News Nvidia Seemingly Uses Revamped Power Connector for RTX 4070 FE

[edzieba]

no your arguement is plain wrong, able to yank it loose while the clip is in place IS a design flaw, a well designed clip would not allow that and will unclip when the yanking is hard enough to make it loose.

By that logic the 6-pin and 8-pin PCIe power connectors, EPS12V, and ATX connectors (all Molex Mini-fit Jr) have the same design flaw, as to PCIe card-edge slots and DIMM slots. All will release if pulled on whilst wiggling.
These retention latches are designed to be proof against unintentional dislodging, not against deliberate malice. Pulling and wiggling a connector is the latter.

 

[YSCCC]

By that logic the 6-pin and 8-pin PCIe power connectors, EPS12V, and ATX connectors (all Molex Mini-fit Jr) have the same design flaw, as to PCIe card-edge slots and DIMM slots. All will release if pulled on whilst wiggling.
These retention latches are designed to be proof against unintentional dislodging, not against deliberate malice. Pulling and wiggling a connector is the latter.

No again, all those other connections you can't (or at least that easily) yank it to a point where it could melt AND with the clip still in place, or the slots you can't yank it out without physically breaking the DRAM/Card PCB. that is what called a proper design, say for 8Pin plugs, you either never actually let the clip click, or you pull out enough the clip breaks/ unclip, once set in, the cable management pull and bending within 30mm clearance from the plug do not pull it out and cause the melting, which the original 12VHPWR connector allows with the clip still "locked" but the contact is poor enough to cause melting.

By logic, all those other plugs and slots don't have remotely close in numbers as many cases when it first release or being used over a decade vs this brand new advanced standard, and the exact same users who built PC for decades do not suffer from melting, yet in the 4090 suddenly all those long term PC builders have quite a portion of melting connector. Even those with cable mod 90degrees connector did melt, those who buy sure enough would make sure they plug it in enough to let the clip engage and yet still some cases of melting.

And as Northridge repair have noticed, all those adapted melting cases the melting point is always at the socket on the card, not on the cable-> adapter end, that alone is sound proof the design is flawed at least on the card-> cable interface. if not, why it's always the card end melting, at least one case of melting in the PSU->cable end which uses the same spec and not the cable->adapter end

 

[TJ Hooker]

Gamer Nexus and Jayz did demoed it can be yanked.

no your arguement is plain wrong, able to yank it loose while the clip is in place IS a design flaw, a well designed clip would not allow that and will unclip when the yanking is hard enough to make it loose.

Provide the links (with timestamps for relevant parts, if they're long videos).

The gamers nexus video I'm familiar with demonstrated it can be yanked out when not fully inserted/latched.

Starts around 25 min in:

View: https://m.youtube.com/watch?v=ig2px7ofKhQ

 

[TJ Hooker]

No again, all those other connections you can't (or at least that easily) yank it to a point where it could melt AND with the clip still in place, or the slots you can't yank it out without physically breaking the DRAM/Card PCB. that is what called a proper design, say for 8Pin plugs, you either never actually let the clip click, or you pull out enough the clip breaks/ unclip, once set in, the cable management pull and bending within 30mm clearance from the plug do not pull it out and cause the melting, which the original 12VHPWR connector allows with the clip still "locked" but the contact is poor enough to cause melting.

Source for the connector melting even with the clip engaged?

 

[YSCCC]

Provide the links (with timestamps for relevant parts, if they're long videos).

The gamers nexus video I'm familiar with demonstrated it can be yanked out when not fully inserted/latched.

Starts around 25 min in:

View: https://m.youtube.com/watch?v=ig2px7ofKhQ

View: https://youtu.be/RT89QmrNisE

5:41 onwards

That’s how it easily unseated slightly with the clip didn’t declick.

The issue is that is virtually impossible to prove it’s impossible to prove that is x% of the case as the pull out partly with the clip still in is so easy and no easy way to identify it. The ddr and PCIe latch will have your whole card pop out slight when disengaged, the old 8pin connector have a big clip unlock handle which when partly unseated will be very visible for not in the clipped angle.

