NVIDIA CEO Jensen Huang hints at ‘exciting’ next-generation GPU update on September 20th Tuesday

I really don't understand the 12VHPWR connector. Usually when you are dealing with increased current (amps), you want to increase the size of the connectors and/or increase the number of connectors (split the current between a larger number of connectors). The 12VHPWR connector seems to have gone the opposite route on both of those fronts. This goes against basic elementary-school knowledge, the kind of thing you learn when you are 8 years old using batteries with your RC car, etc. Spreading the load between three or four 8-pin PCI-E power connectors mounted directly onto the card seems like it would have been a much better way to go, instead of essentially creating an artificial bottleneck. But surely the engineers who designed the 12VHPWR connector knew what they were doing, right? So what am I missing?

The one and only "advantage" that I can think of related to the 12VHPWR connector is decreased production costs for the cards... So is it really that simple? Cards burning up because they wanted to save a dollar by consolidating everything into one connector?
 
I really don't understand the 12VHPWR connector. Usually when you are dealing with increased current (amps), you want to increase the size of the connectors and/or increase the number of connectors (split the current between a larger number of connectors). The 12VHPWR connector seems to have gone the opposite route on both of those fronts. This goes against basic elementary-school knowledge, the kind of thing you learn when you are 8 years old using batteries with your RC car, etc. Spreading the load between three or four 8-pin PCI-E power connectors mounted directly onto the card seems like it would have been a much better way to go, instead of essentially creating an artificial bottleneck. But surely the engineers who designed the 12VHPWR connector knew what they were doing, right? So what am I missing?

The one and only "advantage" that I can think of related to the 12VHPWR connector is decreased production costs for the cards... So is it really that simple? Cards burning up because they wanted to save a dollar by consolidating everything into one connector?

I think the main issue was that with how small the PCB was on the 4090 that they physically couldnt fit 3 or 4 8 pin connectors on it. So they did a dry run with the 3090Ti to see how it would work. Then the problems began once they started really pushing the juice through it. Plus with how big the card is, most cases just cant fit it and avoid any bending.
 
We're up to around 10 melted cable reports now. Something will have to be done.
I suspect Nvidia will release a new adapter.
 
We're up to around 10 melted cable reports now. Something will have to be done.
I suspect Nvidia will release a new adapter.
Good.

I got a brand new case for my 4090 (Fractal Pop XL Air), and the cable is still being bent against the side panel. As of right now, I have my side panel off and will keep it that way until a solution is found.
 
Maybe, going forward. Won't affect the cards already in the wild.
It's fine. Only a very few cards. How many 8-pin PCIe cables fried? Hmmmm? I suspect it was a fairly large number since they have been around for quite a while. There's a nice market for anti-NV these days. Not surprising since so many are butt-hurt over the prices the last few years.

Anyway, I'll own up to being wrong if this post ends up being proven wrong by unseemly numbers of adapter failures on 4090s that NV owns up to.

IOW, let this play out.
 
So Nvidia and AIBs designed a card that one cannot put in most cases without bending the power cable beyond specification. Seems like the new pcie 5 power cable plug spec is much more limiting and unforgiving when not properly inserted/bend radius. Hence the card design and I would have thought that the adaptor would at least have some 45 Degree plugs, 90 Degree and better yet a attachment point and support for the new cable to prevent this, like a 90 degree or less box/extension on card for the cable. Tabs on the card to support cable with tie wraps for a cheap design (at least something). Now for new owners, I would recommend actually checking the temperature of the plug while at power to see if it is getting excessively hot, better yet a contact thermometer, infrared imagining if one has one. Also make a support that keeps the cable weight and force off the plug.

150w pcie plugs to 600w pcie 5 plugs with smaller pins and contacts (much less surface area) which any slight resistance, poor pin contact will heat up the connector (I^2R losses, current through resistance -> gives you heat) which being smaller itself will also have less surface area to dissipate the heat.
 
Good.

