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

Minimize Mania at NVIDIA

With the Ampere generation and the NVIDIA GeForce RTX 3080 and RTX 3090 Founders Edition, NVIDIA launched an avant-garde shaped design onto the market. All well and good, but the question arises: where to put the space-consuming headers? When the board designer runs out of space, in the worst case scenario this is replaced by a daring combination of courage and improvisation – or, as one might critically put it, by “creative ruin”. It was particularly annoying that there were no upright headers for this design at the time, not even at Molex in South Korea, who have been producing plugs since 1984 and should actually know what they are doing (or rather not doing).

Let’s take a closer look at the headers of the GeForce RTX 3090 Founders Edition. Molex’s original Micro-Fit 3.0 design serves as the basis here, but the header differs slightly. The profiles (still) match, but the design of the header is already different. The side guides, which require a matching cable plug (connector socket) and prevent crooked insertion, are interesting. Molex officially describes this component as a “Micro-Fit 3.0 Right-Angle Header, 3.00mm Pitch, Dual Row, 12 Circuits, with Snap-in”.

With the introduction of the Founders Edition (FE) of the NVIDIA GeForce RTX 3080 and RTX 3090, not only the avant-garde shaped design came into play, but also a clear lack of space on the boards. This problem became particularly apparent as the design no longer followed the function, but the functions increasingly had to be subordinated to the design. The elegant design of the cards, characterized by their special heat sink and airflow layout, left the board developers little scope for practical solutions. The space problems on the board meant that technical requirements had to be bent in order to fit into the overall aesthetic concept.

This leads to a paradoxical situation: while the shaped design is impressive on the outside, there are technical compromises beneath the surface. The limited space not only meant that creative solutions were required, but in some cases also that existing standards had to be pressed into a form that they were not originally intended to fulfill – with consequences for long-term functionality and durability that are still unclear today. This approach puts to the test the extent to which function can be subordinated to aesthetics without jeopardizing technical integrity. The question remains as to whether this “design follows Jensen” mantra meets the long-term expectations of reliability and usability or is ultimately more a case of “creative ruin”.

Train of thought two: lack of balance

This is exactly where Buildzoid comes into play. In my archive, I also have an interesting example of a well thought-out implementation and wiring of the 12VHPWR / 12V2X6 and so the MSI GeForce RTX 3090 Ti Suprim was a perfect solution at the time, because its 450-watt layout was already adapted to the requirements of the 12VHPWR standard in a future-oriented manner. Although the sideband channels (sideband signals) have not yet been wired, this card can already be seen as a test platform for the RTX 4090 that will be released later. MSI has implemented a design here that not only responds to the increased performance requirements, but also focuses on security and efficiency.

MSI’s exemplary power distribution solution is particularly noteworthy. Instead of treating all six current-carrying 12V lines of the 12VHPWR connector as a common unit as in the Founders Edition, MSI has divided the power supply into three independent rails. Each of these rails is monitored and protected separately, which not only improves load distribution but also increases safety. If overloads or other anomalies occur on one rail, it is specifically shut down without affecting the entire power supply!

In addition, MSI has integrated appropriate filtering to minimize high-frequency interference and voltage peaks. This filtering makes a decisive contribution to the stability of the power supply and protects the sensitive components of the graphics card from possible damage. This is particularly important for high-frequency, pulsed currents, such as those generated by modern voltage converters, and shows that MSI fully understood both the technical challenges and the practical requirements of the future 12VHPWR standard. Back then…

Although the MSI RTX 4090 Suprim then continued to rely on an initially separate power supply design, some changes had been made compared to the RTX 3090 Ti Suprim. Instead of using three independent rails as before, only two rails were now used, which were also combined into one rail after the two shunts. Even though this still offers the advantage of separate monitoring compared to the Founders Edition from NVIDIA, balancing is no longer possible here. Unfortunately, this path has been blocked. It’s unfortunate, but the reference board that NVIDIA cites as the motherboard solution simply no longer provides for this. But at least there was still the one fuse.

It is also particularly striking that MSI has completely dispensed with fuses on the 12V2X6 connection on the RTX 5090 Suprim (and only provides them for the PCIe slot), although such fuses are explicitly advertised in the product description. This raises the question of whether the decision is due to savings or a different prioritization in the design.

Nevertheless, the use of two rails on the 12V2X6 remains at least a compromise solution, which at least offers a certain level of protection and load separation, even if the level of protection no longer comes close to earlier models, because after the longitudinal chokes everything is brought together on one rail. Somehow this is all annoying when you consider the prices for such cards. And it’s like the Titanic: lifeboats and protective vests are completely overrated if you hope that an iceberg won’t come along and, if it does, that you can at least swim.

Which brings us back to the image of the VU, where everything is combined in the header to form a single bill and only a single shunt is used to monitor the overall current. The effect of the remaining coil as a filter is debatable, but never mind. What is not installed cannot break, said the accountant to the usher and laughed himself to death. And if they haven’t died, they are still celebrating happy plug parties with the designer today.

A balanced design, as with the GeForce RTX 3090 Ti board partner card, ensures that the current load is distributed evenly across the existing cables and rails. This reduces the selective overloading of individual contacts and thus avoids hotspots that can occur due to increased contact resistance and the resulting heat development. In practice, poor load balancing can lead to some cables overheating while others are hardly loaded, which significantly shortens the service life of the connector and increases the risk of thermal failure. And unfortunately, this is what we have been seeing for some time now.

The advantage of intelligent balancing also lies in the possibility of efficiently integrating protective mechanisms. With several rails, power monitoring and protective circuits can be used in a more targeted manner to detect overloads at an early stage and switch off individual rails in an emergency without jeopardizing the entire power supply to the card. This not only offers an increase in safety, but also greater robustness against fault conditions that can occur in practice due to fluctuating loads or faulty plug connections. This should actually be the purpose of good multi-rail power supplies. However, NVIDIA no longer seems to be interested in this.

Especially with modern GPUs, whose power requirements are often at the limit of what the 12V2X6 connector and the associated cables can safely transmit, the combination of good balancing and protection mechanisms would be crucial. Without these measures, both the connectors and the internal rails of the graphics card are exposed to unnecessary stress, which in the worst case can lead to irreversible damage. Manufacturers who pay attention to such aspects ensure that the function not only meets short-term requirements, but also remains stable under load changes and continuous operation. So please rethink!

Note: A well-balanced power distribution and the implementation of protection mechanisms are of crucial importance for powerful graphics cards with such high currents and connectors such as the 12V2X6 connector that are sewn close to the edge. Especially with few rails and the high power requirements of modern GPUs, a good design is essential to ensure the safety, reliability and longevity of the entire power supply.