make them waaaaaay safer.

And personally I prefer the old PCIe connectors since they've proven working since 2004. Luckily Radeons still prefer the classic connectors with few exceptions.

 

[Endymio]

Any non-linear relationship with an exponent >1 will explode (tend to infinity) at some point on the scale. f(x)=y^2 is a curve with increasing gradient, and that gradient will never stop getting steeper.

To be precise:
- Any polynomial relationship with exponent >0 (not 1) will tend to infinity. Those with exponents <1, however, do so at a decreasing rate. In mathematical terms, they are not convex functions.
- 'Non-linear' doesn't imply an exponent. There are many functions (most of mathematics, in fact) which are non-linear, yet lack the operation of exponentiation.

An interesting feature of the log function is that, while it also tends to infinity, it does so slower than any polynomial function, no matter how small its (positive) exponent. In other words, ln(x) grows slower than all these functions:

x^0.1
x^0.01
....
x^0.000000000001

No matter how close that exponent gets to zero, the function eventually exceeds ln(x) (or log(x), if you prefer).

 

[LittleBro]

MSI’s yellow 12V-2x6 power connector turns black after burning on GeForce RTX 5090 graphics card - VideoCardz.com

Not so safe after all: MSI RTX 5090 SUPRIM turns yellow power connector black MSI was the first company to come up with the idea of a colored power connector for new graphics cards. The bright yellow color on the female side was designed to indicate how far the connector should be inserted into...

videocardz.com

I wonder how many black partially burned but not melted connectors are in use right now.

 

[SmookinJoe]

Just wanted to share..
Very funny when he was actually attempting to disconnect the cable 11.35 minute mark and beyond

 

[ir_cow]

Another 3rd party cable. This one from MOD-DIY. Seems like the most likely cause is the ends were crimped instead of soldered. High resistance on 2 pins caused the others to exceed 9a.

Once again, don't buy 3rd party cables...

 

[ty_ger]

I heard from some other very smart people on this forum that Roman (Der8auer) is a liar scumbag salesman.

Seems like the most likely cause is the ends were crimped instead of soldered.

What are you talking about? Show me the connector spec.

 

[ir_cow]

What are you talking about? Show me the connector spec.

Watch the video above my post.

 

[AsRock]

PC hardware used to be fool proof on everything related to power at least.
Not anymore I guess

Statistically mistakes (on way or the other) are happening everywhere and you can not relay on the human factor. Its a major principal in a lot of cases.
In this case when a connector is pushed close to the edge of what a physical connection of this size can sustain like the 600W, and there is little or no room for mistakes you have to force the user to do the right thing. You can't just hope for the best.
Add monitoring to each of the pin is a sensible thing to do, just like @buildzoid is suggesting in this video, especially on a $2000+ product.
I dont think he is unreasonable here.

Its simple... The second the GPU sees a deviation above a certain level between current on the pins it will prevent the user from running anything that can damage the card.
This is the simplest way to force the user to the proper connection.

But the 4090 and 5090 don't have such detection as i understand it, unlike 3090 and such.

Maybe some kind of load balancing is needed on the card or PSU side of things ?.

 

[phanbuey]

But the 4090 and 5090 don't have such detection as i understand it, unlike 3090 and such.

Maybe some kind of load balancing is needed on the card or PSU side of things ?.

either would work tbh.

 

[boppingXQWB]

Current balancing should be GPU side as it always has been for a good reason.

 

[dragontamer5788]

Another 3rd party cable. This one from MOD-DIY. Seems like the most likely cause is the ends were crimped instead of soldered. High resistance on 2 pins caused the others to exceed 9a.

Once again, don't buy 3rd party cables...

Soldering is actually considered worse today than crimping. Just an FYI.

EDIT: A proper crimper should cold-fuse the copper together. Meanwhile, solder is a mixture of metals, often lead + the original copper. The lead is more brittle than copper, and so the lead WILL fail before copper. Especially in cables. Solder is better on "solid" surfaces (like PCBs which should never move). But cables? Crimping 100% of the time. Copper-crimp-to-copper just will be more flexible and less brittle. No internal stress fractures.