The 12VHPWR connector sucks
HTML-код
- Опубликовано: 23 окт 2022
- TL;DW: cablemod.com/12vhpwr/
My Patreon: / buildzoid
Teespring: teespring.com/stores/actually...
Bandcamp: machineforscreams.bandcamp.co...
Twitch: / buildzoid
Twitter: / buildzoid1
Facebook: / actuallyhardcoreovercl...
#nvidia #rtx4090 
Наука
EDIT: PCI SIG was aware of the physical weakness of the connector: t.bilibili.com/720822338533195796
I would like to thank Nvidia for providing me with material for low effort content.
Thanks for your informative video!!
just you wait for rdna3 😉...
Never feels low effort, it's good stuff.
Us: "This is irresponsible. NVidia should fix this"
Nvidia: "HAAAAAAAAAAHAHAHAHAHAHAHAHAHAHAA"
they don't care at all about users or complaints
they will start caring when the class action lawsuits are brought
HINT HINT
lmao...
if they didnt want you to bend the cable, they should've put that part of the cable (35mm) in hard casing
Keep in mind the 35 mm suggestion is NOT from Nvidia, but from Cablemod. We don't know how thorough their power connector testing was but they also don't indicate anything like this in their materials
@Entropium the 35mm comes from PCIESig which is the body that governs pcie devices.
You're right. Unfortunately then it wouldn't fit in any case. Nvidia wants to leave the end user with that problem. Truth is the connector should just be a 90 degree connector to go down the way most people install their GPU.
@@All_I_can_say_is_Wow I have a feeling we are going to see someone (like cable mod) make an adaptor or and actual cable that does the 90. Maybe for those of us that mount vertically we will see some solution for that.
@@Dragonsrage012 the 90 degree cablemod adapter releases in 6 days
This card has only been on the market for 2 weeks and it's already frying. Thanks Nvidia! ;)
450W fryer 🍟
Planned obsolesce gone wrong... All of these 10+ year old cards you can buy 2nd hand used reliable power connectors, it was about time to cut GPU lifespan to warranty period and increase demand for new ones.
On top of it, while Philips was not involved for this, they are still going to benefit since they are selling machines to TSMC and others.
Strictly speaking the connector has been around for 7 month and I believe I have seen 1 case where someones 3090ti melted
30 series had the capacitor issue. I am seeing a pattern here.
@@Vatharian capacitor?
If they really had to go this route... Ideally they would have adopted a more (high end) automotive style connector. It could even be nearly the same size, just *slightly* tighter tolerances between the crimp terminals and the plastic, as well as strain relief at least in the form of a semi elastic boot and preferably o rings around each pin, not necessarily for weather sealing but for its secondary effect of giving you more strain relief in the form of the pliability of the o rings (and this would all still allow for a bit of flex for alignment)
An Anderson plug would do, big pins and handles amps well. might be a bit overkill actually
@@draconightwalker4964 hey overkill usually means less fires/melting lol
PC hardware manufacturers that are trying to save every penny like using cheap TIM (paste and pads) would never use high-end connectors like that... They will instead blame consumers for any potential problems like always.
@@veduci22 sounds like they need to be sued again. That's the only way to get the message across
An 'automotive style connector' does not exist for >50 amps, except for Ford who is famous for having connectors that burn up and fail to the point that new components like alternators come with a new connector attached because you can't reliably use a push on/in connector at tens of amps. To use fewer wires and a smaller connector and increase reliability would require using fasteners, as in using small screws to attach the power connector. Interestingly, the EPS 8pin CPU power connector is claimed to be good for >350 watts, not sure why that connector wasn't used in the first place and the PCI-E simply never having existed. The ATX 3.0 standard is for douche bag OEMs like Dell and HP to save costs and cut corners. Why Nvidia chose to use such an obviously stupid connector is astounding considering how ugly that shit show of a dongle they include in the box that seeming is designed to make the situatation as bad as possible. Simply making the cable 12" to 15" long instead and putting the chip at the same end as the 8pin connectors would go a long way to reducing the problem. Moving to high tension bladed connectors would also improve things dramatically.
The new standard is so that OEMs can ship 4090s with 550watt power supplies and the GPU will simply throttle down so the computer won't crash and the sucker who was stpupied enough to buy a mass market computer won't know he was screwed. It's just anothe rway the PC industry can commit deceit.
Solving a non existent problem by creating a real one.
Your avatar is perfect, lol!
that's nvidia for you. even with ray tracing. "uber ray tracing" for slowingcyberpunk to single didgits just to show off frame generation on the 4090. tessellation bombing games back in the day to show off their cards but slowing performance badly universally.
Next they will try to sell us the solution
Typical Nvidia
Correct
Let's see how many of these cases occur before it becomes a proper fire hazard.
Yeah. I am honest here, i thought we would have seen Way more allready. So i guess it means it doesnt Suck that much...
Still hope i am wrong here and only really broken Connectors are an Issue (Which every other Connector would have an Isse as well).
@@flimermithrandir based on the reddit thread some users said if you google it there are multiple threads or websites have people saying the same thing. Its just a matter of time tbh..
Hopefully there will be better designed aftermarket cables.
@@randomdude8897 needs atx3 psu and perhaps a metal, or at least angled connector too
Oh its absolutely gonna happen. As someone that understand electrical things. Pushing that wattage through something like that connector is just insane to me.
It's almost as if the previus pcie design was already considering thermal limits and had a size acording to that, who would have thought that making wires smaller and more compact would be an issue while pulling the same or more power
Yeah its almost like the Standard PCIe connectors "just work" 😅
Hey there, I've got some decent experience with crimped cable loadings, and while most of your analysis is on the mark, the seam on top of the crimp connector is not likely the reason for the horizontal bending comment. any horizontal side loading will not be transmitted to that part of the crimp as the bottom is just solid.
I'd imagine the bigger issue with horizontal bends is that, they cause much more strain on the cable due to the larger difference in bend length between the inner and outer cable. This tension force on the cable causes a pulling force on the crimp, which can cause a variety of effects:
First of all, it can cause deformation of the crimp. See those flanges on the side of the barrel? Those are what holds the crimp in place in the plastic connector. If you pull on them hard enough, you might cause them to buckle, either 1: just pulling the crimp straight out of the connector or 2: causing the crimp to sit much further recessed in the connector than intended. 1 sounds worse but at least it's easily noticeable (you have a loose cable). 2 however is what will cause your thermal runaway most likely. The crimp barrel contains a small intended to push the pin against the barrel to ensure contact. In the one you show it's located on the bottom of the crimp near the front, so it's hard to see. However, if the pin is not fully inside the crimp barrel this contact can be much worse than intended, leading to high resistance in the connection ending in thermal runaway.
Alternatively, if the crimps are bad or too much force is applied the cable itself might (partially) slip out of the crimp causing a worse connection for the same effect.
