ScrizzYep, not a problem here.

Is your GB RTX 4090 with NTK or ASTRON?

ZoneDymopretty amazing stuff, how hard is it to develop a plug....hard apparently.

Im still in favor of an update to the PCI-E slot so it can carry all the power needed.

Do you realise how much that would cost in Motherboard and connector costs?? You thought PCI-E 5.0 Mobos were expensive? Wait until they basically double the copper requirements to cope with this.

Panther_SeraphinDo you realise how much that would cost in Motherboard and connector costs?? You thought PCI-E 5.0 Mobos were expensive? Wait until they basically double the copper requirements to cope with this.

His point still stands: a slot that's unable to power the cards it services and needs ever more ingenious solutions for additional power is in dire needs of an update. Whether that's feasible at this moment or not, that's another story.

LabRat 891Having actual experienced professionals publicize their research and results allows cross-referencing with the "baseless opinions" of the "aparently qualified".
The egotistical ignore they're wrong, but the genuinely inquisitive learn something new; something that otherwise would've taken years of dedicated work/study (or more).

The 5600X3D discussion on TPU is an example, IMHO:
Some of the conjecture from 'armchair experts' proved out correct, or *very* close to what was revealed just before and with the announcement.

Just because someone is 'un(der) qualified' doesn't change how accurate/inaccurate their opinions and statements are. (vis-a-vis for 'the well-qualified expert' that history often proves wrong).

Some things are honestly just very simple if you have the skill to read between the lines.

Additionally, many things are remarkably similar, as history repeats. Qualifications only teach you the actual technical knowhow, but nothing about the world around it and how it moves, and has always moved. Qualified engineers are nice if you need to actually build something right.

There is one major problem with un(der)qualified experts. Everyone thinks they are an expert these days, but the reality is, the vast majority just isn't smart (note, I think TPU has a rather different demographic), but have the ability to click a few search results and skim through them, missing half the context along the way.

Example. When these things started melting, the playing it down move ('it happens to PCIe pins too so this is fine', yadayada) was straight up armchair expert nonsense, placed against other armchair experts seeing logic in the initial findings. This is where context comes in: the connector is smaller, there is more current, so the headroom for any kind of mismatch / bad contact / whatever else reduces the quality of the connection is going to have a much greater impact. The next hurdle was 'is the user to blame'... I remember the same advocates of the PCIe comparison parrot on about that one. User error, that must be it, so haha you're an idiot and everyone else is better. Today we know the initial impressions of those who just saw things for what they were, were spot on: the plug is a bad engineering job.

It'll always be a difficult question for us... when can we follow our gut, and is it ever wrong? I'll admit the gut isn't always right, but its definitely not baseless. You feel something, it has a reason. The real quest is finding the real reason :)

Just got my Seasonic 2x 8-Pin to 12VHPWR cable, but i'm anything but impressed.

It's not even using 3-dimple terminals. Though i guess it's fine for a 4070 Ti.

Those look like ASTRON female pins with the dimples, there will be two seams in the pin (unless they have been revised), the NTK female pin only has one seam.



As for powering an RTX 4070 Ti with this cable, it should be fine, most of the RTX 4070 Ti cards are limited to 320 watts max in the BIOS, there are a few that have a max limit of 350 watts though.

TheDeeGeeJust got my Seasonic 2x 8-Pin to 12VHPWR cable, but i'm anything but impressed.

It's not even using 3-dimple terminals. Though i guess it's fine for a 4070 Ti.

Those pins look like they are sticking out of their holes, is that true? If that's the case, you can actually press the connector against a metal surface (such as the inside of the case) and make a short circuit.

STSMinerThose look like ASTRON female pins with the dimples, there will be two seams in the pin, the NTK female pin only has one seam.

