In an ideal world, devoid of pesky details like contact resistance and manufacturing imperfections, you would be able to double the current that can be provided to a device by doubling the number of conductors without altering the device’s circuitry, as each conductor would carry the exact same amount of current as its neighbors. Since we do not actually live inside a simplified physics question’s scenario, multi-wire powering of devices comes with a range of headaches, succinctly summarized in the well-known rule that electricity always seeks the path of least resistance.

As recently shown by NVidia with their newly released RTX 50-series graphics cards, failure to provide current balancing between said different conductors will quickly turn it into a practical physics demonstration of this rule. Initially pinned down as an issue with the new-ish 12VHPWR connector that was supposed to replace the 6-pin and 8-pin PCIe power connectors, it turns out that a lack of current balancing is plaguing NVidia GPUs, with predictably melty results when combined with low safety margins.

So what exactly changed that caused what seems to be a new problem, and why do you want multi-wire, multi-phase current balancing in your life when pumping hundreds of watts through copper wiring inside your PC?

Resistance Is Not Futile

In the absence of cheap room-temperature superconducting wires, we have to treat each conductor as a combination of a resistor, inductor and capacitor. These parameters set limitations on properties such as how much current a conductor can carry without changing phase from solid to gaseous. The contact resistance between the conductors of both sides in a connector adds another variable here, especially when a connector wears out or the contacts become corroded.

In the case of the 6-pin and 8-pin PCIe power connector, these are based on the Molex Mini-Fit series, with the commonly used Mini-Fit Plus HCS (high current system) rated for 100 mating cycles in tin plating or 250 cycles in gold, and a current rating of 8.5 A to 10 A per pin depending on whether 18 AWG or 16 AWG wire is used. The much smaller connector of the 12-pin 12VHPWR, and equivalent 12V-2×6, standard is rated for only 30 mating cycles, and 9.5 A per pin. It is based on the Molex Micro-Fit+ connector.

The smaller pin size and lower endurance increases the possibility of poor contact, as first demonstrated with the 12VHPWR connector back in 2022 when NVidia RTX 40-series cards experienced run-away thermal events where this power connector on the GPU side melted. Subsequent research by the team at Gamers Nexus showed this to be due to poor contact within the connector with resulting high resistance and thus a massive thermal hot spot. Following this event, the 12V-2×6 update to 12VHPWR increased the length of the power pins and decreased that of the four sense pins.

The idea behind this change is that by extending the length of the power and ground pins by 0.25 mm and shortening the sense pins by 1.5 mm there’s a higher chance of there being an actual good contact on the ground and power pins when the sense lines signal the GPU that it can start drawing hundreds of watts.

This change did only affect the male side of the connector, and not the cable itself. This made it very surprising to some when after the much higher wattage RTX 5090 GPUs were released and suddenly cables began burning up,with clear melting visible on the GPU and power supply side. What was going on here?

Multi-Phase Balance

Shortly after the first melting cable event involving an RTX 5090 Founders Edition (FE) GPU popped up on the internet, Roman [Der8auer] Hartung reached out to this lucky person and – since both live quite close – borrowed the damaged GPU, PSU and cable for an investigative video. Involved were not only an RTX 5090 FE GPU, but also the PSU with its 12VHPWR connector. On each side the plastic around one pin was completely melted, with the cable having to be forcibly removed.

During Roman’s testing with another RTX 9050 FE and 12VHPWR cable he found that two of the six 12V wires were significantly warmer than the rest, courtesy of these carrying over 22 A versus around 2 A for the others while the PSU-side connector side hit a blistering 150 °C. This result was replicated by some and seems to be fully due to how the NVidia RTX 9050 FE card handles the incoming power, by tying all incoming power lines together. This a practice that began with the RTX 4090, but the RTX 5090 is the first to pull close to the rated 600 watts of the 12VHPWR/12V-2×6 connector. This was explained quite comprehensively in a comparison video by Buildzoid.

Because with the RTX 4090 and 5090 FE GPUs – as well as some GPUs by third-party manufacturers – these 12V lines are treated as a singular line, it is essential that the resistance on these lines is matched quite closely. If this is not the case, then physics does what it’s supposed to and the wires with the lowest resistance carry the most current. Because the 12V-2×6 connector on the GPU side sees only happy sense pins, it assumes that everything is fine and will pull 575 watts, or more, through a single 16 AWG wire if need be.