Finally, even if nothing breaks, the crimps are generally loose in the connector, allowing them to orient themselves around the pin properly. Apply enough force though on the cable end, and they will be stuck in place, causing worse contact. Still in a properly rated cable bad pin contact should really not be such an issue. But just think of the power they're pulling through there. 600W @ 12 means a total of 50A being pulled through 6 pins. That is ~8.5 A per pin. For reference, normal ATX molex pins on 16AWG cable (www.molex.com/pdm_docs/ps/PS-45750-001.pdf#page=6) are rated up to 12.5A individually, but as soon as you put 12 of them in a connector that rating drops to 9A. This is because the limiting factor is the amount of heat generated by the connection resistance that can be dissipated per connection drops as soon as you start stuffing them together. Now I don't know what black magic Nvidia used for the new connector, but 8.5 per pin, on all pins (mind you a normal 8pin has 3x12V and 5xGND, so the ground pins take only 60% the current of the power pins) in the same space with just basic crimps just seems like asking for trouble to me by finding the absolute limits
This man knows his stuff.. spot on analysis..
Good info. BTW, it's 600W at 12v, so nominally 50A total or 8.3A per contact.
@@tinfever whoops I was definitely not awake enough when I wrote that
Fixed!
@@CensoredUsername_ so basically the cable is fine, just dont bend it? the picture of the pc on reddit shows some bending in the cable, right around that black material that im guessing is supposed to reinforce the cable, it looks like the user has hard bent that for cable management
feels bad man, dude bent the cable to take a picture of his 4090 with perfect cable mangement for reddit n boom lol 😂
Well, normal cabels uses 18AWG so the max current is lower 8.5A. Also it depends how many wires are used for power. You can have 12wires, but if you use only 6 of them for power the max current is then higher.
but buildzoid, if they made the 4090 pcb bigger, it wouldn't have the cool pacman mouth
@n n Personally I thought of the package manager for Arch Linux.
WAKKA WAKKA WAKKA, BITCHES!
@@nathangamble125 hmmm. can't tell if Pacman or Fozzie Bear...
Nvidia: Let's reduce the connector size by 4 while doubling the power. What could go wrong?
Nvidia didn’t create the ATX 3.0 connector. The ATX standard is maintained by Intel.
Idiots should really stop parroting dumb info, but they wouldn’t be idiots then.
Law of physics for resistance of electricity: Let me stop you right there
I wonder how many house fires started since the 4090 released.
Moores Law is dead, duh.
@@thierryfaquet7405 Going by ur comment, isnt Nvidia more DUMB to use the DUMB ATX 3.0 connector created by DUMB Intel 😂😂
NVidia: Doing dangerous things to save a few mm of PCB space. Also NVidia: Building a GPU the size of a freaking house.
Intel made this connector. Not Nvidia.
@@pokealong NVidia chose to use it, for this purpose, and with a bad implementation (see BZ's next video). All on 2 grand cards. Intel can share the blame a bit, maybe, but not most of it.
@@LeoDavidson Nice, RUclips keeps deleting my comments. Search "PCI SIG melt" and you will find articles discussing how PCI SIG found the issue is the CONNECTORS designed by INTEL that is the problem, and this identical metling is happening on the PSU ports as well. This issue is 1000% an Intel design issue, and Nvidia nor AMD has any control over it in any way. They are just using a standard created by somebody else.
@@pokealong When, in the entire history of mankind, has the existence of one standard prevented another standard from being introduced?
Its like squeezing 3 cameras on a Smartphone but apparently there is no room for a headphone jack.
Good thing Sony still makes Flagship phones with headphone jack 4 & micro sd slot.
I decided to figure out if there is a large difference between the new 12-pin (Amphenol Minitek Pwr 3.0) and the old 8-pin connectors (Molex Mini-fit Jr). I found out that they are basically the same size and the plate thickness is 0.2mm for both. The maximum contact resistance (the important part) is also the same 10mOhm according to the datasheets. The material of the 8-pin (either brass, nickel or phosphor bronze) has more internal resistance than the copper alloy of the 12-pin but also more area so the difference between the two is small. The connectors for the 8-pin are rated at 13A each where those of the 12-pin are rated at 12A. The heat produced by the contact resistance is the interesting part. The 12-pin at 600W has a current of 8.33A per contact, resulting in 0.7W of heat per contact or 8.33W of heat in total. An 8-pin at 150W has a current of 4.17A per contact, resulting in 0.17W of heat per contact, or 1.04W in total. Funny thing is, a single 8-pin connector is actually larger than Nvidia's 12-pin and thus the 12-pin has to handle 8x the heat in a smaller size.
So someone had to approve putting on a connector that outputs twice the heat of 4 8pins put in less than a quarter of the space. Genius.
Power connector cooling fins/fans when?
@@mendez256 NOW
@@mendez256 Lmao now we need heatsink for cable? When will we water cool it lmao
I already do lol
PCI-SIG put out a report about dangerous thermal variance with these connectors just a few weeks ago. The cable in question probably predates that warning and the changes we should see in cable design moving forward. This isn't the first time I've seen connection issues with the folded terminal molex micro-fit size connectors, but it is the first time I've seen them melt.
Ultimately, these cables have installation requirements that are too strict to expect consumers to use them properly on a wide scale so I suspect the design will be short lived.
And these reports were regarding the connections at the PSU, everyone ignores that. They were melting AT THE POWER SUPPLY, yet everyone is blaming Nvidia for this new INTEL standard that is failing because INTEL did not thoroughly test it.
@@pokealong LMFAO AT YOUR STUPIDITY!!!
Its bugging me from the first day ive seen that connector. no one took it seriously.
I'm glad i wasn't wrong about it. Disappointed that others didn't even bother to talk about it.
Wwooowww
A few people have brought it up actually
Didn't GN and J2C already talked about this issue?
@@First-Name_Last-Name Yeah I remember Jay specifically saying he wished each of the terminals was at least the same size as an 8-pin terminal. Sure, the 12/16 pin connector would therefore have been bigger, but also better.
@@shadowprice8100 a few. It's a power connector . A potential fire hazard that a few ppl didn't even mention it. They said "its just weird and what its purpose with the new sensor bs is" no one said that is smaller and weaker than probably just only two regular 8pins
It's just infuriating that ppl agree to what nvidia and who ever comes up and say whatever to impress the crowd and grab money without any results or proofs. They selling things to you for testing!
Reminds me of cheap molex splitters that were broken out of the factory. I've had to bend them back into shape before. Good thing those don't pull 600 Watts, LOL 😜
MOLEX to SATA, wipes your DATA.
They really made a cable that shouldn't be bent for areas with often tight cable management
I really enjoy how they expect a cable to just stay bent at a certain distance away from the connector like its a paper clip and will hold that bend exactly where you bend it.
if the wires were solid and not stranded, it definately would... but that would also mean they would need to be the proper gauge as well.
not only that, with the size of the 4090 theres no way a cable with 35mm of straight wire is gonna fit half of the cases in the market.
@@PhantomBlank Half? Only half??
Keep in mind that is a third party analysis. I'm sure safety was a reason of that measure.
@@Splarkszter Come-on man, that's almost as bad as the "they have engineers" argument I hear about fanboys for things like smartphones. I see it way too often, it's like people think engineers can't overlook something. Didn't stop iphones from coming out of the box bent, didn't stop Samsung phones from exploding (or them apparently expanding only after a few months sitting unused), or the PS3s over heating and failing, or 360s heating up and bending the literal board, etc. Engineers fail sometimes.
Maybe they just should make the connector housing longer, like 35 mm, to prevent this. But then some users will have trouble with the space it will require...
better than a fire hazard
@@dioneto6855 what are you talking about? Can't you see all that space being taken? It's gonna eat like, what, 1 cm bigger?/s
@@wisdoom9153 that can between a case closing or not.