Fun fact: NGK and NTK are the same company.
I find it funny that (IMO) one of the best brands for automotive spark plugs also was the one company that didn't F-up their pin design.
It's almost like they have experience with spring/friction retention of electrical connections, in imperfect conditions (like an engine bay). :laugh:

LabRat 891Fun fact: NGK and NTK are the same company.
I find it funny that (IMO) one of the best brands for automotive spark plugs also was the one company that didn't F-up their pin design.
It's almost like they have experience with spring/friction retention of electrical connections, in imperfect conditions (like an engine bay). :laugh:

Not sure if they are the same company / related, but anything is possible.

bugHis point still stands: a slot that's unable to power the cards it services and needs ever more ingenious solutions for additional power is in dire needs of an update. Whether that's feasible at this moment or not, that's another story.

With the size that cards are becoming I am starting to think it would be worht looking at perhaps drawing power off 2 slots next to each other.

The other thing that could be done is an extenstion to the front of the PCI-e slot to give more dedicated power pins so the slot could provide say 150 watts.

I worry about trying to pull 600+ watts on consumer boards just purely due to the extra layers/copper content that would be required. We all complained how DDR5 boards suddenly jumped up $100 due to the tighter restrictions in the signalling. Well trying to pull 600+ watts through the board on top of the CPU power draw will incur more power planes being required and probably a similar jump in pricing again due to the extra layers.

Panther_SeraphinWith the size that cards are becoming I am starting to think it would be worht looking at perhaps drawing power off 2 slots next to each other.

The other thing that could be done is an extenstion to the front of the PCI-e slot to give more dedicated power pins so the slot could provide say 150 watts.

I worry about trying to pull 600+ watts on consumer boards just purely due to the extra layers/copper content that would be required. We all complained how DDR5 boards suddenly jumped up $100 due to the tighter restrictions in the signalling. Well trying to pull 600+ watts through the board on top of the CPU power draw will incur more power planes being required and probably a similar jump in pricing again due to the extra layers.

Yes, this will add to costs to the motherboard, ASUS and MSI have prototypes with the power delivery to GPU's via a special slot, but they had to also adapt the GPU for it.




I don't like the idea of this personally.

usinameThere is fundamental problem with the 12/16/2x6 pin connector

They keep ignoring that for some reason. Bet this new one is gonna melt as well, it's physics.

STSMinerNot sure if they are the same company / related, but anything is possible.

www.ngkntk.com/

Wikipedia:
Niterra Co., Ltd. (⽇本特殊陶業株式 社, Nippon Tokushu Tōgyō kabushiki gaisha) is a public company established in 1936 and based in Nagoya, Japan. Formerly known as NGK Spark Plug Co., Ltd., the company’s automotive business (with its brands NGK Ignition Parts and NTK Vehicle Electronics)

www.ntktech.com/

Copyright © 2023 Niterra NTK Technologies, Inc.

STSMinerYes, this will add to costs to the motherboard, ASUS and MSI have prototypes with the power delivery to GPU's via a special slot, but they had to also adapt the GPU for it.




I don't like the idea of this personally.

I do not either. That would be an immense step backwards.
AGPpro and Apple's custom AGPpro slots did exactly that.
(Server/Industrial board manufacturers inconsistently use an 'extra inline slot' for power too. Usually, for proprietary riser assemblies)

bugOr, here's another wild idea: maybe this is how the industry has always worked. Someone builds something for the first time and then, based on the first implementation(s), we iron out the kinks and develop a standard.

Except the "kink" in this case is a faulty connector that's a literal fire hazard on a $1600+ dollar GPU. "Ironing out the kinks" should've been done for at least a year before the connector was rolled out -- long before using it on your entire line of products. With that line of thinking, I guess the Galaxy Note 7 should've never been recalled -- after all, it was only a "small number of units" that were affected and Samsung were just "ironing out the kinks" with their batteries, right?

Also it's hilariously ironic how Ngreedia didn't have the "courage" to roll out the latest DisplayPort 2.1 standard on their overpriced $1600 GPU (using outdated DP1.4a ports and pocketing those few pennies), but beta-testing a flammable power connector was perfectly OK.