While the new connector just need to shine a torch in and look carefully for any gap, presumably better everytime you open the case and do something? That’s why msi come out with a yellow pin connector and now NVIDIA is secretly revising the socket, and Intel have proposed a revision of the internal contact change. When these can be done so easily to warn the users, they should be there before releasing.

 

[cryoburner]

I still remember that dual GPU cards maxed out at 500Watts only using two high end GPU chips ... now a single card needs more ... Nvidia and AMD are not doing their homework right anymore

The number of GPU chips on a card doesn't matter as much for power consumption as how many total cores it has, and how high they are clocked. The chip used in the 4090 is more than double the size of the one used in the 4070 and 4070 Ti, and in turn has more than double the cores of those cards. All the cores being on one chip does allow the card to utilize those cores more effectively though, getting more reliable performance out of them and allowing the card to scale better with that additional power.

Had they chose to slap two 4070 Ti chips on a single card for the 4090, the power consumption would have likely been similar, but the total performance would have been worse, at least if they were designed like those older multi-GPU cards, that generally tended to treat each of the graphics processors more like separate cards in SLI with limited communication between them. So this is really a more efficient solution than those multi-chip cards, allowing for increased performance at a given power level. I'm sure multi-GPU cards could also work efficiently if they were designed from the ground up with direct communication between separate processors in mind though, and we will probably see chiplet-based cards relatively soon, but that will be more about improving yields and allowing cards to scale further.

PSUs capable of delivering higher amounts of power are also more common these days, so companies like Nvidia are probably more comfortable with making cards that can utilize those higher power ceilings for a bit more performance for the niche market looking for the fastest card possible. It's likely that a 4090's efficiency could improve a good amount just by underclocking it though.

the 4060 is "technically" a pretty efficient gpu while using frame gen for the frames u get out of it.....if you ignore downsides & extremely small support for it.

Frame generation provides worse artifacts and high input latency at lower base framerates though, and it's questionable whether the 4060 has what it takes to properly utilize the feature in many of the demanding titles that might otherwise benefit from it the most. In many cases, you would likely be better off just sticking to upscaling to increase framerates instead, where the artifacts will tend to be less noticeable and latency actually improves. The feature might effectively help smooth performance in some CPU-limited titles that are not latency-sensitive, but the delayed filler-frames it generates are not a directly-comparable substitute for actual rendered frames.

 

[edzieba]

View: https://youtu.be/RT89QmrNisE

5:41 onwards

That’s how it easily unseated slightly with the clip didn’t declick.

That is CableMod's adapter (a part manufactured by CableMod, not by Molex or another certified OEM), not the actual 12-pin connector. If it's already violating design spec by adding 90° cable exit and clipped pins, there is no guarantee it is meeting spec when it comes to dimensional tolerance.

So thus far we're two-for-two on any actual evidence the 12-pin connector can be pulled out without yanking and wiggling (and one showing yanking and wiggling did not dislodge a correctly seated adapter, only one not fully seated) or without use of noncompliant adapters.

 

[YSCCC]

That is CableMod's adapter (a part manufactured by CableMod, not by Molex or another certified OEM), not the actual 12-pin connector. If it's already violating design spec by adding 90° cable exit and clipped pins, there is no guarantee it is meeting spec when it comes to dimensional tolerance.

So thus far we're two-for-two on any actual evidence the 12-pin connector can be pulled out without yanking and wiggling (and one showing yanking and wiggling did not dislodge a correctly seated adapter, only one not fully seated) or without use of noncompliant adapters.

View: https://www.youtube.com/watch?v=mKqC2wJIeEI

I suggest you watch on Northridgefix on why he believes it's a poorly designed connector, look at 18:12 where they replayed GN's video of wiggling the official adapter, that properly seated one have clipped in, and wiggling didn't pull the plug out, BUT if you pause it there, you could see the pins are showing and there's a slight gap on the connector-plug interface. That alone is a potential for poor contact and burning without the obvious sign of it being unseated.