@@soi8739 it's likely heat shrink tubes, it acts as a strain relief.
Which will add enough leverage to make it easier to rip the connector off the board.
Think JTC was saying this was gonna be a problem with cases and being too bent , but they blew him off
Pretty much anyone with a basic understanding of the subject complained about it and was shut down by people who thought they knew better somehow.
@@Waldherz do you think it will be the same when cablemod and others 3.0 psu come out with their wires sets . Or a 90 adapter you have to connect to gpu to keep away from side panel. (Sorry if dumb ? I’m a noob but very interested in this content, wish I got into this 10 yr ago haha)
@@farrez_gump With a 90 degree adapter you wouldn't really need to bend the cable that much, so should be fine. Custom cables or direct 16 pin to 16 pin cables still use the same connector, so it wouldn't improve the bending issue. Although not needing to fit that gigantic 1x 16 to 4x 8 adapter might make running the cable without major bends somewhat easier.
@@farrez_gump Good question. Most of these 3rd party cables dont have this problem, so you shouldnt need a 90° adapter.
But only time will tell.
Definitely looks like 4000 series is the generation to skip :D let the rich kids do the beta testing, they can afford a new PC if current one burns down.
the power rating is mostly given by the standard gauge of wires the connector is meant to be used with. PCIe 6 and 8pin connectors, as well as pretty much all other standard connectors prior to this new 12vhpwr one were either 20 or 22awg gauge. Having no guarantee the 20awg gauge would be used, manufacturers had to design everything around the 5A current limit of 22awg (when no active cooling is applied). So they probably rated everything around at that same limit. Now the new connector requires 18awg gauge, which can handle 16A or 14A passively I think? Point is, now that the wire itself is no longer a limiting factor, the job goes to the connector. The other thing also is that 18awg is much thicker and less flexible, unless you use a ton of strands count cable, but then it gets even bigger and it might not crimp as cleanly.
The crimps themselves (terminals added to the end of each wire) look like standard microfit crimps from Molex. They come in a lot of different flavors such as tinned, gold plated, lower or higher gauge, and sometimes the current rating can vary from each of those. For example the tinned crimps will usually be made of much thicker sheet metal than the gold plated ones. I don't know if the new 12vhpwr standard mentions a specific flavor of crimps, but it's another variable to keep in mind.
From the other document from Zotac saying the connector can only handle 30 cycles and with all this stuff, the only conclusion to have is all these components are meant for a single time assembly with minimal cable flex. This is probably fine for system integrators, but the culture around pc parts is "you can disassemble it all anytime", so it's really bad to give those parts to customers without proper training or knowledge of the parts' limitations. I think the main culprit is simply that this is all a new standard and the industry wasn't ready to deploy it properly, so the cables aren't made as well as they are supposed to be and they have not spent the time to ensure usability will be the same as it was with previous cables.
Time will fix these issues and we will see lots of company selling "premium" 12vhpwr cables while it will be exactly what they were supposed to ship from the start.
You should know the 30 cycles specification is a conservative number given as a guarantee for all cables, and as a legal cover for themselves. In reality the cable will likely be good for a lot more than that number.
@@Safetytrousers sure, the point is more that unlike for example a USB connector, it's not engineered to be connected and disconnected often, it's meant for a one time assembly or extremely infrequent disconnected/reconnect cycles. It's important to understand the engineers' intentions before using a product for something it wasn't designed to be used for. (not gonna talk about what it should have been designed for tho)
@@Safetytrousers *less, as experimentally demonstrated by NVidia
@@erkinalp The reported failure wasn't due to getting plugged in multiple times, so not a hidden 10 use limit.
@@EMILE12345678901 Key is that the engineers building this card and choosing the connector didn't consider the standard requirements for PC components. Instead they wrote warning pamphlets explaining how fragile their design was.
AMP/TE creates connectors for everything and write out descriptions of how to use them safely in products. 30 use connectors with cable bend limits are clearly not suited for user juggled PC components.
suggestion for another video: buy some cables and do a live dissection on them. also measure the resistance of the pin to receptacle connection on both the new 12VHPWR and the old 8-pin PCI-E connectors. Then you can calculate based on the current what the voltage drop over them is and the heat generation over that connector is. maybe even see what happens when you bend the connection a bit causing the less than optimal connection and measure the resistance on that too.
Was there also limit how many times you can insert that connector before it's out of specs
@@LimbaZero The limit is 30 cycles, but according to Nvidia the 8pin connectors have the same 30 cycle limit.
@@Gamer-df9xt the Mini-Fit spec is public: www.content.molex.com/dxdam/25/25cce481-dbb4-467e-ad4c-1deb2f5122cc/987651-8272.pdf
Looks like it's 70 or 100 cycles depending on the surface finish.
you would need a REALLY good ohmmeter for that
Best to just send a ton of current trough the connectors and measure the drop
@@alouisschafer7212 or literal milliohm meter and that doesn't sound cheap. Although it will be cool to see
I would love if we could get back to the actual pci-e spec handling of connectors at the end of the card and not on the side. it would mean cleaner handling of hiding those cables.
yes please!
They should be allowed to choose 1 or the other, but you have to have one obviously, Nvidia tried to optimize both away and invent something new and it was overly prioritized towards aesthetics.
I was thinking of this last night when I was looking at Xilinx FPGAs, why don’t AMD and Nvidia do it when Xilinx does it already?
@@Fractal_32 Are you referring to some FPGA based acceleration boards for specialized computations? Because I weren't aware those didn't just use existing interface standards to fit in existing computers.
@@johndododoe1411 I’m referring to PCIE FPGA cards from Xilinx that put a 6 pin on the opposite side of the card from the rear panel IO, typically closest to the motherboard. Boards like the “Xilinx A7 Artix-7 200T XC7A200T” (Development board)
For anyone not having seen this already, GN reported back in September that Nvidia itself had found some potential overheating with that connector, which was likely tied to bending:
ruclips.net/video/p48T1Mo9D3Q/видео.html
And in a later video interview they stated that all of the issues with the adapter cable have been fixed and that they're "very confident in it".
ruclips.net/video/MxU1AkGNnT0/видео.html
I think you nailed it right on. High resistance loose connections are the issue. I'm no electronics engineer but I've been an electrician for many years and we see this all the time. I don't understand why manufacturers are thinking some test in lab for 24hr period means it's good
Electronics engineer here. There is also going to be a possible issue over longer period of time in case there is a higher contact resistance on one or few of the pins, but initially it works fine, but as the card gets more power cycling and the connector pins experience high and low currents at times times, the thermal expansion and contraction might degrade the contact resistance even more.
Like in my 3d-printer where originally all connectors were fine at the start, but after hundreds of prints and the connectors cycling between high and low current, the heated bed's connector finally melted it's contact housing and the connection broke, so on one of the pins the contact resistance had slowly degraded over time and increased the resistance, until finally it was so high the the contact heated up so much that it melted the plastic housing.
One of the causes also seem to be the very stiff wires that they use, which causes more mechanical stress to the connector, especially if it seems to be on any kind of sideways direction. I wonder if for example, silicone wires (basically wet noodle stiffness) would allow for tighter turns, because they aren't so stiff...