Several things could be done, and I think a transition to a higher voltage for the GPU would be among the most justified ones. In such a case, the voltage would be negotiated between the PSU and the GPU, similiar to what USB PD is designed to do.

WirkoSeveral things could be done, and I think a transition to a higher voltage for the GPU would be among the most justified ones. In such a case, the voltage would be negotiated between the PSU and the GPU, similiar to what USB PD is designed to do.

I immediately have hesitation with that idea, but:
-48VDC is already a common standard in computing and telecommunications. Though, I'm not sure of the full potential consequences that would come up with an inverted polarity and higher voltage.
(negative voltage rails have been on/off 'a thing' in PCs, but have almost always been fairly low-current)

Funny thought: If they went with 24VDC, I could see tinkerer-makers modding Power Wheelchair chargers into dedicated GPU PSUs.

caroline!They keep ignoring that for some reason. Bet this new one is gonna melt as well, it's physics.

'The numbers' say that *if the current can be equally-distributed*, there's more than enough 'meat' to the connector for .6KW+. (I think that's what the standard was trying to facilitate, with the sense pins)
OTOH, Real-life (as opposed to 'on-paper') physics tells us this is a very bad idea.
IIRC, this is why it's HIGHLY DISCOURAGED to use multiple wires in 'AWG-equivalency' for both DC in cars and AC in mains wiring. Even with a de-rating, factors come up that can suddenly change resistance on one of the multiple conductors; potentially causing a cascading failure.

Am*Except the "kink" in this case is a faulty connector that's a literal fire hazard on a $1600+ dollar GPU. "Ironing out the kinks" should've been done for at least a year before the connector was rolled out -- long before using it on your entire line of products.

It's not the connector that's faulty, Nvidia proved that with the first gen connector (does not have the 4 sense pins) works just fine with the RTX 30x0 FE series using the NTK female pins in the cable that came with the RTX 3080 FE, RTX 3080 Ti FE and RTX 3090 FE and then with the RTX 3090 Ti FE and all the Nvidia board partners that provided the same power cable with the NTK female pins in it with their RTX 3090 Ti cards.
None of these cards had the melted connector issue, not even the RTX 3090 Ti that was basically a test run for the RTX 4090.

For some reason someone decided to use the ASTRON female pins in the cable that was supplied with the RTX 4090 cards, those ASTRON pins did not get tested properly to see if they could handle the power draw and make a proper connection with the male pin.

We also got other 3rd parties using this ASTRON female pin and seeing the connectors melting (PSU, CableMod cables, CableMod right angled adapter, cables from Seasonic 8x2 to 12+4 etc).

STSMinerIt's not the connector that's faulty, Nvidia proved that with the first gen connector (does not have the 4 sense pins) works just fine with the RTX 30x0 FE series using the NTK female pins in the cable that came with the RTX 3080 FE, RTX 3080 Ti FE and RTX 3090 FE and then with the RTX 3090 Ti FE and all the Nvidia board partners that provided the same power cable with the NTK female pins in it with their RTX 3090 Ti cards.
None of these cards had the melted connector issue, not even the RTX 3090 Ti that was basically a test run for the RTX 4090.

For some reason someone decided to use the ASTRON female pins in the cable that was supplied with the RTX 4090 cards, those ASTRON pins did not get tested properly to see if they could handle the power draw and make a proper connection with the male pin.

We also got other 3rd parties using this ASTRON female pin and seeing the connectors melting (PSU, CableMod cables, CableMod right angled adapter, cables from Seasonic 8x2 to 12+4 etc).

