Fire Hazard! Design Flaw? NVIDIA 12VHPWR Connector Explained

I’m sure you’ve seen the pictures of burned and melted 4090’s on Reddit already, but since I finally have my hands on one of these, I think we should talk about it in a bit more detail and explain what’s going on here. For a bit of context, let’s talk power connectors. The PCIe slot on its own can deliver up to 75W of power to any add-in card – that’s why you can find lower end GPUs that don’t need any power connections, and why things like network cards don’t need any extra power either. Of course, for higher end cards, they need some extra juice.

So, enter the 6 pin PCIe power connector. This is a universal standard that can add an extra 75W of power to the GPU through its three pairs of 12V and ground pins. Nice and simple. Then GPUs started needing more power… This HD6870 was a mid-tier card, but even it needed two 6 pin connections, with the top end (especially dual GPU) cards needing up to 4 connections. It all got a bit silly really, so in comes the 8 pin connection instead. The 8 pin is a superset of the 6 pin, meaning any 8 pin connectors are all backwards compatible with 6 pins thanks to the optional two pins. Strangely, those extra two pins don’t actually supply any extra power. They are two more ground pins which basically act as sense pins to tell both the GPU and power supply that the PSU supports offering 150W through the existing 6 pins, and that the GPU might need that much.

Perfect, we couldn’t possibly need more than 225W right? RIGHT? Yeah that didn’t last long. It’s now pretty common to find higher end GPUs with THREE 8 pin connections for a total of 525W of total power supply – and the new 4090 can benefit from FOUR. So, NVIDIA thought it was time to change things up again, which I totally get. We’ve been using the 8 pin for over a decade at this point so a new and improved solution is about due. And that’s where we run into problems. See, instead of coming up with another superset design, say with an extra four pin block for a total of 12 pins but with the option to run it as a 6 or 8 pin as well, NVIDIA decided a wholly different connector is needed. I mean ok, I get it. Having TWO optional blocks per connector is a little silly really and I’m sure you could streamline the process a little more with a new connector.

Sadly, the connector they picked was this. A 12 pin (plus four sense pins separately) that is considerably smaller than the existing connectors. Here’s a look at a 12 pin connector of the same pin size as an 8 pin – this is what BeQuiet uses on their PSU side connections to supply up to 300W through two 8 pin connectors on the same cable. Now compare that to the size of NVIDIA’s 12VHPWR connection. It looks like it’s been shrunk in the wash, it’s comically small, and that’s not all. Instead of offering a doubling of the power from 150W to 300W like we saw from 6 to 8 pin connectors, NVIDIA is running up to 600W through this. Yes, that’s right. This tiny thing is meant to run FOUR TIMES AS MUCH POWER AS AN 8 PIN CONNECTION.

Even if you pretended this was the larger pin size connector, running 4x the power through six pairs of wires seems a little much, but this isn’t the larger connector. In fact, I measured the pin sizes here and found that the pins are around 0.5mm with 3mm pitch – that’s the distance between pins – compared to 1mm pins on the standard 8 pin connector. That means the pins themselves are just a quarter of the size – while being asked to draw FOUR TIMES MORE CURRENT THROUGH THEM. In fact, working out how much current each pin is drawing here, 600W divided by 12V to get the current gives 50A. 50A divided by 6 for the number of pairs gives 8.3A per pair.

Now, looking up the current rating for 0.5mm wires from JST, one of the major standards and connector makers, they say that 0.51mm wires should only carry 2A safely. Oh, oh dear. Now it’s worth noting that that rating is for wires not connectors, although for 0.5mm2 wires they aren’t meant to exceed 5A. It’s also worth noting that the tiny wires that come out of these adapters need to be 1.3mm in diameter to safely carry 8.3A. Now I should note that the 12VHPWR connector is actually made by Amphenol, and they say the connector is rated for up to 12A per pin, meaning NVIDIA does have a good safety margin here, but compare that to JST and MOLEX who, for the same 3mm pitch, say they’d only want 4A or best case 6.5A. That’s something to keep in mind.

But, you know what? I’ve got a 4090 and a connector right here, let’s test it and see if it blows up. I’ve attached some K type thermal probes to the connector and left one in ambient air for reference. I’ll fire up FurMark to get the GPU pumping all 450W it needs through this tiny thing, and.. Oh wow that’s jumping up pretty quick. I’ll speed up the footage a little here but wow it went from 30°c to 40°c, 45°c, oh, 50°c… It’s… It’s gonna blow! Oh my.. Just kidding. This is definitely warmer than I’d like, but remember this cable is strapped to a board outputting 450W of heat which is soaking into the PCB and the connector. In fact, in all my testing with the 4090 so far I haven’t had any problems, and honestly I don’t expect to.

But.. that doesn’t mean I think this new connector isn’t a problem. Here’s a bit of conjecture for you. You know that reddit post? The OP also linked a picture of how they had their GPU which was in a vertical mount. The cable is pulled sideways out the card, and if I look at my connector I’ll find a fair bit of play side to side. You can actually see the connector lifting up a little with very little force on the cables. If you look at the burn pattern you’ll notice it’s mostly the pins on the outside. My speculation – and it’s just that – is that because the connector is so short compared to the standard 8 pin – the insertable length is just 6.5mm instead of over 9mm – and so wide compared to its depth, and thanks to such tiny pins and high power, any level of bad contact will cause resistance and therefore heating, potentially to the point of burning.

Does that mean we need to avoid vertical mounting these cards then? Well, not specifically. Any position that stresses the connector should be avoided, regardless of the card’s orientation. The fact the adapters are so short and need three or four 8 pin connections isn’t helping the situation, although I’d expect when PSUs can offer dedicated 12VHPWR connectors this might be a lot less of a problem.

I don’t think all that many people will have significant problems with these connectors, although considering within weeks of the launch we’ve seen multiple reports of failures, I would worry about what will happen in a few months or even years down the line. Will these prematurely fail compared to standard 8 pin cards? Will warranties be honoured in a year or two’s time? And I still don’t think this 12VHPWR connector is a good choice, nor the future of GPU power connectors. The power connector shouldn’t be something you should even need to consider as a potential failure point. It should be the most reliable part on the card, but I can’t say this connector is. I’m not sure I’d be comfortable having a card with that connector in my system, although it is worth noting that part of the problem here is that these are 4090’s. They are drawing ¾ of the power this connector is designed to handle, compared to say a 4080 that’s more likely to need more like 300W from it, or 4A per wire which is much more in line with JST and MOLEX’s recommended specs for connectors of that size.