@@antecboy The stiffness is the copper itself, not the plastic insulation. These are multiple 0.8mm² wires in a cable that parallels them to carry 63A like the heaviest common industrial cables with IEC 309 plugs. Those big IEC 309 plugs aren't even rated for use below 20V (purple color code) and wouldn't physically fit.
@@johndododoe1411 0.8mm²(~18AWG) wire is still quite small and silicone wire at that size would _very_ flexible. I can only assume, having not opened these harnesses that the strand count for the copper is just average, not special in either direction. I'm guessing the insulation is PVC, which isn't the most flexible material either, but it is cheap. But if the problem comes from the copper being too stiff, then a higher strand count wire would make it less stiff for a given mm² area.
I think it would be interesting to build a custom wire harness with silicone wire to test and see if that does help with the problem.
@@antecboy Don't forget the combined stiffness of all those parallel wires. Hence my suggestion to use a single braided copper strap for each rail and a connector that is inherently good for 67A. The one I suggested can be trivially integrated in the back of the card with an airflow optimized cable attachment.
@@johndododoe1411 You are right The 3090 ti has the exact same connector but a 3 cable adapter instead of 4. The 4 cable adapter is way too huge and stiff so its prone to damaging the connector. The PCI 5.0 cable should solve the problem or using a smaller 3 cable adapter.
Nvidia should be forced to recall all of these 4090 gpu's due to a fire hazard....
Or ship out the cable with the fixed connector to anyone who purchased one for free
@@SamLoki Or designed a right angle connector to begin with.
AMD will also be using this new ATX PSU standard. In case you were not paying attention.
@@henryvaneyk3769 AMD will probably put more thought about it though.
@@henryvaneyk3769 We don't know if they'll be using the 12VHPWR connector yet.
Yeah when I saw this design I figured they'd done a bunch of testing and the pins were huge and tolerances were tight so moving the wires wouldn't effect the connectors.
Welp I was wrong on all accounts. That's insane to rate this at 600w. Good luck 40 series owners.
I love your wrap up statement in this video - quoting Buildzoid verbatim (spoiler alert): "You can always count on NVIDIA to come up with a complicated solution to something that isn't really a problem that is actually worse for reliability."- Gold statement sir - great work !!!!!
That quip was the cherry on top of the sundae. :-)
So I just looked up both the 12VHPWR connector and the 8 pin ATX power connector datasheets (the 12VHPWR model is seen on the video and the 8 pin power connector I found on Digikey with MPN: 2354120-1). Each pin for the 12VHPWR is rated for up to 9.5A while each pin of the 8 pin power connector is rated for up to 12.5A. The 12VHPWR has 6 energised sets of PWR+GND pins (12 pins total) which would yield 57A of max current and at 12V this is equivalent to 684W of power (although the manufacturer only specifies up to 600W of power for whatever reason). The specified contact resistance of both of these connectors is the same, 5-6 mOhms and the receptacle terminal style is the same in both. I couldn't find any specifications for the housing of the 12VHPWR, but the rated dielectric withstand voltage in the 12VHPWR is 1000V whereas in the 8 pin power it's 600V. Given that the pin spacing is definitely not bigger in the 12VHPWR, that would imply that there is a thicker dielectric there, not thinner. So, on paper at least, the connector doesn't look stupid at all. Whatever the problem is with the melting connectors, it doesn't look like it's in the specifications.
I remember, the Olivetti M24's gfx card had 2 heavy cables go to it because the monitor connector also powers the CRT.
yup, it was something like 21V or so. Screwed and bolted on.
Olivetti - wow, blast from the past there!
@@richmanricho The first thing I thought of when reading Olivetti was a typewriter.
@@wargamingrefugee9065 They made one of the first precursors to a 'personal computer' a regular person could buy. The Programma 101 in 1965.
Also, Jonney Guru was saying the 150 watt "rating" of a single 8-pin PCIe connection isn't an actual power load rating. It was more of a classification rating that NVIDIA came up with for their own cards. As he stated in a Q&A video with eTeknix, smaller terminals with less 12v leads can't handle more current than a larger terminal with more 12v leads.
Mr Guru has not been relevant in more than 5 years..
@@godslayer1415 Based on what exactly? If if that were the case, that little tidbit of information is over five years old anyways. So, even if you were correct... which you're not, that information is from when he was still "relevant" and thus, still true.
sure but the end result of the entire industry treating the 8pin as a 150W connector is that the 8pin has a massive safety margin. The new 12VHPWR has pretty much no margin for sub optimal operating conditions.
@@TheGameBench Maybe he's referring to the fact that you can't rely on an employee of a power supply manufacturer for impartiality, nor to say candid things about the products designed by their corporate partners such as those Buildzoid said in this video.
@@ActuallyHardcoreOverclocking Absolutely. Just saying that, all other things being equal, if the 12-pin can handle 600 watts, so can two 8-pins. And if two pins can't really support 600 watts, neither can the 12-pin. I agree with what you said in the video completely, the 12VHPWR connector is a terrible design that had no reason to exist outside of it being small enough to fit on their reference boards. Happy to hear that AMD isn't going to be using it, hopefully partners as well.
Then NVIDIA needs to provide a 35mm rigid support coming out of the board if that is a requirement. They need to guarantee requirements are met.
My attempt at a simple explanation on why horizontal bending is so bad: First, imagine the power cable is a race track, then realize that all of the wires in these cables are the same length, but if the cable bends side to side (horizontally) the cables on the outer lanes will have to travel a longer distance than the ones in the inner lane. What's happening is either the wire is getting pulled backwards out of the 10132447 receptacle terminal if the crimp isn't secure enough or if the crimp is secure enough then the whole 10132447 receptacle is getting pulled backwards from the plastic housing. Regardless of which of these are happening, both would result in a smaller contact surface transferring the same amount of power through it. This is bad because you need all of that contact surface in order to 'spread the load' keeping the thermals in check. If the load of energy is un-spread then it's concentrating into a small spot which results in melted plastics.
I honestly expected them to use the extra pins to automatically measure the resistance of the connector. That's literally the first useful application I could think of for direct communication between the power supply and graphics card.
yeah if the GPU sees "oh shit we are getting 12.5V on Sense but only 11V on the +12V plane we need to shut down the connector has bad contact" it would make sense to have the sense pins in the first place. Not using them for that is kinda dumb.
Perfect idea. Unfortunately that would require new PSUs. "ATX3.0" PSUs are the same as older PSUs, just with the new connector slapped on the housing.
@@jonnyGURU It would be a feature of the cable, not the PSU.
@@johndododoe1411 If you have a wire that measures resistance or voltage, what's it going to report to? If you look at the sense wires on a 24-pin, those double wires report voltage back to the PSU. What good is a cable that measures resistance if it has nowhere to report that data to in order to take action?
@@jonnyGURU couldn't it be made so that it talks to the GPU VRM controller and shutdown/power limit the card?
It seems like the only way this exists is to save one square inch on PCB. Although those PCBs are so small I struggle to understand why was it an issue.
even 2 8pins would be much much better than this nvidia money grab and don't look back BS
Worst part this is 12 pins. Even if they keep the old gauge it would be equivalent to 2×6 pins.
This is beyond stupid.
@Yeltnerb 1 Let's save 25% of space on the PCB while adding 100% more height to the cooler. We are obviously starving for space out here.