That's not true, though. The RTX 4090 was way closer to testing the limits of the connector than the 3090 was (mostly seemingly due to much higher transient spikes -- even than the 3090 TI, which was already nuts) -- that's why the issue wasn't immediately evident with last gen. NTK connectors still melted just like Astrons, as the cable heats up and slowly ejects itself from the connector over time. Any off the shelf PC buyer that buys pre builds from the likes of Dell is never going to check for this and Nvidia can't blame that on user error, since the PCIE connector doesn't do this and never has (due to the PCIE connector being way over-engineered), nor does Nvidia warn people to check for this past installation. My friend was running one of these cards and we both noticed this (I didn't believe him until I saw other people confirm this on Reddit). He had to remember to push in the connector at least every month after heavy use, as we both noticed it wiggling out slowly after heavy usage. Eventually he had enough and sold the card back in April, as this wasn't worth the headache (and I can't blame him either). It's not a one-off occurrence, nor is it with a specific manufacturer (although the Astron ones were of poorer quality than the NTKs, that's for sure).

STSMinerThose look like ASTRON female pins with the dimples, there will be two seams in the pin (unless they have been revised), the NTK female pin only has one seam.



As for powering an RTX 4070 Ti with this cable, it should be fine, most of the RTX 4070 Ti cards are limited to 320 watts max in the BIOS, there are a few that have a max limit of 350 watts though.

I did mail Seasonic, see what they have to say.

I will also check my 2-way NVIDIA adapter tomorrow, see how that looks inside. I've been using that ever since i got my 4070 Ti. Initially i was going to go for a CableMod 2x 8-Pin to 12VHPWR cable, but to this day they still don't sell the Pro ModFlex version of that. Only inside the configurator for triple the price. And since they're not using NTK either i decided to get the Seasonic, because pictures i saw looked like it had single-split terminals, but i guess that's not the case after all.

Not sure if NTK has been adopted yet actually.

Am*That's not true, though. The RTX 4090 was way closer to testing the limits of the connector than the 3090 was (mostly seemingly due to much higher transient spikes -- even than the 3090 TI, which was already nuts) -- that's why the issue wasn't immediately evident with last gen. NTK connectors still melted just like Astrons, as the cable heats up and slowly ejects itself from the connector over time. Any off the shelf PC buyer that buys pre builds from the likes of Dell is never going to check for this and Nvidia can't blame that on user error, since the PCIE connector doesn't do this and never has (due to the PCIE connector being way over-engineered), nor does Nvidia warn people to check for this past installation. My friend was running one of these cards and we both noticed this (I didn't believe him until I saw other people confirm this on Reddit). He had to remember to push in the connector at least every month after heavy use, as we both noticed it wiggling out slowly after heavy usage. Eventually he had enough and sold the card back in April, as this wasn't worth the headache (and I can't blame him either). It's not a one-off occurrence, nor is it with a specific manufacturer (although the Astron ones were of poorer quality than the NTKs, that's for sure).

I've not see this issue, I have an RTX 3090 Ti here.

Google search does not come back with any results of this issue either, and I've not seen any of the major news sites (like this one) carry that news.

Even with that said, where are the melted connectors that use the NTK female pins ? I can not find any reports on the internet for them (RTX 3090 FE, RTX 3090 Ti or RTX 4090) !

Am*That's not true, though. The RTX 4090 was way closer to testing the limits of the connector than the 3090 was (mostly seemingly due to much higher transient spikes -- even than the 3090 TI, which was already nuts) -- that's why the issue wasn't immediately evident with last gen. NTK connectors still melted just like Astrons, as the cable heats up and slowly ejects itself from the connector over time. Any off the shelf PC buyer that buys pre builds from the likes of Dell is never going to check for this and Nvidia can't blame that on user error, since the PCIE connector doesn't do this and never has (due to the PCIE connector being way over-engineered), nor does Nvidia warn people to check for this past installation. My friend was running one of these cards and we both noticed this (I didn't believe him until I saw other people confirm this on Reddit). He had to remember to push in the connector at least every month after heavy use, as we both noticed it wiggling out slowly after heavy usage. Eventually he had enough and sold the card back in April, as this wasn't worth the headache (and I can't blame him either). It's not a one-off occurrence, nor is it with a specific manufacturer (although the Astron ones were of poorer quality than the NTKs, that's for sure).

