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

1 - 2x EPS and over 700 watts? It could have been so easy! 2 - NVIDIA runs out of space with the Founders Edition 3 - GeForce RTX 3090 Ti: Playground for Ada and the 12VHPWR 4 - NVIDIA's mystic 12VHPWR adapter 5 - From 12VHPWR to 12V2x6 connector - user error de luxe? 6 - Flip headers, quality issues and a conclusion

After numerous articles, explanations, and follow-up tests, today we’re taking a different approach. I have three intriguing topics for you: the possible origin story of the 12VHPWR connector, the deliberate ignorance of the originally intended 8-pin EPS connector, which continues to prove its resilience against defeat in power supplies and on motherboards, and a patented flip-header on Asus cards that playfully violates PCI SIG standards, often ending up delicately melted on the PCB as a costly demise. While this may sound a bit casual, these topics have, unfortunately, serious underpinnings.

For my article today, I interviewed many people and received responses quite often. Those who were forthcoming shared some truly astonishing information. Many pieces of information aligned, so I used them accordingly, although there are always some uncertainties. I don’t claim to hold the absolute truth, but most of it should be pretty close.

The solid 8-pin EPS connector was initially intended to be the solution until it was “executed”

Few may know this, but even before the launch of Ampere consumer cards, like the GeForce RTX 3080 and 3090, NVIDIA seriously considered replacing the triple connector strip with 6×2 headers on high-end graphics cards with solid EPS 8-pin headers! It’s important to note that the PCI Express 2×4 Auxiliary Power Connector, as referred to in the PCI SIG, fundamentally differs from the EPS with 2×4 pins. Essentially, it’s only a 2×3 connector, with two additional pins (one per row) acting as mere sense pins, not providing additional power.

The 2×4 EPS connector, on the other hand, has true 4 pins for 12 volts and ground, without any superfluous sense pins to report on the quality of connected cables. For backward compatibility, one could even plug a standard 6-pin PCIe connector into it. Considering NVIDIA’s vision for cards with over 300 watts of power consumption, they initially sought a feasible solution within the existing repertoire of PC and power supply builders. And yes, there were NVIDIA cards with a real EPS connection (see image below)! The connector profiles clearly show that a 6+2 cable wouldn’t fit.

And so, the first power adapter was created, which didn’t yet use the 12-pin Micro Fit 3.0 connector but divided a single 8-pin EPS into two 6+2 cables. I keep this header in my collection of rarities, just like the RTX A6000. Two such adapters with four 6+2 cables could have solidly powered an RTX 4090 with 600 watts (or more). This was even considered back then, but we know how it turned out in the end.

During that time, NVIDIA advised various power supply manufacturers and builders to include at least two additional native 8-pin EPS connectors and the corresponding cables in their future plans for powerful power supplies. This approach would have significantly minimized compatibility issues! Corsair cleverly implemented this in their RM series. Below, we see two native EPS headers that only fit EPS cables, while the four headers above are “unisex.” Any 8-pin, 4.2 mm pitch profile would fit there, hence the explicit instruction to use only original, proprietary Type-4 cables. Native 6+2 cables could cause a serious short circuit, so the included 6+2 cables also have an EPS profile on the power supply side. However, since it wasn’t clear what NVIDIA really wanted and trust was low, Corsair cleverly kept all options open.

The original Korean Molex connector (shown here in a 3D model) allows up to 13 amps per pin and at least 7 amps when all are used simultaneously (so-called de-rating). This translates to 28 amps per header at full load, or 336 watts. The 2×4 headers (known 6+2 connections) of the PCI SIG are also rated at 7 amps. That equates to only 252 watts for three 12-volt lines, making it insufficient for a GeForce RTX 3080, considering the additional power from the PCIe slot on the motherboard.

It’s also worth knowing that these Molex connectors come in 10- or 12-pin variants (2×5 or 2×6). This could provide up to 504 watts for a single connection using the 7 amps de-rating! Thus, the original 8-pin EPS could have easily powered graphics cards up to 350 watts and more, using the PCIe slot. Even an extremely overclocked GeForce RTX 4090 could have been powered with over 600 watts using two EPS connectors.

The decision to eventually discard this connector likely stems partly from the playful nature of the graphics card manufacturer and possibly the challenge of modifying this connector without licensing issues. Molex offers a nearly endless array of suitable connectors, all functional. To set the stage for the next chapter, I’ve included a nice overview that shows both the Mini Fit with 4.2 mm pitch (as with EPS and 6+2) and the variants of the Micro Fit 3.0 with 3.0 mm pitch, including those capable of handling even higher currents. Pitch refers to the distance between the centers of the pins, effectively defining a grid pattern.

With Micro-Fit 3.0, there’s a lot of potential, and even more with Micro-Fit (Plus)! And that’s the next topic, so please turn the page.