It supposed to be also replace EPS cable on server. Yes server GPU used CPU power cable if you didn’t know.
The RTX40 series is made for the sole purpose of maximizing profit. Terrible quality, terrible efficiency, terrible thermals; I wouldn't even pay $300 for this pile of shit.
I love all the tape and heat shrink making the cable more rigid to more efficiently transfer all of the force to the connector.
LMAO
I have a very roomy full tower case. If I mount a 4090 in standard orientation; There isn't room to bend a cable at 35mm out from the gpu without taking off the side panel. You can start the bend at 35mm but unless you bend the cable at a hard 90° the cable is still going to make contact with the side panel which means the curve of the cable pressing on the side panel will still put an undesirable amount of stress on the connector. I Can't think of any REASONABLE cases that have enough room to achieve that 35mm of cable length before the bend. It doesn't matter if you cable manage up, down or to the side of the gpu; any bend you put on it will stress the connector. The connector should have just come as a 90° connector and it should have been built better
i was looking forward to a replacement of the ancient 6+2 pcie connectors, but i was expecting something better than this crap. more alike the 12v 8pin EPS connector but with 12 or 16 Pins in a 50/50 12v/gnd split without some stupid sense lines.
well... think sense lines was to try and make sure gpu's didn't blow up the psu's but well... we are at a time where most psu's are reliable (except from gigabyte....)
Sense lines make sense as a safety feature, IDK what they are actually used for, but they could for example be used to supply a reference voltage signal for the GPU to figure out if the voltage drop over the connector is too large, implying a bad connector/connection.
@@JohnDoe-jp4em thats not what the sense lines of the new connector do. basically those sense lines are connected to ground to show the current capability of the power supply, because the new connector has different ratings for different power supplies - theres a 450w rating and a 600w rating, maybe even a lower one. i dont know exactly how its configured, but essentially of all sense lines are grounded it means full spec, half connected means 450w and so on.
Its not a smart connector, its a 6+2 with extra steps.
So here comes the kicker: if every power supply, that offers the new connector, had the proper full power rating, there wouldnt be any sense lines necessary because they wouldnt do anything, but here we are, making basically 2 or 3 "Versions" of that connector right from the getgo, so we can retrofit old and weak power supplies.
instead of 3 or 4 pcie 8 pin we could just use 2 EPS 8 pin - EPS has a much higher specced power rating than pcie 6+2 because it has one more 12v line and is a more recent standard.
@@sgredsch the 12vhpwr has a spec for 150, 300, 450, and 600. It's intended at least from what we are told to completely replace pcie power connectors, even on lower power gpus and the sense pins are not intended for older psus but lower capacity ones, a 450w psu has some reasonable qualms about delivering 600w through one connection. Instead of running different amounts of pcie cables, this is just different amounts of sense pins.
@@Tcatracing older psu are being adapted via pcie 6+2, just like nvidia has been doing it for two gpu generations now.
Because nvidia decided they need mini-ITX PCB size on their flagship cards with some weird triangular cut out for blow thru fan. Use normal size PCB and you can fit fucking ten 8pin PCIE connectors.
This bigger size PCB will cost Nvidia 10$ more per rtx4090. They cant afford to THAT 😂
mini itx my ass, it's a lot taller than the bracket making it hard to fit in most SFF cases, even with a waterblock (about that: why is no one making waterblocks with fitttings at the end of thecrad instead of the top?)
The joke is, the tech media outlets are saying that one factor is wrong connector usage by the user. As you stated, if the 4090 would be cramped into a case that's way too small and the connector were to be bent on the sidepanel, I'd say yes, that is wrong usage by the end user. But this piece of garbage being pushed on the customer is a joke. How can you make a fragile connector and not instantly design a 90° full-plastic adapter, while knowing that your connector is garbage.
A few points: A) As someone who worked with Data Center servers and cabling, I agree with the graphic showing the bend starting 35mm away from the cable. This is almost exactly what we would tell new people during training (1 inch, instead of 35mm) because it is advice that, if followed, gets precisely what you want. Yes, there is no such thing as the middle of the cable bending on its own, but that isn't really the point. The real point is leverage. Even a small amount of force pulling on a cable to make a neat turn that looks good can end up putting a lot of sideways leverage on the connector. Telling someone to make the cable come out straight from the connector doesn't sound like it makes sense, but it is advice that will produce the right result instinctively if followed (because it means pushing the cable to provide slack, instead of pulling the cable tighter).
B) At 12:10 - I feel like this could use clarification. I don't believe the sense pins are to prevent over-current to the power supply, but to prevent over-current down a single cable. Since Nvidia knows that people will be using adapters, it is entirely possible that someone could plug a 600w GPU into a 1000w+ PSU, but not plug in all of the 8-pin connectors. Instead of changing their power design, they engineered this to avoid pulling 600w down a 150w rated cable. I agree with your characterization heartily - this is an over-engineered solution for a problem that they created.
C) 16:02 - missed opportunity to say, "Truly, this will go down as one of the designs in history."
I wonder how many manufacturers will do returns avoidance using the cable mods wiring page when a customer fills out a RMA.
"Warranty Voided for misuse of the 12VHPWR connector". 🙄
just showed your video to our company electrical engineer and he nearly had a heart attsck seeing this connector
No, you did not
As an aspiring electrical engineer I reacted the same to that bullshit connector.
Nothing beats sheer size and contact surface area when you have high currents. You cant cheat physics... electrons gonna electron and that means give them beefy connectors or they will burn your stuff to a crisp.
600W at 12V is 50 Amps... for 50 Amps you would use big beefy bolt on lug terminals or screw terminals in the real world NOT a tiny GPU power plug.
Two 8 Pins would be perfectly fine at 600W you have 8 current carrying pins (Sense Pins are useless imo) so each pin would have a 6 amp load on them that's acceptable.
Three 8 pins would be better though.
@@BungeeGum1 not of the top of my head 😅
The XT60 is so good because it uses round barrel jack type connectors that are springloaded with a fair amount of force and they are usually gold plated. Spring force is a huge factor too. If you press two surfaces together they make better contact.
Surface area for one XT60 is probably the same or even less maybe than one of those stupid Nvidia Plugs.
Well regardless the XT60 is very capable there are very good reasons why beefy LiPo packs use that exact connector. Its rated for 30a continuous and 60a peak that says a lot...
Two little XT60s can do what three big bulky 8 pins can do in theory.
Why isnt that connector introduced to PC Hardware? The more I think about it the more I realize that those connectors we trust with so much current in our computers are kinda shit.
Like it would genuinely make sense to desolder the PCIe power connectors from a GPU and put XT60s in their place and to the same with the PSU Side.
@@BungeeGum1 For real! They should use that connector on the high end enthusiast cards and make corresponding PSUs.
EVGA could have done that but well they got too fed up with Nvidia.
The PCIe Connector works but its bulky and a very crude solution compared to something like the XT60.
@@alouisschafer7212 for reference cranking the starter of a car tends to be 120-130A, and it lasts few seconds... It has massive wires, terminals and they are not friction fit. This thing is a tiny friction fit connector that's bound to wear off.
Extending the cable straight out is going to be worse because of the simple fact of torque multiplying force, the more you extend out from the pivot point, plus the fact that the retaining clip being at the bottom rather than on top to anchor the connector from such cantilever force. Think of a breaker bar, the longer the bar, the more you multiply the torque you apply.
