News 16-Pin Power Connector Gets A Much-Needed Revision, Meet The New 12V-2x6 Connector

[russell_john]

Or use 48V at 1/4th the current, but that would require a new PSU.

Do you realize how hard it would be to convert 48V at 450 Watts down to >1V at 450 watts.

For cripes sake graphics card are big enough already without increasing the size of the VRM components (including cooling). You like have to resort to a two step process from 48V down to 12V and then your normal 12V to >1V

 

[InvalidError]

Safety was never an issue with this. Flameproof plastic melting is not a fire hazard which is why they use flameproof plastics. You can point a blow torch at that connector and it still won't catch fire, all it does is melt.

94V-0 materials (what most electronics with valid fire safety certifications are made with and is a PCI-SIG requirement for HPWR connectors) are flame-retardant, not flame-proof. If you point a flame at them, they will burn as long as external heat is supplied and there is something left to burn. Remove the external heat source and they should self-extinguish within seconds.

In short, 94V-0 only means that materials cannot self-sustain a flame.

 

[InvalidError]

Do you realize how hard it would be to convert 48V at 450 Watts down to >1V at 450 watts.

Not that difficult: use transformers instead of chokes, then you can directly convert 1A@48V into 48A@1V instead of using buck regulator stages and you can even use cheeky little 6V FETs on the low-voltage side! Slightly more complicated but not tragically so.

 

[bit_user]

it is quite easy to over-torque it and strip the hole. I never bothered screwing cables in more than a few turns after my first stripped connector threads.

Oh, yes. Not only stripping the threads, but also torque the head off the screw-nut, or even break off the connector's screw inside the screw-nut.

Gosh, so many failure modes! It's a wonder those darned things kept getting used time and again!

 

[purpleduggy]

again with the molex...

 

[purpleduggy]

Whether the root cause for the melting connectors is/was "user error" or not doesn't really matter, at best they messed up by making the specification too loose and/or too hard to use properly/safely.

The retention clip on them are clearly not good enough for example, on many it's possible to wiggle the connector far too much with it inserted fully and latched, and the latch also lacks a "positive feedback" which the previous connector had. Arguably they should probably have gone to two latches for this wider connector but it was likely hard to do and stay compatible.

In short, it may well be that technically all the melted connectors ARE user errors, but that doesn't mean the PCI-SIG is blameless here or that they shouldn't improve it so that it's no longer a "gotcha" connector.

user error is a lie. the molex connector cannot handle the wattage. the connector is badly engineered. it doesn't seat properly. it is high torque because molex is a bad design that has been abandoned since the 70s in automotive and aerospace because it is a terrible design. it melts because the power it is tasked with supplying is above what it can handle. user error is a cope manufacturers use to explain away their bad engineering.

 

[purpleduggy]

We engineers have a saying that's at least 40 years old

"There is no such a thing as idiot proof because idiots are so damned ingenious"

If there is a way to screw something up an idiot is going to find it

this is engineer speak for when they are coping about their bad designs. automotive and aerospace engineers should teach computer engineers about how connectors have moved on from the abandoned ancient molex connectors of the 1970s. there is a reason why they are no longer used in automotive and aerospace and by the way, Computers only started using molex because it was already in use in aerospace and automotive. they need to get with the times and abandon molex's terrible design.

 

[InvalidError]

there is a reason why they are no longer used in automotive and aerospace

Different connectors for different applications. Automotive and aeronautical applications need to cope with extreme environments from vibrations and grease monkeys to weather and chemical exposure. If any of that got in your PC, there are pretty good chances it won't last long no matter how fancy your connectors are.

 

[purpleduggy]

Different connectors for different applications. Automotive and aeronautical applications need to cope with extreme environments from vibrations and grease monkeys to weather and chemical exposure. If any of that got in your PC, there are pretty good chances it won't last long no matter how fancy your connectors are.

molex was never made for PC hardware applications. so the idea that it is suited to the application isn't true. its was developed for aerospace and automotive in the 1950s and used up to the 1970. PC hardware started using it, probably first by IBM because it was cheaper than developing their own connector. so its not application specific. they just took what is already available. the automotive and aerospace industry stopped using molex because of its inherent failures and unreliability. there are far better connectors available today and will be far more suited to PC hardware and user friendly as has been proven with connectors on cars and aerospace, and at the same cost, probably even cheaper because they will not fail anymore like molex will.

Molex connector - Wikipedia

en.wikipedia.org

 

[purpleduggy]

To give you an idea of how far behind the times the Molex Mini-fit Jr. 39-28-1203 ATX specification is to the low torque, vibration resistant, moisture ingress proof, high temperature, triple insulated connectors used in automotive and aerospace. Listen how easily they seat with an ensuring click, and how they are made to connect and disconnect multiple times:

View: https://www.youtube.com/watch?v=4rW8zx-dYBc


why are we still powering advanced computer hardware with stone age connectors? because pc hardware engineers suck. ATX molex sucks. come out of the 1970s and join us in 2023.