I got a brand new case for my 4090 (Fractal Pop XL Air), and the cable is still being bent against the side panel. As of right now, I have my side panel off and will keep it that way until a solution is found.
Cablemods 90 degree connector on a custom cable for your specific powersupply. Comes out Oct 31st
 
They really should have mounted it the same as they did on the 3090Fe. It used early version of this connector (didn't have the sense pins), and it came out of the card at an angle. 0 issues and my card pulls up to about 417 Watts. And the current cable I have only has 2 8 pin power supply side connectors. I think the connector itself is ok, but they should have either used an angled connector on the card, or made those adapters with a 90 degree 12vhpwr connector.

More power in a smaller area, so it was bound to happen. The old 8 pin connector was more forgiving since those can handle 332W per cable but are only counted on supplying less than half that (150W). Even the new connector is electrically adequate with some overhead, but it should have been obvious that they needed to come with 90 degree connectors.

Growing pains...
 
the connector (I^2R losses, current through resistance -> gives you heat)
More heat if current achieve to stay the same (the more important variable), V=RI would mean it require for the tension to be raised and the current to go through the hardest route instead of the easier pin.

I could be all 100% out of touch here, for an important disclaimer, but what is more likely to happen is when I pin get more resistance via a bad contact, more ampere current use the other available less resistive pins and it is the well connected pin that melt because of that imbalance, the high resistance one being fine.



You see at 8:29 the SIG study back in the days, were they monitor the 6 pins, the pin with the least ohm is the one getting the most current and being the variable that is square for heat generation by far the more power-heat.

it seem to be constant highest the resistance, cooler the pins get
 
Last edited:
Last edited:
Cablemods 90 degree connector on a custom cable for your specific powersupply. Comes out Oct 31st
Preorders start October 31.

1666876131190.png
 
Gotta make sure all cards are at the HQ so no contradictory evidence shows up to Nvidia's narrative?
They use this connector on the RTX workstation cards and the new A100 series as well. You better believe if they cause fires there a class action from the likes of us is barely a blip on their “oh shit” radar.
 
LOL. The real reason they didn't go with 600W is that the connector would catch fire even if it wasn't bent.
 
The 3090TI had the same connector and power demand, which open speculation of a possible fix on the next batch of cards I would imagine, the difference would be according to reddit:
3090ti (and similar) use multiple 12V power planes that balance the current across the different pins on the connector, while the 4090 uses a single 12V power plane that can result in excessive current across and individual pins of the connector when the resistance between the pins is significantly different.
 
I think it's funny people are still going out and buying these things even when they know it has a serious problem with the power connecter. Guess they got to have the best even if it burns their home to the ground.
Well, hardly any more 4090 inventory has been scheduled into NA over the next couple of quarters. NV shot their wad, got the press for having the fastest card on the market, and now will shift their focus to running out high end Ampere inventory so that it can fit the 4070 into the stack. Incoming inventory for the 4080 cards look to be about 75% of what 4090 was at launch.
 
Well, hardly any more 4090 inventory has been scheduled into NA over the next couple of quarters. NV shot their wad, got the press for having the fastest card on the market, and now will shift their focus to running out high end Ampere inventory so that it can fit the 4070 into the stack. Incoming inventory for the 4080 cards look to be about 75% of what 4090 was at launch.


3090 for $10 when?
 
I can't help but think about the recall clip in Fight Club,

Take the number of vehicles [GPUs] in the field, A. Multiply it by the probable rate of failure, B. Multiply the result by the average out-of-court settlement, C. A x B x C equals X. If X is less than the cost of a recall, we don't do one."
 
Nothing wrong with undervolting if you do it right.
That’s the case for about everything, just too many people are convinced their way is the right way.

I’m too tired to sort it out, I’ll just keep things stock and play what ever I’ve been told I’m playing. Which is Gotham Knights this week, it’s been fun. The team needed a Robin so it’s me.
 
Back
Top