Any electrician would see than it could become a problem, we use different wire gauge for a reason also plugs or fuse.
Even using everything the way it's supposed too, any wire not secured properly will cause problem. I've seen melted plastic on the fuse with wire just handling lighting and not even drawing 400W.
Has anyone dismantled this connector to have a look at the build quality? I would imagine that poor quality control could impact its reliability.
You’d think that if given the opportunity to create your own, custom, high power connector, they’d have designed it to be able to work for the application it’s designed for. Well done, Nvidia!
@@_Clivey shhh don’t speak facts and logic in the comments section. When AMD uses the cable it will be ok
@@_Clivey See, this is what's confusing. Nvidia claim to have designed it, some people say keep saying they designed it, but it's also a standard.
Either way, it's obvious this connector is crap for high power PC applications .
@@_Clivey Nvidia designed its 12-pin precursor (which they used on the RTX 3090) though.
The must be hiring their management from Boeing rejects, cause they're certainly crashing and burning just the same
@@dennisp8520 nvidiots seething their 600w card is on fire
In the automotive sector, a 12V connector rated to carry 50 amps would likely have massive terminals that would be firmly secured in the connector with two separate terminal position locks. Not to mention the first 4 - 5 inches of the connector's wiring would be taped up and zip-tied securely. Alternatively, it'd be a fat ring terminal secured with a nut. Asking a connector that small and chintzy to carry that kind of power leaves no margin for error.
Problem with the 35mm bend protection is that it's stupid, too. It just ends up providing more leverage on the connection. It'd be like using a longer handled wrench on a stuck bolt. In this analogy, we want the bolt to be stuck and having any kind of extra pressure on moving it is going to be bad. A shorter handled wrench (keeping the analogy) would also just be more likely to break the handle instead of the bolt. Also super bad. Instead, why put everyone on one single bolt when everyone has been doing it with 3 or 4 bolts for over a decade already? Oh so you dont have room for it? Look at the size of those GPUs, the lack of space is a lie. It's down to simply trying to be A) useless materials B) use proprietary materials to try to force consumers to be locked into their products in any tiny way they can implement.
I do like that one cable is 600 W. It will take up less space once new PSUs have it built-in. However, they should have come up with a more robust connector with thicker pins.
How about a XT60 connector? It's rated for 60A which results in 720W at 12V and is not only a single cable, it also has just two leads, so it's even easier to route. Also, this connector has profen its worth in many RC devices and is about as big as a 6Pin PCIe connector.
@@Donnerwamp That is almost good enough. It would be better if there was also a small single twisted pair for communication to the PSU, as one important factor is that the PSU may supply less than 720W. They have a similar problem with EV battery charging where they have smaller pins for communication separate from the big pins used for power. I agree generally that these connectors being used today are just ridiculous. I think we are due for an upgrade to something that securely snaps on and doesn't bend at the connection. This problem has been solved in so many other places, so why do we still have these antiquated connectors?
@@geordonworley5618 Hm, maybe a "XT60B" with two sense pins like the bad plug has?
Also, yes, there are so many connectors that avoid so many problems this new thing has, it's kinda sad that we have to deal with that.
This makes me wonder why 24V systems aren't made a thing, I know it's a bit more efficient to step down 12V to Vcore voltage, but it would enable you to use way less material and space for wire and connectors. Doubling the voltage and thus halving the current would in theory mean 75% less wire gauge and connector size. And I'm not even sure if running it at 12V isn't just shifting the efficiency loss to the PSU.
We have a legacy of 12V systems in PCs so a lot of accessories wouldn't work.
Intel wanted to move to 24V or 48V but that never got off the ground.
PSU gets to use a transformer, which is required anyway for safety isolation, so it can do the large-ratio stepdown more efficiently.
They could make like a 48V 4 pin connector for graphics cards.
@@groundzero_-lm4md The funny thing is that power-electronics have become small enough, where you could basically come up with an adapter system that interfaces from 24V to 12V that has the power-electronics built into the plug, at least for the stuff that consumes less than 20W or so for auxiliary/low power devices and such. This would probably only be HDDs/SSDs since the rest is usually powered from the main board which could be constructed to accept both voltages even. The rest could be produced with the new voltage standart like it's done for switching to new CPU sockets.
I would go even further, like Automotive was moving with mild-hybrids. They started adopting 42 V. and remember, automotive is also using more and more computing stuff ( teslas having consoles from AMD..., most high-tier multimedia / navigations are powered by NViddia, so... )
I build mining equipment wiring harnesses and even I know you don't ever want to do anything lime this. Just use the correct 8 pins.
I have checked on similar pin assembly and yes similar only and the amount of force needed to split appart the casing was so much that the plastic housing cracked first. So while its only similar hardware I have a tendencie to think something else is happening inside this connector. I can provide image or video material if needed. Awesome work as always 👍😀
Yes, that connector seems insane... Why didn't they go larger for each pin if they're running more power through each?
To save $0.00001 per card.
To make it look "More Advanced" when they showed it off to the tech illiterate execs and investors
I don't understand why they would adopt such a dumb standard. Is this supposed to trick a consumer into thinking it doesn't use much power, because one 12pin is less than four 8pins? I hope AMD learn from this.
Rumors stated that AMD will get rid of this connector and stay on 8 PIN pci-e (probably 3 connectors).
I have checked the connector design of the plug itself and saw that you are 100% correct with your assumption. Just look at it. This IS just folded metal. So the connector itself is not really 100% at fault but (to be very specific) rather the way those terminals inside the plug are made. If this notch/fold/groove/overlap got welded shut I believe there would not be such a problem like this. It would still be flawed but at least somewhat safer to use when bent.
Also, you can't even go in a Corsair 1000D with the cablemod recommendation of 35mm after the bend because the GPUs are just so massive.
Here are a few options. The cables are soldered with tin. The problem may be in soldering or broken solder joints. A loose pin can also get hot. Oxidation or dirty pin. Perhaps, only two connectors have been used and the other two have been left unconnected.
I get the idea behind remote current and voltage sensing, but it's not really important here. At this point, we might as well just start using a pair of fat screw terminals or bullet connectors and 6-gauge cable. Surely there's an existing industrial solution to the problem of *transporting 50A of 12V a distance
That sounds remarkably practical.
@@wargamingrefugee9065 Right? The 90A relay I replaced at work? Screw terminals. The portable spot-welder tool? 8mm bullet plugs. The CPU+VRM module in a Power Mac G5? Surprisingly, screw terminals. This is so far from a new idea, but the industry is obsessed with shit-tier Molex connectors.
@@asm_nop Hello, igorsLAB posted this video, "Das Grauen hat ein Gesicht: NVIDIAs feuriger 12+4 Pin Adapter für die GeForce RTX 4090 analysiert!" (Yes, it's a German language channel, but he posts an English translation of his written articles too. Sadly, the translations are a bit rough [Google Translate and a fast edit?]. However, they're still worth a read.) If you haven't already seen it, you won't believe what's inside of the nVidia supplied 12VHPWR connectors. The wires are soldered onto tiny little pads made from stunningly thin metal. You hit the nail on the head when you wrote "shit-tier connectors". Have a good one. EDIT: The link to the article is in the video's description.