 

[InvalidError]

To give you an idea of how far behind the times the Molex Mini-fit Jr. 39-28-1203 ATX specification is to the low torque, vibration resistant, moisture ingress proof, high temperature, triple insulated connectors used in automotive and aerospace.

Which of automotive connector features are of any benefit to PCs? None. You just get connectors that are 5X the size for a given job and no meaningful benefits. If you run a PC in a high-vibration environment, solder joints are likely to break all over the place since nothing in a PC is vibrations-rated. If you run a PC in a high-moisture environment, your components are going to short out from moisture ingress and corrosion because nothing in your PC is moisture-rated. Having ludicrously high temperature-rated connectors in your PC are pretty much pointless when the case temperature should hardly ever exceed 50C, maybe 60C at high-power connectors. Triple-insulating stuff is a waste of insulation at 12V with no high-voltage anywhere within reach. "Ingress-proof" connectors are only "ingress-proof" while connected, which is kind of pointless in PCs where any FODding is practically guaranteed to only ever occur while the connector is disconnected.

Also, if you split current 6-ways through a connector where each pin is operating at its rated limit, you'd likely still get the occasional melting connector from poor current balance no matter how fancy your connector is. Practically nothing else in consumer-serviceable tech uses multiple parallel pins and wires to carry non-trivial current.

 

[purpleduggy]

Which of automotive connector features are of any benefit to PCs? None. You just get connectors that are 5X the size for a given job and no meaningful benefits. If you run a PC in a high-vibration environment, solder joints are likely to break all over the place since nothing in a PC is vibrations-rated. If you run a PC in a high-moisture environment, your components are going to short out from moisture ingress and corrosion because nothing in your PC is moisture-rated. Having ludicrously high temperature-rated connectors in your PC are pretty much pointless when the case temperature should hardly ever exceed 50C, maybe 60C at high-power connectors. Triple-insulating stuff is a waste of insulation at 12V with no high-voltage anywhere within reach. "Ingress-proof" connectors are only "ingress-proof" while connected, which is kind of pointless in PCs where any FODding is practically guaranteed to only ever occur while the connector is disconnected.

Also, if you split current 6-ways through a connector where each pin is operating at its rated limit, you'd likely still get the occasional melting connector from poor current balance no matter how fancy your connector is. Practically nothing else in consumer-serviceable tech uses multiple parallel pins and wires to carry non-trivial current.

well for one, they don't catch on fire.

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

 

[InvalidError]

well for one, they don't catch on fire.

Most of them will if you get them hot enough and even GN's most successful melting attempt still didn't get hot enough to actually set the HPWR connector on fire.

 

[purpleduggy]

Most of them will if you get them hot enough and even GN's most successful melting attempt still didn't get hot enough to actually set the HPWR connector on fire.

its not the manufacturer, its actually the user's fault they installed it wrong. the user is always to blame. they should have just bought geforce now cloud gaming instead. or better yet buy a console. all the RTX4090s that caught on fire were the user. Nvidia and AIB partners will never design something badly. they are always innocent. their innocence is eternal. for they can never do any wrong. using a better connector is a bad idea, all the fires and melting are just because of users who don't know how to install them. its not a bad connector, its just balancing/bending/badly inserted. these are divine innocence interdicts that always shifts the holy blame to that of the user. how dare any of us accuse the holy light of nvidia of a bad connector. we should be ashamed to even think it. I'm so thankful we have nvidia and all the AIB saints who never do anything wrong or make any bad decisions. We don't deserve them. We users make so many mistakes and the holy nvidia always gets the blame. If only we unwashed peasants could hope to achieve a crumb of the same divinity some day, then we will have less 4090s on fire. Forgive us Jensen for we have sinned. Thy holy leather jacket, thy stage AI songs, as we art in greed, deliver us from our wallets. Amen.

"I am here at computex, I will make you like me best, yeah. Sing, sing it with me, I really like NVIDIA"

View: https://youtu.be/IlWT_TdOK6s?t=781

 

[InvalidError]

all the RTX4090s that caught on fire were the user.

Can you link even one story of an RTX4090 that actually caught on fire? The closest I can find is people calling the melting connector a fire. No actual fire damage to the PCB.

 

[purpleduggy]

no i am nvidias biggest fan. I have a lifesize poster of jensen in my house. my whole family only buys nvidia gpus. the rtx4090 power connector is the best connector ever. when it starts smoking we breathe in the holy nvidia smoke we've been blessed with by our lord jensen. its by design.

+5000 nvidia social points

 

[-Fran-]

For all I'd like to see better (modern or not) connectors for high powered applications in PCs, I do have to say there's a balance to strike in practicality and security. A bulkier connector will be secure for sure, but PC components are not big or massive to support such a connector. Plus, while I personally don't care about this too much, there's the price consideration. Those more secure, "better built" connectors do look expensive to use; not just cost of the connector, but actually build around it to support them and such.

I consider the current 24, 6 and 8 pins designs are sufficient for most powered things in a PC, and this new 12+4 pin with this new revision I hope doesn't have any more problems with wiggling and such that has been causing some users headaches.