Meanwhile the Asus RTX 5090 Astral GPU does have individual shunt resistors to measure the current on the individual 12V lines, but no features to balance current or throttle/shutdown the GPU to prevent damage. This is actually a feature that used to be quite common, as demonstrated by this EVGA RTX 3090 Ti GPU:

On the top right the triple sense resistors (shunts) are visible, each of which is followed by its own filter coil and feeding its own set of power phases, marked in either red, green or blue. The yellow phases are for the RAM, and are fed from the PCIe slot’s 75 Watt. The bottom right controller controls the phases and based on the measured currents can balance the current per channel by shifting the load between parts of the phases.

This is a design that is completely omitted in the RTX 5090 FE design, which – as Igor Wallossek at Igor’s Lab describes it – has been minimized to the point where crucial parts have begun to be omitted. He also covers an MSI RTX 3090 Ti Suprim card which does a similar kind of phase balancing before the RTX 4090 and RTX 5090 versions of MSI’s GPUs begin to shed such features as well. It would seem that even as power demands by GPUs have increased, crucial safety features such as current balancing have been sacrificed. As it turns out, safety margins have also been sacrificed along with these features.

Safety Margins

The ugly truth about the switch from 8-pin PCIe connectors to 12-pin 12VHPWR connectors is that while the former is rated officially for 150 watts, this power level would be hit easily even by the cheapest implementation using crummy 18 AWG wiring. With the HCS connectors and 16 AWG wiring, you are looking at 10 A × 12 V × 3 = 360 watts, or a safety margin of 2.4. With cheaper connectors and a maximum of 7 A per wire it would still be a safety margin of 1.68.

Meanwhile, the 12VHPWR/12V-2×6 with the required 16 AWG wiring is rated for 9.5 A × 12 V × 6 = 684 watts, or a safety margin of 1.14. In a situation where one or more wires suddenly decide to become higher-resistance paths this means that the remaining wires have to pick up the slack, which in the case of a 575 watt RTX 5090 GPU means overloading these wires.

Meanwhile a 8-pin PCIe connector would be somewhat unhappy in this case and show elevated temperatures, but worst case even a single wire could carry 150 watts and be happier than the case demonstrated by [Der8auer] where two 12V-2×6 connector wires were forced to carry 260 watts each for the exact same wire gauge.

This is also the reason why [Der8auer]’s Corsair PSU 12V-2×6 cable is provided with two 8-pin PCIe-style connectors on the PSU side. Each of these is rated at 300 watts by Corsair, with Corsair PSU designer Jon Gerow, of JonnyGuru PSU review fame, going over the details on his personal site for the HCS connectors. As it turns out, two 8-pin PCIe connectors are an easy match for a ‘600 watt’ 12VHPWR connector, with over 680 watt available within margins.

There’s a good chance that this was the reason why [Der8auer]’s PSU and cable did not melt, even though it clearly really wanted to do so.

Balance Is Everything

Although it is doubtful that we have seen the last of this GPU power connector saga, it is telling that so far only GPUs with NVidia chips have gone full-in on the 12VHPWR/12V-2×6 connectors, no doubt also because the reference boards provided to board partners come with these connectors. Over in the Intel and AMD GPU camps there’s not even a tepid push for a change from PCIe power connectors, with so far just one still-to-be-released AMD GPU featuring the connector.

That said, the connector itself is not terrible by itself, with Jon Gerow making the case here quite clearly too. It’s simply a very fiddly and somewhat fragile connector that’s being pushed far beyond its specifications by PCI-SIG. Along the way it has also made it painfully clear that current balancing features which used to exist on GPUs have been quietly dropped for a few years now.

Obviously, adding multiple shunts and associated monitoring and phase balancing is not the easiest task, and will eat up a chunk of board real-estate while boosting BOM size. But as we can see, it can also prevent a lot of bad publicity and melting parts. Even if things should work fine without it – and they usually will – eating into safety margins and cutting components tends to be one of those things that will absolutely backfire in a spectacular fashion that should surprise absolutely nobody.

Featured image: [ivan6953]’s burnt cables.