NGL I was quite shocked when I was installing my 4090FE. The nvidia adapter is super loose and you can very easily pull it out because the latch to hold in place is very small and have almost no grip on the female plug on the card. The special 12VHPWR cable from Corsair has the same problem BTW.
As someone who has spent 17 years working in public transportation maintenance, this looks all too familiar. Commercial aftermarket systems (HVAC, etc) often use high power molex style connectors. Even in a space where there isn't any force put on the connection they're still the first place to look for failure. Heat, vibration, etc always seem to burn these up before their rated capacity. Weak contact, high resistance, more heat, see ya later.
They need to make the connection below the housing more robust. The pins need to freely move, but at the same time, they need to have something to reduce the tension on the area where the pins connect. It really comes done to making the connection from the cables to the housing to small. It's classic Apple design.
This conector needed to be 400w max, if you want 600w then a extra one or even a 8 pin extra one.
This connector needed not to exist. These 400W+ cards are pure idiocy.
@@Leopardipzg I agree
@@Leopardipzg 400W+ cards do have a purpose though.
They're a means for people with more money than sense to fund the development of more powerful next-gen midrange GPUs which match the performance while using a sane amount of power and not costing over $1000.
i want to see a case where this 35mm bend is possibbe while normally mounting your gfx card
The heating up of the plastic and metal itself can cause leveraging to pry it away. So even if you install it corectly, there is still the possibility of causing this problem through normal usage.
The thing about the bending if you read that statement carefully appears less concerned with the bending of the receptacle around the pin or a consistent load bending the plug in the socket.
Instead, the issue appears to be more what happens to a multi-wire cable where the wires are held fixed relative to each other, and what happens if you then bend that bunch of cables near the plug end. The illustrations show that the wires on the outside of the bend are then stretched due to the difference in radius between the inside and outside of the bend. This may weaken and damage the crimp where the wire connects to the pin receptacle. Thus if one of those power wires gets disconnected by pulling out, it will either cause a very poor connection and potentially heat, or the gpu may try to pull that lost power through the remaining good wires, hence the warning about increased load through the wires. (Not sure about that last part, I'm a mechanical rather than electrical engineer).
But if you bend the wires further away from the plug end, firstly there may not be held as right relative to each other, so you don't get the same level of pull. This tension is also spread out more across the wire and less into the crimp connector at the plug receptacle.
Hopefully the requirement for 600w is the *only* reason this connector exists and the AIB 4080 and lower cards might have standard 8 pin connectors
in my case the connector is, exactly like the example you gave, badly smashed up against the side panel bending the cable up and diagonally and yeah.. so far no problem even while running benchmarks on a mild overclock with the 4090, but im not risking a sustained full load. I really need one of these right angle connectors (apparently not available yet)
I also felt very anxious bending the adaptor but I prefer to use side panel of my case. Nvidia forgot ppl dont use test benches in real life.
@@simptrix007 don't buy expensive cheaply designed crap from Nvidia next time.
Lesson learnt the hard way.
Thanks *god* these 90º adaptors are becoming the norm. This was an issue before the release and still is after. Unfortunate this happens to people due to misinformation. At the same time though, I understand the complexity of having 4 8 pins connectors on a PCB being *very* inconvenient.
@@alouisschafer7212 I don't doubt that, But I would love to see the results of 2x8pins powering the 4090.
@@Symba_Lysm I think they would handle it. 600W might be really pushing it but Stock or undervolted power draw would be fine.
Three or four 8 pins would be better just for redundancy and maximized current handling capacity for lets say a worst case like some insane overclock, a weak connection or one plug not being plugged in.
@@alouisschafer7212 I agree with that, but, IF they can fix the main connection, as the cable itself seems fine, It's just a matter of fixing that connection (Which I think they can).
I'm expecting Lisa to point out the 3x 8 pin power connectors on November the 3rd now.. lol
Comment from someone who runs RC models. Up to 60V and 80-100A sustained load, spikes double. "The ability to transfer 10A through one of these pins without heat damage over time is very optimistic". Surely the companies know better than me... And I also did not see the new connector personally. For the normal 8 (6) pin, I think it should be capable of double of 150W as a whole cable, but at the same time the AWG is not thick enough usually. The number of plug in/out on these connectors also matter a lot. It is a bad choice probably motivated by saving the PCB space and as side benefit, opportunity to produce new cables. However splitting the phases per "cable bundle" is a good thing, but this still can be done with independent connectors. This 12pin brings nothing in that regard. You do not want to pull more current than what the cable can deliver, using multiple cables in parallel is not good solution, because difference in length or terminal resistance will create a current difference.
Why make it with internals that aren't long enough to support possible bending though? It's like, the first thing you'll see when you look at power cables inside a PC...cables bending in various ways.
The problem is bending the cable also making the pins lose, hence make it a fire hazard. Long internal won’t help anything.
Well, 12 small cables/pins which should do 600w ? Sure they get hot xD
I've seen miners suffer burn ups of their 8pins when they use the daisy chained style of PCI-e power cable. The way to fix this is to use silicone insulated wires with heavier gauge wire to sink more heat away from the pins. The Pins on the card can sink heat in to the PCB. (The PCB mount versions of this style amphenol connector are higher rated than the inline version). Having thin gauge wires that are braided don't help with heat dissipation. And looking at the test conections they tested at 26c on a bench, not at 60c in a case (the exhaust temp of cards are likely to be near the GPU core temperature).
So "do not flex this flexible cable" is not going to cut it for most consumer protection agencies in many countries. I would argue the install shown is "best case using a riser cable install and a slightly bend since you can't route the cable through the MB. A more typical install is direct to MB with side panel installed. I do not think any case has 3.5cm clearance to the side panel. You will need at least 90 degree bend and likely more 150 degrees to route to back. If they did not want 3.5cm region to be bent, they need to have it solid plastic. Of course, this would mean most users could not install this rigid adapter since it would hit the side panel. You need at least a 90 degree adapter. A consumer safety agency expects a typical user (e.g. grandma), following a typical installation (vertical mount to the motherboard with side panel on) to not melt, perhaps spark, and burn their house down.
I'd say you need a special case for these 4090 / 4080 cards that allows for PSU and video card placement to be done in a way that doesn't bend that cable anywhere near the card.
ah yes, the legendary pizza box case.
@@Born_Stellar The Fractal Torrent tower with the vertical GPU riser will work since it places the PSU at the top you have a strait shot down to the card with no bends.
Guess instead of the 4090 I'm getting the 7900 XT or whatever flagship AMD releases next month.
Rumours stated that AMD will give up this connector and stays on 8 PIN PCI Express
If I had to guess, the 12 pin connector probably costs a lot less than an equivalent number of 8 pin power connectors. Although technically true, I think the board space reason is a bit of a red herring. Connectors and most through hole components are usually a non-trivial cost in the automotive rated controllers that I work on, I assume the case is similar for consumer electronics as well. Additionally they usually require a separate selective solder step after the SMD line. Depending on the process it can be slow. So I think it is mainly a profit margin or manufacturing thing.
I wouldn't even think so. those 8-pin connectors have been around for a while. maybe they save 2c of plastic per card, but the development and research would make it way more expensive. (although it seems they didn't spend a lot on R and D either)
Who knew miniaturizing a power connector and push even more power trough it, would create problems... very smart.