And moreover as I've said previously, this sounds like the Boeing 737Max poor design MCAS defending all over again,

When better design which don't cost any more (shorter sense pin to stop sending huge current when not properly seated, different colour to hint it's not fully seated (should have warning/reminder in manual also), better pin engaging mechanism) than the original is there could warn the user, or just completely stop the said "user error" which is quite prone to happen and happen unnoticed, why on earth such better design is not in place as a proper standard is not a poor design and every blame is just the consumer, those fan like behaviour is questionable.

When same group of experience level users use a new product, a lot more and severe accidents occurs compared to the old standard, it's the new standard's problem, not like the "you are not holding it right" thing again.

 

[YSCCC]

And one more thing to add in as a consumer perspective: Even if I love the company which produce the said debated products, I would hope they just admit it's a bad connector and revise it ASAP on their newly produced products, why? If the new connector need a lot more care to make sure it don't burst into flames and render my PC useless for the RMA time period compared to competitions, and the competitions though being slower, is also much cheaper and still adequate for the user's needs, a certain % of users will just don't take the risk and trouble and go to the next best thing, worse still if the best product like the 4090 cost a lot more copmared to the 7900XTX.

Worst still is if you accodentally killed the connector for your $1600 GPU and the vendor just said: "O it's your error, no warranty coz it need to rock hard into the socket and check in detail again after you pull the cable for cable management". you will just lose market or at least sales to the customer, and selling less of a product will render the company produce less/slower refresh on products, which is a bad sprial towards the bottom. Don't tell me there isn't significant potential buyers of 4090 hold on to their card for another generation, or jump to options using the good old 8pin cards, this is not only physical, but also commercially bad design.

 

[edzieba]

And moreover as I've said previously, this sounds like the Boeing 737Max poor design MCAS defending all over again,

The difference is that MCAS full-endstop actuation from a faulty attitude sensor has a reproducible problem that has been independently reproduced.
In contrast, nobody managed to independently reproduce the burning behaviour with different connector models, soldered vs crimped connectors, different internal pin designs, and aligned/tangent pressing of the connector (all things that the finger was initially pointed at). Lots of youtube talking-heads theorising over the cause, but nobody actually manged to put a card on a test bench and make it smoke. However, once the 'connectors not seated' theory was advanced, the same behaviour was recreated reliably and in short order by multiple independent testers.

 

[YSCCC]

The difference is that MCAS full-endstop actuation from a faulty attitude sensor has a reproducible problem that has been independently reproduced.
In contrast, nobody managed to independently reproduce the burning behaviour with different connector models, soldered vs crimped connectors, different internal pin designs, and aligned/tangent pressing of the connector (all things that the finger was initially pointed at). Lots of youtube talking-heads theorising over the cause, but nobody actually manged to put a card on a test bench and make it smoke. However, once the 'connectors not seated' theory was advanced, the same behaviour was recreated reliably and in short order by multiple independent testers.

Which, the unseating can be done overtime with normal use of the computer, slightly bouncing the case during transportation, and adding or changing hardware over time. None of the melting cases are reported in the short time as the first power up issue in the bench of the YouTube tests, they go slow in days or weeks of heavy use. In that sense any reasonable design would have sense pin to just black screen your card and notify you that it’s unseated.

If it’s so poorly seated as in the GN extreme poor contact test, it will have ppl saying they don’t burn or they burn in their first hr of usage, which, isn’t the case.

The MCAS have been blamed on user error on the Malaysian pilots and claimed not replicable and not any design fault for a year and more before agencies step in and test rigorously. And this connector since it’s not stepped in by gov agencies, they just don’t got the legal sue as the Boeing

And if it’s just pure user error, the PCISIG would sure as hack just do the second throughout test, share the results and point at the users, which didn’t happen. What happen is they revised and reviewing a 6x2 variant to improve the plug, which is a better design