Regards.

 

[yngndrw]

Datacenter stuff can run on 48V because they don't mind having to spend $300 more on fancier FETs, transformers and support components to make their 50k$ machines work more efficiently from a higher system power distribution voltage. For consumer stuff though, you need to stay within the realm of cheap-and-cheerful components and minimum parts count designs.

PoE is 48V, it's already commonplace.

768W from a 4 pin connector using the old tried-and-trusted 8A per pin sounds good to me.

ATX48VO is where it's at.

 

[InvalidError]

PoE is 48V, it's already commonplace.

Being "commonplace" for 15-30W devices where convenience outweighs cost and efficiency is one thing, bringing it to $100 motherboards is another. At 24V, you can still use buck regulator built with high-current 30V MOSFETs. At 48V, you have to either switch to far more expensive 100V GaN FETs (need the extra headroom so the FETs don't self-destruct from their own stupidly fast switching transients) or use a transformer-coupled DC-DC converter topology. Either way, the cost is going to be substantially higher.

 

[yngndrw]

Being "commonplace" for 15-30W devices where convenience outweighs cost and efficiency is one thing, bringing it to $100 motherboards is another. At 24V, you can still use buck regulator built with high-current 30V MOSFETs. At 48V, you have to either switch to far more expensive 100V GaN FETs (need the extra headroom so the FETs don't self-destruct from their own stupidly fast switching transients) or use a transformer-coupled DC-DC converter topology. Either way, the cost is going to be substantially higher.

If they can afford to throw around 14+ phase converters, I don't think cost is the limiting factor here. At the scale they operate at, they'll drive the manufacturing costs down if they need to - Which doesn't necessarily mean they won't use it as an excuse to charge more of course. In theory, the extra cost on the components should be saved on the PSU, although they won't pass that saving on.

 

[InvalidError]

If they can afford to throw around 14+ phase converters, I don't think cost is the limiting factor here. At the scale they operate at, they'll drive the manufacturing costs down if they need to - Which doesn't necessarily mean they won't use it as an excuse to charge more of course. In theory, the extra cost on the components should be saved on the PSU, although they won't pass that saving on.

Entry-level boards don't have 14-phases VRMs, they typically have 4-8. Switching everything to 80-100V GaN to efficiently handle 48V input at 150-200A out would add $20-30 to entry-level board cost, $50+ to higher-end boards boasting 500+A of peak overclocking juice. Motherboards and GPUs don't really mind having 60A @ 12V when they already have all of the copper layers necessary to deal with 200+A between the VRM and CPU/GPU Vcore pads which isn't really needed for anything else anywhere else on the board.

On the PSU side, you don't really save anything: the primary side remains exactly as-is and on the secondary side, most of the money you save by needing fewer parallel synchronous rectifiers gets lost in needing to use 100V GaN instead of conventional FETs, you trade high-capacity low-voltage (ultra-)low-ESR for about as many lower capacity, worse ESR, more expensive 60V ones, the transformer copper you save from downgrading the secondary-side wire gauge at reduced current gets spent on needing twice as many turns as you would on 24V or four times as many as 12V, etc. You are trading current for a proportional increase in voltage and the overall difficulty remains fundamentally unchanged, no miracle savings vs 12VO or 24VO besides how many wires/pins you need to carry a given amount of power from the PSU to loads or the size of those wires/pins.

For PC use, 24V is more cost-effective and convenient to work with at the point-of-load. 48V is the norm in datacenter and telecoms primarily because it was the highest nominal battery voltage (from the days where telecomms used on-bus lead-acid battery backup) that was still considered "low-voltage work" exempt from most regulations. Otherwise, they'd be running 400-450V (rectified Y240) all the way into the backplane and do two-steps conversion (board-level power distribution -> PoL regulators) to whatever they need from there instead of 48V that doesn't run anything natively besides stupidly powerful and noisy fans.

 

[doughillman]

no i am nvidias biggest fan. I have a lifesize poster of jensen in my house. my whole family only buys nvidia gpus. the rtx4090 power connector is the best connector ever. when it starts smoking we breathe in the holy nvidia smoke we've been blessed with by our lord jensen. its by design.

+5000 nvidia social points

When asked if you could point to a single incident of a 4090 actually catching fire, you could have just stopped after the first word of your response. The rest was just childish babbling.

 

[RandomWan]

99%+ of users aren't having issues, but it's a "much-needed revision." I think users just need to learn to cross their Ts and dot their Is.

 

[jonnyguru]

Wouldn't it be cool if we lived in a world where the press confirmed things with the industry?

 

[bit_user]

Wouldn't it be cool if we lived in a world where the press confirmed things with the industry?

Is this in reference to news of the upcoming connector revision, or regarding problems with 12VHPWR to date?

Since we're talking about confirming things, this thread claims you went to work for Corsair. Is that true and are you still there?

https://pcpartpicker.com/forums/topic/394446-what-happened-to-jonnyguru

P.S. Glad you dropped by! Please stay a while.