The closing image from the last post: a Hardware Unboxed technician reached over, felt a cable that seemed unusually hot, and checked the meter. That wire: 22 amps. The other five: zero.
You might assume that if one wire is carrying more, the other five are splitting what's left — uneven, but at least being tracked somehow. Here's what's actually strange: most high-end cards on the market today have no way to even know how much current each of the six wires is carrying individually. It's not that the card chooses to ignore it. The circuit design means it literally can't measure it.
This post opens that black box: why the card can't measure current on each wire separately, what the one consumer card with per-pin sensing actually saw when it looked, whether repeated plugging wears out the contacts, and an experiment that overturns the intuitive story — current concentration on its own doesn't automatically mean melting.
Where the current on each of the six wires goes, most cards genuinely don't know
Whether current gets shared evenly comes down first to how the card measures current at all — not to the six wires themselves. EVGA and MSI's versions of the RTX 3090 Ti split the six 12V wires across three independent shunt resistors, each tied to a different power phase. The controller can see each of the three groups separately, catch one phase running high, and in principle has room to adjust.
If your card is an RTX 4090 or RTX 5090 Founders Edition, it does this differently: all six 12V wires feed into a single shared shunt resistor. Well-known overclocker Buildzoid described this design as "one big blob of 12V" (reported by Hackaday in February 2025, and by HotHardware). The consequence is direct: your card can only see the combined total across all six wires, not what any individual one is carrying — which means it cannot detect, and cannot correct, a single wire running hot. Current doesn't share itself out evenly on its own. It simply takes whichever path has the least resistance, whichever wire that happens to be.
| RTX 3090 Ti (EVGA / MSI) | RTX 4090 / 5090 reference | |
|---|---|---|
| Shunt resistor grouping | Three independent groups, two wires each | All six wires into one group |
| What the controller can see | Each group's own current | Only the combined total of all six |
| Can it catch one wire misbehaving | Roughly, by phase | Not at all |
Which means whether your card can even see a problem coming depends entirely on which of these two designs it uses. Which raises an obvious question: what if a card actually did spend the money to give all six wires their own sensor? One does.
The one card with six independent meters can see the problem and still can't save you
The ASUS ROG Astral is currently the only consumer graphics card with true per-pin current sensing — all six 12V wires have their own measurement circuit, which is in principle exactly the fix for the "one shared safe" problem above.
But it only solves the "seeing it" half. When the Astral detects abnormal current on a specific wire, what it can do is alert you — it doesn't automatically shift load away from the overloaded wire, doesn't throttle on its own, and doesn't shut itself down when the numbers get extreme. Monitoring answers "will you find out sooner," not "will it happen." And there's a sharper irony here: the Astral itself had a documented melting case in September 2026. Having the sensor doesn't make a card immune.
Where does the resistance difference the sensor is measuring actually come from? Two things stacked together: the resistance of the wire itself, plus contact resistance at both ends of the connector — one end plugged into the card, the other into the power supply or cable (reported by HWCooling). Wire resistance is fixed at manufacture. Whether contact resistance changes based on how you use it is where the record starts to disagree.
Whether mating cycles wear out a connector: two tests that flatly contradict each other
You've probably plugged your card in and never given it a second thought since. Germany's Igor's Lab ran an aging test: after several hundred mating cycles, measured contact resistance (LLCR) rose from 1.3 to over 2 milliohms. That's not a small gap — at the same 9.5 amps, it means over 50% more heat generated at that point.
JayzTwoCents got a completely different result. After 100 mating cycles, contact resistance showed no observable change. What actually failed was the latch, not the contact itself.
| Igor's Lab | JayzTwoCents | |
|---|---|---|
| Mating cycles | Several hundred | 100 |
| Contact resistance change | 1.3 mΩ → over 2 mΩ | No observable change |
| What actually failed | The contact itself | The latch, not the contact |
Both are serious, deliberate tests, and their conclusions flatly disagree — this one is unresolved, and this post isn't going to pick a side. The more plausible explanation comes from a separate observation by GamersNexus: what actually drives up contact resistance may not be "how many times you plugged it in" but bending and lateral stress. Pulling a plug sideways can leave it improperly seated, and over time may deform the terminals — if that's the real mechanism, mating-cycle count isn't the variable that matters. How you route and handle the cable is.
