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

Square Wire / Profile Wire

To better understand the following, I will first provide a brief insight into the significance and manufacturing of so-called square wire, from which the pins of the headers are cut, pressed in, and then bent at the back. The wire that header manufacturers use has a square cross-section and is produced from simple, drawn round wire. To shape square profile wires, a so-called “Turk’s head” is used, the term stemming from the arrangement of the rollers, reminiscent of the traditional tying of headscarves in Turkish culture. Learned something new… The Turk’s head consists of four rollers symmetrically positioned at right angles to each other, exerting even pressure from all four sides when the wire is drawn through.

This very process is pivotal for the quality of the square wire and its dimensions, and along with the electroplated coating and its tolerances, the final outer dimensions are determined. We will soon see a pin whose wire was not perfectly rolled and still shows remnants of the round wire’s curvature. If such a wire is processed anyway, it means the Quality Control Inspection (QCI) of the connector manufacturer has failed. And we also recall the relatively soft copper and its mechanical disadvantages compared to a more suitable alloy.

Clamping surface in spring clamp connections

This term also requires a brief explanation. In connectors with spring contacts, the clamping surface is the part of the contact system that presses the conductor against a counter-surface by spring force to establish an electrical connection. In such connectors, the clamping surface is usually part of the spring element, designed to press the conductor securely and firmly against a stationary contact surface without the need for a screw or other tool. The spring force ensures a constant pressure, maintaining a reliable electrical connection throughout the connector’s lifespan. The clamping surface must have high conductivity and must not damage the conductor. It is often designed to accommodate different conductor cross-sections and to compensate for slight movements or vibrations and temperature fluctuations, which could otherwise lead to a loosening of the connection. Now you also understand why I insist on the lack of definition for these in the specifications.

The near-ideal pins of the 3rd party headers

Yes, they do exist. The pins in the headers from CableMod (at the adapter output to the power supply) all conform to standards regarding pin strength, alignment, and positioning. It can be done when the header manufacturer really wants it to be so! This should be emphasized, because aside from the tolerances PCI SIG allows for emergencies: The required 0.64 mm should be the norm, not the exception! However, CableMod also uses headers with pins made of pure copper instead of brass. But the wire used could not be more precise, and even the edges are only slightly rounded. I tested the headers from a total of 18 unused CableMod adapters from various batches and found the same, nearly perfect measurements in all 12 pins of each header.

Moving on to the second model, which represents the replacement headers used by KrisFix to replace scorched original headers. Here too, we see a very well-aligned pin with clean edges and outer dimensions that are slightly above the norm but still well within the upper tolerance range. Here, the spring should hold even better because the clamping force increases slightly as a result.

The pins on the original headers: Fishing in the tolerance range

I have also examined headers from Asus, because I had many of them, providing a certain statistical reliability. The wires used are often uneven, with slight bulges (compression) instead of straight edges and also often uneven dimensions. Yes, it’s all in the micrometer range, but when it comes to borderline high currents, the effective usable clamping surface is important and it certainly suffers from such deformations.

The next image was taken with different lighting and maximum depth of field. We see a pin where the plastic at the feedthrough has already melted a bit. The measurements here are the absolute maximum values, which again show that the contact surfaces of the pin are partially even slightly bent and not truly flat. Let’s quickly recall the sentence at the beginning of the page about round wire and the problems in manufacturing.

And another thing: I have not been able to find a pin with 0.64 millimeters or more in outer dimensions that would have been damaged in this way! It was always only the thinner ones, with 0.63 mm and significantly less, or those that were not really rolled smoothly! Perhaps the PCI-SIG should define the lower limits a bit tighter and no longer allow negative tolerance ranges of this magnitude. And the connector manufacturers really should be advised to check the quality of the wires more often and thoroughly.

It must also be taken into account that the top layer is made of soft tin. There is definitely material wear and tear during plugging and unplugging, so this lower tolerance limit could be quickly undercut. But it gets even less accurate!

In fact, there are pins that are twisted internally! A 3-degree deviation of the axes from the vertical on this non-charred pin is already significant. This brings us back to the crimped cable plugs. Such a spring contact, since it sits loosely in the housing, can absorb and compensate for this twist to a certain extent, thus adjusting the positioning. However, if the counterpart is fixed-wired, such as a true socket like the one we find on the CableMod adapter and whose pins are soldered onto a PCB, no compensation can occur, and the clamping surface is also significantly minimized in part because the contacts are askew and therefore do not lie flat completely.

Position Tolerances

Now I will attempt to cut through the upper row of 12-volt pins. Of course, the inaccuracies of the optical measurement must be taken into account, but I have taken the picture with a 200-fold magnification and maximum depth of field, and also had the image stitched to largely eliminate optical distortions. Thus, it’s an undistorted top view at all positions. However, I must note for the sake of accuracy that these deviations are an exception and not all headers fall out of norm to the same extent. This also applies to the pins, which, however, even varied more noticeably among themselves in the defective headers. So, it will come down to manufacturing tolerances of certain batches.

From the cross-sectional measurement, it can be seen that the pins are not necessarily in the correct position, but this moves within the official tolerance specifications that were given. However, this only works with more flexible cable plugs, as they can adapt better. If the counterpart is also soldered onto a PCB, there is little room for maneuver and it only goes over the spring contacts. The possible consequence would then be an unequal pressure on the respective four side surfaces of the pins. And the very left pin, i.e., the one that has tended to char first in the past, is even below the tolerance limit for pin dimensions.

Intermediate Conclusion

It is quite obviously the sum of pin size, positioning, and possibly twisting that negatively affects the contact surfaces (“clamping surfaces”). Such a borderline construct in terms of currents to be delivered demands the highest manufacturing quality to really be considered safe. And it is not acceptable that I can short circuit three CableMod adapters in series at a limited 600 watts on the laboratory power supply and nothing happens, while a single inferior header is enough to cause damage to the expensive graphics card.

I can only advise manufacturers of these connectors to adhere exactly to the specifications and to check this continuously and more often within a series. Yes, these are higher costs, but the customer will surely appreciate it. Part of the problem certainly lies in the wire used for the pins, as the incoming goods must be inspected more closely by quality control from the start. Since there are many manufacturers of connectors, the quality of the products varies greatly. While the plastic housings were all cast very correctly, it becomes more problematic with the square wire and its initial quality.