Ah, someone else that's observed self-unseating 'previously, fully-retained' connections.

More than a couple times, I've had internal DC connections in a PC come loose after AC power failures. Semi-related: I just-yesterday, I had an AC IEC plug/receptacle come loose in a power surge (that also killed the UPS, but not the connected-equipment).
I can't blame my instances completely on thermal-cycling (assuming electromechanical/magnetic forces ???) but you're definitely not crazy.

I wouldn't be surprised if the rapidly-shifting (high moment-to-moment deltas) current consumption of a GPU is part of this 'wiggling out'. IIRC, that's part of the reason these (and other high-current interconnects) use a clip-in/on retention mechanism. Clearly, that mitigation is insufficient in some scenarios.

LabRat 891I immediately have hesitation with that idea, but:
-48VDC is already a common standard in computing and telecommunications. Though, I'm not sure of the full potential consequences that would come up with an inverted polarity and higher voltage.
(negative voltage rails have been on/off 'a thing' in PCs, but have almost always been fairly low-current)

Negative voltage is not a big technical issue; but it's also unnecessary.
A variable voltage, for example 12V/24V, is also not something complex to implement. However, PSU manufacturing cost would go up because it would need to supply two voltages, one fixed 12V for the mobo and one variable for the GPU.

A side note ... 12VO for mobos doesn't seem to become mainstream anytime soon but if it ever does, everyone will suddeny discover, surprise surprise, that we actually need more volts there.

LabRat 891'The numbers' say that *if the current can be equally-distributed*, there's more than enough 'meat' to the connector for .6KW+. (I think that's what the standard was trying to facilitate, with the sense pins)
OTOH, Real-life (as opposed to 'on-paper') physics tells us this is a very bad idea.
IIRC, this is why it's HIGHLY DISCOURAGED to use multiple wires in 'AWG-equivalency' for both DC in cars and AC in mains wiring. Even with a de-rating, factors come up that can suddenly change resistance on one of the multiple conductors; potentially causing a cascading failure.

I don't know but it can't be just universally discouraged. PC connectors have had multiple parallel wires at least since the IBM PC XT (main power was through three +5V and four grounds, other voltages were auxiliary).

De-rating is a thing however, and I am amazed that no one has mentioned it so far. If one contact pin and one wire can carry 12A, six of them in parallel can't carry 72A because you can't guarantee an exact, even distribution of current. So, to give an example, the capacity drops to 6 x 11A. Then there's another obvious reason for de-rating: heat density in connectors. So six pins close together can't carry 6 x 11A but only 6 x 9A if the connector is expected to operate at up to, say, 60°C ambient temperature.

Here's an example from Molex's documents for Micro-fit pins (I'm not sure if the 12+4-pin PCIe connector counts as 6-circuit or 12-circuit):


I found another interesting bit of information at Molex. They make female pins with dimples (only two on each of the two sides, four in total!) as well as female pins with springs on each of the four sides. The latter is marketed as RMF or Reduced Mating Force. Available with lubricant too but let's leave that for another day. So I must assume that one type of pin is not necessarily inferior or less reliable than the other, but there's any number of details that Astron could have screwed up, such as manufacturing tolerances, the choice of materials, thickness of metal and so on.

Ah ! So there is a 3rd player now with this female pin with only 4 dimples, that makes sense now with that Seasonic cable (8x2 to 12+4).

STSMinerAh ! So there is a 3rd player now with this female pin with only 4 dimples, that makes sense now with that Seasonic cable (8x2 to 12+4).

Molex is probably too expensive, as is Amphenol in the example I gave another day (another type of high-current connector). But of course there can be several more cost-cutting players around, not just two.

STSMinerYes, this will add to costs to the motherboard, ASUS and MSI have prototypes with the power delivery to GPU's via a special slot, but they had to also adapt the GPU for it.




I don't like the idea of this personally.

Oh please no, horrid idea.