IDK if someone else has already posted a similar theory/hypothesis on the web, there are so many reports of this specific problem I am not going to waste my time searching the web to find out. I am not going to read all of those >900 posts in that locked thread "EVGA X58 759 Clasified Limited Edition #551 down..." either, so if I should know something posted please let me know. This problem is not limited to EVGA, nor X58 boards, so I am posting in "General hardware." I will post links to this in a few current threads in the "EVGA X58 SLI" forum. This thread is not for complaints to EVGA, nor arguments about 'who is to blame' for the problems. Please keep the posts oriented toward mitigating the problem, or proving/disproving my hypothesis only.


Foreword

Many people with enthusiast computers with high-powered SLI video nowadays are having a meltdown of one or two pins on their 24-pin ATX mainboard power connector (actually one of those pins is on the 4-pin add-on plug, or one could call it the 20+4 pin plug), particularly pins 10 and 11 (+12V). This has only begun happening since high-end video cards have started drawing a lot of power - like those that draw >100W. Said video cards have extra power connectors on them, which must be used to supplement the video card power, but apparently they are not supplying enough extra power in all these cases. It doesn't help that the sockets that carry this current have to be made very cheaply, meaning the manufacturers cannot afford very good quality control for them. And I have not seen any that are gold-plated for better conductivity/lower resistance. EVGA has issued a "powerboost" PCIe cable, that plugs in to a 4-pin molex power connector from your PSU, and into any available PCIe slot, to help mitigate this melting 24-pin ATX plug problem.

EVGA "powerboost" PCIe power adapter mod:


EVGA "powerboost" PCIe power adapter mod installed:

 
 
This is a theory, or dare I call it hypothesis, regarding the reason this is happening (without getting too deep into the consumer product reasoning) and perhaps a workaround to avoid such a problem, and a measure one could use to assess the likelihood of contracting said problem. I will concentrate only on the +12V supply in this hypothesis as it is those pins which are melting on the 24-pin ATX plug.


(E760 Classified Melted my PS 24 Pin Connectors)


Non-exclusive list of video cards affected:

This is a list of video cards presently known to elevate the risk of a 24-pin ATX plug meltdown, when two or more are installed on one mainboard.

ATI Radeon
"38XX top ends in Quad Fire" - per donta1979 10/6/10
4890 in Quad Fire - per donta1979 10/6/10
5830
5850
5870
5950
5970 (all 5900 series)
6850
6870
6950
6970 (all 6900 series)

Nvidia GeForce
295
450
460
465
470
480
570
580 (all 500 series)
 
This affects all motherboards designed for SLI/Xfire except those with the extra 4-pin molex plug(s) added for PCIe auxiliary power, like these:


Images courtesy of LionRed
 
I dont believe any of these boards with auxiliary PCIe power Molex plugs were distributed before 2010.
 
 
Non-exclusive list of EVGA motherboards affected (there are also other mfgr's mobo's affected):
E756 EVGA X58 Micro
E757 EVGA X58 LE
E758 EVGA X58 SLI
E759 Classified Limited Edition
E760 EVGA X58 Classified
E761 EVGA X58 Classified
E762 EVGA X58 Classified 4-way
E768 EVGA X58 FTW3
 

Reasoning (logic behind my hypothesis)

The root of this problem lies in the way electricity moves, and the use of plugs not designed to withstand such electrical current. The video cards draw 12V power primarily, and this power is available through both the PCIe bus and the extra card power plugs (6-8 pin). But electricity will follow the path of least resistance, thus meaning if a bad connection exists on the 6-8 pin plugs it will flow more from the PCIe bus, which in turn comes from the 24 pin ATX power plug - +12V pins 10 and 11. So bad connections could cause this problem, and I am not referring to the 24 pin plug exclusively. But all this theory assumes that the voltages supplied to each plug are exactly equal!


Image courtesy of kpablo (Re:How many Corsair HX1000's does it take?)
 
Present day consumer computer power supply units (PSUs) are made quite cheaply due to market competition. As you may know, the cheap units are generally much less reliable than the more expensive ones (there are exceptions). And most PSUs that are rated more than 650W use multiple "rails." These are basically separate circuits of the same general voltage. The single-rail units are generally significantly more expensive, and thus most consumers buy the multi-rail PSUs. I could get into a discussion here of the advantages/disadvantages of a single-rail PSU, but that's another subject. Each rail contains its own current limiters and regulators. These components are not perfectly matched, nor professionally adjusted to an exact voltage, thus each rail can vary a bit in voltage output. This is fine in many cases, but in the case that the voltage is higher on the primary rail, this means more of the amperage supplied to the video cards is likely to come through the 24 pin connector!



