Groundhog Day: The 12V2X6, melting contacts and unbalanced loads – what we know and what we don’t know

1 - Introduction, foreword and pulsed currents 2 - Intelligent load balancing vs. minimise mania 3 - Measurements and conclusion

I had actually decided not to write anything more about the 12V2X6 connector (formerly 12VHPWR). But after the numerous reports and discussions on the subject, I decided to take another look at it and record my thoughts. In view of the whole issue, where even the (surely accidental) loss of a connector due to the unequal load was documented in a video, I thought a little about how this topic could perhaps add real value to the discussion after all – and without riding the usual clickbait wave. Well, you can try it first.

Of course, you can complicate things unnecessarily or get lost in the details. Today, however, I would like to make a sober and unemotional note of what went through my mind when I took my measurements and the conclusions I drew from them. Special thanks for the mental stimulation go to der8auer and Buildzoid, who have been working intensively on topics such as balancing. Balancing in particular was a point that we discussed from time to time, but then lost sight of in the meantime.

The overheating of contacts on graphics cards could well be due to a combination of mechanical weaknesses, thermal challenges and lack of load balancing. Uneven contact areas, high current densities and pulsating currents could potentially lead to localized resistance increases and increased heat generation, which would be further enhanced by effects such as the skin effect and harmonic components. If the load distribution is inadequate and only a few rails or contacts are overloaded, this could exacerbate the problem and lead to thermal failure. In medicine this is called multiple organ failure, here it is called 12V2X6 for short and cryptic.

I have already done extensive research on the mechanical and electrical problems of the connectors, and I don’t want to repeat this, but I will link it once again as a reminder for everyone, because I completely dislike any gloating about having been proved right in the end. And no, it’s not a buyer and user problem, even if that’s what they tried to sell us on YouTube two years ago. It is and remains a contact issue. Hence my recommendation to read the linked article again:

Smoldering Headers on NVIDIA’s GeForce RTX 4090 – New Insights, Measurements and (Un)known Causes of the 12VHPWR and 12V-2×6 Issue | Update

A possible solution might be to split the load over several busbars (as in the past) or to fundamentally optimize or change the design.

However, I would like to emphasize that I cannot and do not want to commit myself to a single cause here, but will only contribute puzzle pieces to the discussion.

Thought experiment one: Pulsed and high-frequency currents

Before I go into the balancing that Buildzoid discussed later, we should briefly discuss the contacts and the so-called clamping surface as the actual contact area. I have already described this often enough and I don’t want to bring it all up again. But: The voltage converters on graphics cards work with switching regulator technology to provide the operating voltages required for the GPU, memory and other components. They do not generate a constant, but a pulsed DC voltage on the supply lines, which is realized by quickly switching the input current on and off (PWM, pulse width modulation). This pulsed voltage is then smoothed by filters and capacitors in order to generate a DC voltage that is as stable as possible. Simple series chokes or complete LC filters (consisting of a choke and a capacitor) are often used in the input area of voltage converters on graphics cards. These components are used to suppress the high-frequency interference caused by the switching behavior of the voltage converters, which can be fed back into the input voltage and also cause electromagnetic interference (EMI). LC filters are particularly effective in this respect, as the combination of inductance (L) and capacitance (C) enables effective attenuation of interference over a wide frequency range.

However, if only series chokes are used without accompanying capacitors, the filter effect remains incomplete. Although high-frequency interference is reduced by the inductance, the capacitance required to store and dissipate the remaining alternating components is missing without the capacitor. This can lead to residual interference impairing the input supply and possibly also negatively affecting surrounding circuits. There is also a risk that the GPU will react more sensitively to interference peaks as the ripple on the supply voltage increases. NVIDIA pursues a minimization strategy in many of its designs, which aims to save space, costs and components in order to enable more compact and cheaper designs. In practice, however, this has already led to discussions, especially in connection with insufficient filtering. If the filtering at the input or output lines of the voltage converters is too small, this can affect the stability and service life of the card. Too little capacitance or the omission of complete LC filters can, for example, result in increased voltage peaks or electromagnetic interference, which can lead to thermal stress and potentially poorer signal quality. And you shouldn’t skimp on the coils either…

Contact resistances can increase with high-frequency loads, as the electrical properties at the contact points depend heavily on the surface condition, the material behavior and the influences of the alternating currents. In the following, I will try (once again) to explain how the behavior of pulsating currents can lead to an increase in contact resistance and ultimately to thermal failure. This is because high-frequency currents tend to flow preferentially on the surface of conductors due to the skin effect. This effect reduces the effective cross-sectional area available for the current flow, causing the current density to increase locally. However, there are often contact problems at the contact points between two conductors due to a lack of contact pressure or poor fit, as well as unevenness or contamination on a microscopic level, which means that contact is only made effectively at a few points. Due to their small surface area, these punctiform contact points have a high current density, which causes heating.

This heating can in turn lead to a number of undesirable effects. Firstly, the increase in temperature can lead to further deterioration of the contact, for example through oxidation of the contact surfaces or mechanical deformation due to thermal expansion. Oxide layers act as an additional electrical barrier and increase the contact resistance. Secondly, thermal stress can lead to progressive damage to the material, such as a change in the crystal structure or loss of mechanical stability. The increased contact resistance increases the power loss at the contact point, as this is proportional to R⋅I², where R is the resistance and I is the current. The resulting heat development further intensifies the degradation processes, which can lead to a thermal chain reaction. This is referred to as thermal runaway or thermal breakdown, as the process can eventually lead to the destruction of the contact point and possibly the entire system. In addition, high-frequency currents can further increase heat losses through parasitic effects such as eddy currents or the generation of electromagnetic fields in the environment, which further intensifies the mechanisms described above.

Back to the roots: The Molex Micro-Fit connector from Molex

Molex’s Micro-Fit connectors, which were the inspiration for the 12VHPWR and later the 12C2X6 connector, including variants such as Micro-Fit 3.0 and Micro-Fit , are primarily designed for low to medium frequency applications typically found in power and signal transmissions. These connectors are predominantly specified for DC or low frequency AC applications where high frequency effects are generally not a primary concern. However, for applications involving high-frequency currents or signals, additional design considerations are necessary, as parasitic effects such as inductance, capacitance and the skin effect can become significant at higher frequencies. However, the standardization and specifications of the Micro-Fit series, as described in the technical data sheets, do not explicitly take these effects into account. The primary parameters include current and voltage carrying capacity, contact resistance and thermal stability – all under the typical conditions of DC or low-frequency operation.

If micro-fit connectors (or their derivatives such as the 12VHPWR/12V2X6) are nevertheless used in high-frequency applications, the following challenges may arise:

1. Increased contact resistance due to high frequency effects: The skin effect reduces the effective cross-section for current flow.
2. Signal distortion: Reflections or signal delays can occur due to non-optimized impedances.
3. Heating: High-frequency currents lead to local heating due to eddy currents or losses in the contact points.

At this point, it is worth mentioning the second victim in this causal chain: the power supply unit. But I’ll come to that later.

NVIDIA’s connector story: EPS vs. 12VHPWR Connector – Unfortunately, good doesn’t always win, but evil wins more and more often (insights)