There's still one question left unanswered, though: even if a specific wire really does have elevated contact resistance and current really does pile onto it, does that guarantee it melts?
Concentration alone doesn't guarantee melting: a counter-example that complicates the simple story
By now you might be wondering whether your own card is at risk just because one wire happens to be carrying more than the rest. Tech outlet GamersNexus ran an experiment, and it needs a clear label up front: this is a reported test from a media outlet, not raw data this post's author verified directly. They deliberately cut four of the six 12V wires, forcing a card's full 600 W load through only the remaining two — making sure both were fully seated and made good contact. The result: no melting, with temperatures well below the melting point.
That result is worth pausing on, because it cuts against intuition. Two wires carrying 600 W means each one is theoretically running well past its individual rating — if current concentration by itself were the fatal flaw, this setup should have failed. It didn't, which means pure concentration, with good contact, isn't necessarily enough on its own to cause a failure.
Putting every piece from this post together, the more defensible conclusion is that no single factor is fatal by itself — it's concentration plus an already-elevated contact resistance at that specific point, happening at the same time. Either one alone, and the connector seems to hold. Both together is the pattern that shows up across the actual melting cases — which is also why you can't judge your own setup safe just because "my cable feels thick" or "I pushed it in hard."
Frequently asked questions
Why doesn't current split evenly across your six 12V wires?
Because the RTX 4090 and 5090 reference boards feed all six wires into one shared shunt resistor, so the card only sees the total, not what each wire carries. Current follows the path of least resistance, and the card has no way to detect or adjust a single wire — unlike the RTX 3090 Ti, which uses three independent shunt resistors and can at least tell roughly which phase is running high.
The Astral has six independent sensors — does that make it safe?
Not entirely. It's the only consumer card with true per-pin sensing, but that only gets you monitoring — an alert when something's off, not automatic load-shifting, throttling, or shutdown. Monitoring means finding out sooner, not preventing it. The Astral itself had a documented melting case in September 2026.
Does repeated plugging and unplugging wear out your connector's contacts?
The two available independent tests contradict each other, with no resolution yet. Igor's Lab measured contact resistance rising from 1.3 to over 2 milliohms after several hundred cycles; JayzTwoCents saw no observable change after 100 cycles, with the latch failing instead. The more likely explanation is that bending and lateral stress drive up contact resistance, not mating cycles themselves.
If your card's current concentrates on one wire, does that mean it will melt?
Not necessarily. GamersNexus's test (media-reported, not verified firsthand here) forced a 600 W load through just two fully-seated wires and it didn't melt. That suggests concentration with good contact isn't automatically fatal — the dangerous combination is concentration plus already-elevated contact resistance, happening together.
Where this stands, honestly
Everything in this post — how shunt resistors get grouped, whether contact resistance degrades with mating cycles, what condition has to accompany current concentration before it becomes dangerous — happens at the level of graphics card board design and connector materials. We have no way to verify these figures ourselves, and no lab to rerun Igor's Lab's or JayzTwoCents's tests. What we can do is mark sources clearly and write contradictions as contradictions, honestly.
What PowerDoze can actually do here is limited: setting a lower GPU power limit brings down the average current across all six wires together, which widens the safety margin the last post described. But it changes nothing about how current divides between the six wires, and it touches no connector's contact resistance — those are exactly the two problems this post is about, and both sit outside anything software can reach.
If your card runs on traditional dual 8-pin power, the shunt-resistor design problem in this post mostly doesn't apply to you — dual 8-pin cables route independently by design and were never merged into a single shared measurement group, and you don't need PowerDoze or anything else installed on account of it. What else actually differs between 8-pin and 12VHPWR on paper — mating cycles, wire gauge, materials, and a few other things that get misreported often — gets checked line by line in the next post. The full how-to for setting a GPU power limit yourself, including how to find your card's legal range and how to make it survive a reboot, is in this guide.
Wondering whether your GPU's power limit is still set after your last reboot? That's exactly the problem PowerDoze handles — you set the limit once inside a power mode, point an all-day, every-day schedule rule at that mode, and PowerDoze re-runs the nvidia-smi command at every startup instead of you retyping it. It won't balance current across six wires, and it won't fix a connector's contact resistance — this post has explained why: that's a board-design-level problem.