What's more, the voltage supplied is likely to drop slightly as current draw increases. Thus when the video cards begin to gulp power the voltage on their rails drops slightly, meaning more amperage is likely to be drawn through the 24 pin ATX connector at that point. Also, any loose sockets on the video cards plugs will cause additional resistance on that circuit, also causing more current draw from the 24 pin ATX connector instead.




Conclusions (the hypothesis)

Therefore, if a power supply is used that has a bit higher voltage on the main rail as opposed to the video rails, it is more likely to melt the 24 pin ATX connector. This means you can measure the likelihood of your system being highly susceptible to this occurrence. I suggest you measure the voltage on pins 10 and 11, and then on pins 1, 2, and 3 of the PCIe power plug. Pins 1, 2, and 3 should be almost exactly the same voltage. Compare these to pins 10 & 11 voltage, exactly. If 10 and 11 are higher, you have a high-risk of contracting the melted 24-pin ATX plug problem (theoretically). Variations of just 0.02V should be ignored, such small voltage differences could be caused by voltage changes in the time between moving your probes, and probably also beyond the accuracy of your instrument. In my opinion it would take more like 0.05 or 0.1V differences to be significant. Voltage measurements can be made using just about any black wire on the 24-pin plug for ground (a.k.a. 0V), always use the same one for every measurement you report. Measurements may differ during different load conditions, such as CPU load vs SRAM load vs graphics load; and component temperatures may change the readings as well.


Images courtesy of raw2dogmeat ((not just mine...) EVGA said to remove extension before return [R.I.P.] :( )




I wonder what your results might be under differing conditions, whether or not this theoretical risk remains the same. The extension of this theory is if your 24-pin ATX plug pins 10 & 11 are at least 0.1V under your PCIe plug voltages, you should never have the melted 24-pin plug problem. Also, if you have a single-rail PSU you should be less likely to have the melted 24-pin ATX plug problem if this hypothesis proves true, since your voltage measurements at all +12V points should theoretically be exactly the same except very slightly lower where a lot of power is flowing. So I would also be interested in hearing whether or not anyone with a single-rail PSU in use has experienced the 24-pin ATX plug meltdown.




Call for user reports (data) to support/disprove this hypothesis

I would like to hear some reports from many people on what voltages they measured, then perhaps we could tie these to failure reports in the future to prove (or disprove) the hypothesis. Of course, the new EVGA "powerboost" PCIe power adapter mod fix would make a big difference, so please state whether or not you are using one of those. Please concentrate on those voltage differences between the different plugs, as opposed to how far off 12.00V it is, since we don't know how accurate your test instrument is and that is not what this thread is for. (specification says 11.40V to 12.60V FYI.)


Protocol for your reports, I ask this info:

When reporting your system findings, please include system specs including PSU model, #rails, motherboard model, ALL videocard models, SLI (Xfire) or not, whether or not one card is used for PhysX, voltage measurement(s) on 24-pin ATX plug, voltage measurements on vid cards plugs, voltage measurement on "powerboost" PCIe cable (if you have one), date of voltage measurements, voltage measurement instrument used, date of 24 pin ATX plug meltdown (if). Please report condition of 24-pin ATX plug pins 10 & 11 (perfect or burns or loose). Feel free to add more info of course.


Solder modification instead of "powerboost" PCIe cable:




(Re:EVGA X58 759 Classifed Limited Edition #551 down... going to get an ASUS....=/ - Post#364)





Threads I know of addressing this issue (FYI, for reference)
Internal (EVGA forum) links:

EVGA POWERBOOST (PCI-E Molex fix) [cable] 2010


Are you going to do the MOLEX MOD? e759 / e760 owners

Melted 24-pin ATX power connector, on extension of cable
Atx melt down with gtx 285sli

Official Request Letter/Sign in list for people waiting on EVGA PCIe Power Adapter mod

EVGA X58 759/760 Classified 24 pin mobo/psu burning up Defect With High Power GPU's

10 Days of Appreciation--I'd Appreciate the PCIe Power adapter mod we were all promised!!!

EVGA X58 759 Classifed Limited Edition #551 down... going to get an ASUS....=/ (Locked thread, 915 posts)

E760 Classified Melted my PS 24 Pin Connectors (referenced under picture above)

Classified 3, 3-way SLI and the 24-pin plug

Official x58 Meltdown Poll

(not just mine...) EVGA said to remove extension before return [R.I.P.] :(  (referenced in pictures above)

How many Corsair HX1000's does it take? - Post#39



External links:

Guru3D: ****!! 2 pins from ATX cable MELTED!

