I believe it is entirely possible to implement a solid and safe 12VHPWR connector, and its reputation is being destroyed by a shoddy implementation... but that's very much nVidia shooting themselves in the foot.
The linked article gives very good insights in what’s going on.
Unfortunately NVidia has really screwed the pooch with this as this dangerous adaptor is the first encounter most people will have with the 12VHPWR form factor
However, you're missing the point. The spec should be designed such that even "non-reputable" manufacturers with sloppier manufacturing methods wouldn't create a fire hazard. These connectors are extra-delicate even though they're supposed to carry higher loads. They will be less robust in practice, for very little gain. That's a poor tradeoff, which makes it poor design.
These new adaptors are included with products Nvidia themselves ship, with themselves as OEM. They know this standard is not common, and they know they need to ship adaptors with them, If they can’t guarantee the quality of it’s manufacturing, then it says there’s possible more worrying things going on behind the scenes.
70 Celsius is NUTS for that usage, not ‘oh, that’s ok then’
It’s well outside normal expected operating temperatures for wire.
This data point isn't exhaustive, but it does indicate that the actual safety margin is not as tight as is assumed from this news cycle.
That brand new connectors and wire that are custom specced wouldn’t catch on fire (but come close!) at only 2x the power draw means it’s very likely any damage to a connector, wire, etc. especially with normal wire and connectors likely would cause damage and a fire, at much less than that power draw.
Which is what we’ve been getting reports on.
And that is just from minor physical damage it seems, ignoring corrosion or fraying wires over time which is usually the bigger problem.
The truth is the plug is more demanding and expensive than lower density options and it's dangerous when corners are cut. This is always true in power electronics, but the risk surface area is higher with a new and demanding connector.
Even with connectors and cables without those early problems, there are often longer term problems that show up over time - cables fraying at the connectors due to movement, plugs and connectors building up corrosion (and hence having higher resistance), connections loosening or getting bent, etc.
The link you posted, and your later statement seems to support the same view. I'm just repeating it so the folks dismissing this as just an issue with the badly designed adapters Nvidia distributed (in some cases?). It's overall a connector and spec that's on the edge of the performance envelope with some 'obvious' types of field failure modes that don't seem to be properly addressed.
Honestly, I'm a bit shocked they went with 'mountains of parallel power feeds' solution instead of... I don't know, doing 6 gauge stranded, which seems like the obvious choice to me? Parallel power feeds like this are always the source of endless fussing and headaches due to exactly the problems we're seeing. Trying to save a couple cents by using more of the same materials they have on hand I'm guessing?
Yes, if "the right materials" are used and the cables are treated "properly" (not bent out of spec, not too many connection cycles) then it's "OK", but those shouldn't be requirements designed into something that cost-conscious consumers are going to be assembling.
Does the old way of building PCIe cards still make sense when the GPU now draws more power than all of the other system components combined? GPU sag (i.e. the card weighs so damn much) is such an issue that one manufacturer actually put a level in their card.[0] The 4090 probably needs three slots just to keep it attached to the motherboard.
When someone buys a desktop these days, it's very often because they need a discrete GPU. You buy a dedicated desktop GPU because the thermal envelope of a laptop can't support a card that would draw 600W - even if it were possible to shrink the card into a laptop (or make the laptop big enough to fit the 600W card). The top of the line Intel processors draw in the neighborhood of 150W but that doesn't necessitate going to a different form factor. Future motherboard and south bridge designs should mount the GPU independently of the other components.
[0] https://www.tomshardware.com/news/geforce-rtx-4090-has-a-spi...
In your opinion, if you could force power supply makers to add another voltage rail, which voltage would you pick?
Solving for the problems of higher current, lower voltage DC we use in ATX format supplies today requires higher gauge copper, but the cable run between the GPU and PSU is so shot that the cost of extra copper could very well be less than the increased cost of a higher voltage supply.
where is this typical? in my experience, typical is 200A service for residential. is mains service amps dependent on provider/region/etc?
Often, those transformers on the pole have some wiggle room in what they can deliver, and just because you upgraded from 100a to 200a service doesn’t mean you’ll actually see a real world change in usage. You usually don’t. Utility companies can over provision that capacity because not everyone on a street switches on their microwave at the same time, etc.
I don't personally like EVGA, had a bad experience trying (and failing) to get warranty replacement for a defective PSU, but I can certainly follow GP's thinking about how EVGA's involvement with this launch might have made a difference.
I'm sure Nvidia will get it sorted in time, but they're trying to do a lot right now, and it's noticeably bumpy.