Overclockers.com: P6T Melting 24 Pin Connector

HardwareCanucks: E760 Classified Mobo + SLI GTX 480's = Melted 24 Pin Connectors

XtremeSystems.org: Corsair HX1000W USA has a burned 24-pin (referenced under picture above)

Motherboard solder mod to fix this melting issue:
XtremeSystems.org: Note for those running heavy oc'ed Quadfire 4890 (referenced under picture above)

Overclock.net: E760 Classified Mobo + SLI GTX 480s = melted 24 pin connectors



If you know of any other thread you think should be listed here, please post it and I will edit this OP adding it in if I authenticate that it is appropriate. But let's not get too crazy here and add too many external links.

Great write-up.  I've come to the same conclusion in some previous threads on this issue.  Never went as in depth as you have but the conclusions remain the same....draw too much power through those poor little pins and they burn up the connector.  It's an unfortunate limit of the spec for the 24-pin ATX connector.  All those +5V and +3.3V connections on the 24-pin and only 2 +12V is kind of ridiculous in the modern computing era the depends so heavily on +12V power.  To blame also is the board manufacturers and GPU makers not taking into account for this limitation.  ATX and PCIG need to get together with the hardware companies....and each other in order to come out with a unified specification that is open, free, and modular in the sense that it can easily be adapted for future tech.

Awsome Write up man
 
I am just glad EVGA has started placing 4pin molex connecters directly on the board to fix this 12v over draw problem on the 24pin.
 
I know without it my E762 4 Way Classy would have had a serious meltdown by now since I use all 7 of the PCIe slots (4x GTX 400 series cards and 3x add in PCIe x1 cards)
 
I believe the fix for all future motherboards will be to just have the extra 4 pin molex connecter built directly into the PCB layout, thus eliminating the meltdown problem.

Awesome thread, and this got me thinking.  The nasty problem with harness terminations melting and burning is caused by electrical resistance generating heat in the location the damage occurs.  Powerboost cables and solder modifications are simply adding additional paths to supply 12V current to the motherboard.  Wouldn't this create an opportunity for another part of the current path to fail?  You eliminate one source of resistance and there may be other areas that are marginal.  I've seen burned traces and components fail on other types of printed circuit boards when way too much current flows through the circuit. 
 
Kaninja hit the nail on the head with his comment:
 
"To blame also is the board manufacturers and GPU makers not taking into account for this limitation." 

@brknhead; What KMoore is saying, is this:
When you make more connections in a series (in-line) circuit, you add more points of failure, and each will add to the circuit resistance if a bad connection. Series:
 
-----(XXX)-----(XXX)-----(XXX)-----(XXX)----
 
But with a parallel circuit, the primary current flow will travel in the path of least resistance, so any single-point bad connection will simply cause the current to flow in another path. This is somewhat head-scratching, since by this logic the vid cards should start drawing near 100% power from the vid card PCIe power plugs, the ones you plug directly into the vid card; when the 24-pin ATX begins to develop a bad connection. Possibly at that point, then the other 12V devices on the mobo draw enough power to continue the 24-pin meltdown? IDK.
A parallel circuit is much harder to draw with lines of text, I will not attempt.

Edit: He already fixed this
 

ZachA

imagin if someone whipes out and knocks down a power line a few miles away, you will see your service will "black out" momentairily untill another power line in that series works its way around the down power lines and finds your house again giving power back, This could only be done with a parallel circut where more then one path of curreny flow is availible, if it were in series your power will be knocked completly out untill the power company repairs the lines.

LOL, sorry Zach. No in that case the cause of the power failure is usually that the circuit is shorted (to ground). That is a different story, not related to bad connections. The rest of your post is spot-on, but this is a bad analogy - not the same thing. The "black out" occurs when the "hot" wire touches a ground or other grounded object, like a tree for instance. There are "fuseable" connections, that disconnect when too much current moves through them, once this opened, disconnecting that portion of the power grid, the rest of the grid then gets proper voltage again, then the power is restored - except to that portion. That is what you experience when the power "winks." Nice try though. "A" for effort.

Very original posting HolloweenWeed

Thought I was unaffected till I just went in to check if my wires were hot or warm. Of course there is not much room between the 24 pin ATX and the memory on my x58 (E758) so on squishing my finger in there I got a reboot. Tried again and nothing, tried 3 more times and one more time it reboot.So 2 out of 5 times it reboot, some wire or wires in there are definitely not making a real good connection. Only does it in windows though when there is far more current draw and not in the BIOS or log in screen.
Funny thing though is the +12v (yellow) wires are not hot or even warm so I'd not think it was them being loose causing the reboots in my case. Might be in other peoples situations though.
I know not good hardware or even software wise to do what I did but at least now I know and can do something about the bad connection.
 
Think the problems may start as people hide/ tuck away their wires and are having them plugged in and bending them every which way on the final adjustment. Loosens the connections which is part of the theory on these meltdowns I believe.

My system:
No powerboost,
24-pin (rail #1) 12.06V
Pri gfx Rail #2 11.98V
Gfx 2 rail #1 11.99V Xfire mode
Running Kombuster @ 1600x1200, Xtreme burn mode.
 
No evidence of 24-pin heating after 1 yr.

I have a question,will this or could this problem end up on the P67 boards?
There is only a couple out there with the molex plug on the board..
i am getting ready to order parts for a new p67 build and this problem has been nagging at me..

mikotan

Quick question as i plan to upgrade one of my current rigs Halloween.

I'm thinking of getting a gt580 ftw HC from evga.
Will be using it on an e759 rig.
Will be getting only one card for now with moderate oc.
And maybe a 2nd card in the future for sli.

Will I experience the atx burn?
(note: i am currently running 2x gtx 285 oced to 775mhz with no problem for 2 years)

I don't know is the short answer. More accurately, the incidence of this is very low, so it is very unlikely, but possible. The fact that you are only using one card makes it even less likely. But when you add the 2nd card, you are right up there with the rest in terms of likelihood.
 

mikotan

Although i do get the crux of the matter, i am unsure of the parameters within which the atxburn will occur for say the gt580. How much increase in mV or oc will this problem likely manifest itself? It might be noteworthy to find these numbers if possible. 

Thanks in advance.

Yeah, I think I understand what you are asking, and since nobody else responded to my call for data (with measured voltages), I don't have much of an idea, sorry.
 

I've been operating under the assumption that, rather than letting power come from whatever happens to be the path of least resistance, PCI Express devices with multiple power sources specifically control how much power is taken from each source. I've assumed that because I don't know how else these devices could be expected to stay within PCI Express specs (75w max draw per PCI Express x16 slot IIRC) given the variety of motherboards and PSU combinations with which they might be operated. Maybe I'm just misunderstanding because it's past my bed time but am I mistaken?

What would be the best way to determine whether or not this is caused by current taking the path of least resistance?
 
Just checked the temps on pins 10 and 11 of my P7P55-based systems (using a UEI DT304 temp logger with type K thermocouples). Both systems gave the same temps despite having different video cards with different power draws. In both cases, pin 10 read ~120°F and pin 11 read ~113°F. One system has a pair of GTX 470s while the other is a pair of GTS 450s. The 450s should draw a combined ~160W, the 470s a combined ~430W, yet the pin temps were the same. Note that the P7P55 WS Supercomputer has no auxiliary power connectors.
 
Now, to give context to why I've assumed that the cards specifically determine how much power they get from which sources, consider the power consumption test results at XBit Labs for the the cards I have: GTS 450, GTX 470. If you look at more reviews, you'll find many cards, old and new, taking up to but not beyond a limit of 3A from the PCI Express slots. The main exception is "green" cards (ironically) that have no auxiliary power connectors.
 
I'm not trying to disprove your hypothesis, but rather looking for clarification.

I should have noted a few other things (now it's way past my bed time lol):
 
· all of my video cards are at stock clocks.
· ambient temps are 70°F
· cards are Folding 24/7; temps checked with cards at 100% load
· I checked other pin temps as well; temps varied 75°F to 90°F

Thanx for your input, Requital. But if you look at the OP date you will see there is just not enough genuine interest in this thread to hope for enough data to be sufficient to come to any conclusions RE the OP. I left the testing parameters somewhat open bc I knew that adding restrictions = less response from the forum members, meaning no one would have done the test. But if you would like to propose formal test restrictions, go for it. IMO you would just be wasting your time.
 
When I did the OP I knew that it would take many months to first get a database of ppls readings before a meltdown, and then we would have to have a few meltdowns of those ppl that reported in order to see if the readings were significant, and that would take maybe a year. Now that ppl are not responding with readings, there is no hope of this happening before the X58 systems are just old and obsolete, and ppls have lost interest in fixing them.