2x12V braided cables and 2x GND braided cables with 1 sense pin was all that was needed. Scale the mm2 for 600W and put a fat connector on it, problem solved.
Hi Buildzoid 🥰👋. Dumb question and this stuff is outside my understanding…… but would creating a motherboard 20/24-pin power connector for a GPU like the RTX4090 or higher powered solve this issue?
Just slap a XT-60 connector on it. It's smaller and rated for 60A which should allow for 720W at 12V.
24 pin don't just carry 12v, why there is so many wires.
Why do we have these problems? Oh yeah. >300W. 👍
NVIDIA: We don't like how PCI-SIG downrated the Molex mini-fit connectors from 9A to 4.5 A, so let's run 9.5A through smaller pins.
As far as the load across the wires goes: It is an issue is an alignment problem occurs that causes too small of a contact area between a terminal on the connector and the video card. Of course if this is an issue then it is the reason that the connector could be considered a bad choice, since that would indicate that the plastic of the connector is not high enough quality to properly isolate each terminal while rigidly holding each terminal in correct alignment. The fact that there has been the real issue of some pins on the connector melting also indicates that the rated power limit of the connector has already been reached and should not have been used in the first place.
[[[ EDIT: 1) The 30 mating cycles durability rating is what will get you, since it indicates fairly limited durability of the connector for maintaining sufficient contact area at the terminals. 2) Thicker metal improves not only current capacity but also reduces voltage drop. But this requires the terminals to make excellent contact. 3) Tin plating on some pins? Really? Oxidation alone will limit the life of this connector as opposed to gold. Probably made like this since Amphenol knows the connector has low durability anyway. ]]]
I'm thinking this terminal is going to be a backwards step toward the days of the 4=pin Molex power connector.
I am confused: is the card trying to balance the current in each pin somehow?
The point of multi-pin connectors is that each individual pin has random resistance, and you want to carry most of the current over the few pins with least resistance. This is usually achieved by just connecting the pins together on the card end -- which keeps the voltage constant across all of them, which achieves what we want.
The point is, if a pin (or wire etc) suddenly has high resistance, you want to stop the current flow through it. This usually prevents melting until all of them have high resistance...
Nah. The point of the single connector is to safe money on nvidia's part. 30-90c per card they can squeeze more out of the customer. It's a simple as that.
think of the connector as a set of parallel resistors. If the card is pulling 600W at 12V it will pul 50A through those resistors regardless of how much total resistance they have.
No, I'm very sure the point is to decrease contact resistance by creating as many parallel current paths as possible. There is no way to route 20A or so through a single pin of an 8-Pin connector that's not going to make it glow red-hot in no time. Also, while there is some random spread of resistance over the pin, they are all in the same ballpark and because they all carry current that gets combined to flow through a single "resistor" on the board (actually DCDCs converting voltage for the chips) which has a much higher value than the resistance of the pins, the current distributes roughly equally. The point is to lose as little as possible power in those connectors because a) that means a voltage drop which the card will not appreciate possibly and b) the lost power will melt the connectors.
Ehhh that is not how it works. As pointed by AHO wires in this case behave like resistors in parallel and they are splitting total current through the wires (in this case pretty much equally so six 12V wires means 1/6 of total current is going through each wire), basically simulating current carrying capacity of for example single much higher gauge wire.
For this application resistance of copper wires is negligible at these distances, it's in the single digit mΩ territory.
What actually melts these connectors is, you guessed it, bad connection and subsequent arcing which produces MUCH more heat then the constant current passing through the wires.
Just to put it in perspective, even if the wires used for this cable were 18AWG (didn't read the specification but they are probably 16AWG) and you pull two out of the six 12V wires so only four are left, cable could still provide 50+ amps of current without being anywhere close to melting plastic housing of that connector or wire insulation.
If they don't want you to bend it within 35mm of the connector, then the body of the connector should be that long back over the cable,
But am I missing something? In automotive gear, you'd use something like 8mm2 wire and something like an anderson plug for 12v @ 50a.
Yes, it is called crimping. The problem with horizontal stress is moreso that the pressure can pull the connector out of the housing, its only held in place by the tiny metal tangs sticking out the sides. If these get bent enough, the connector can end up making limited and/or no contact with its counterpart.
Good video, but I respectfully disagree in some points:
- Nvidia didn't come up with this connector. It is a new standard designed by PCIe-SIG and Intel, and all manufacturers of PSUs and high powered PCie cards are adopting it.
- The connector is an excellent design, it reduces cable clutter a lot, also size and costs.
- The real problem (a serious fire hazard actually) is the position on top of the card (that is Nvidia failure) and going upwards when the card is already very tall, in most cases there is no way you can have 30mm extra tall card and another 35mm clearance in horizontal orientation (which don't forget is the standard) so users are forced to use vertical mounts ( using expensive and crash-risk risers for pcie-4 due to signal degrataion , and directly impossible for PCIe-5), and then having the cable exerting pressure one way or the other due to cable management needs and space restrictions. So it is a mechanical cause, not electrical, and it could have been easily avoided by including a 90 degree adapter by default, with the cable resting relaxed along the top edge of the board.
- Proof of this is that the melted connector only have melted the upper row of pins, which means uneven pressure pushing the bottom pins and pulling the upper pins.
In fewer words, it is a good design but a bad implementation.
Apple: you holding the phone wrongly...
Nvidia: you bending the power cable wrongly...
Rtx4090 should come with its own cable for the wall outlet.
lol yeah
Or power supplies could add a 24 or 48V rail haha
48V x 12.5A = 600W
24V x 25A = 600W
12V x 50A = 600W
The moment that I saw that connector I knew something like this would happen.
this needs a lot of attention! very unfortunate! someone needs to file the proper complaints to get this recalled!
Surface area contact is important in mating connections such as this. Specifically surface area contact with sufficient pressure to ensure good contact.
Smaller pins will have smaller contact areas. When stress is applied to the connectors, the pins can shift and lose complete contact. The stress on the pins can reduce contact pressure from damage to the contacts as well, as you described with the seam.
More power through smaller contact area, then reducing that contact area and pressure… That is a recipe for increased power transfer resistance, which means more heat.
The biggest thing to watch for is stress on the connectors, not necessarily how/where the wires bend. It’s just that some bend types are more likely to induce lateral or other stress to the connector.
100% agree. No clue on what glue they where on thinking that thing was ok.
The glue that they had left over from the RTX 2000 FE cards.
its call money clue.
smaller pcb = less cost, beefy cooler and take a big slice from the AIB pie with more and more FE sales.
Question though... how is an average consumer supposed to know about this i have a few friends who aren't into PC hardware as i am, and i don't know how they would know about this ever
It is a fair question, however I feel that if you buy a 4090 you kinda are into computers/hardware. Only enthusiasts buy these cards imo. And whenever you buy such an expensive piece of hardware, I think it is common sense to google about it for couple of minutes to see the potential issues/stuff that you need to worry or not worry about.
I'm not saying this is normal, but with the connector should come a manual on how to install it and it should include these diagrams. It's a good practice to go through manuals/instructions even if you know what you are doing.
@@dobrescusilviu4189 That discounts all the people who but the priciest prebuilds because moar $$$ is Moar better. And those builders aren't always very good, and receive trauma during shipping.
The fact that they're doing 8~10A DC over such tiny pins and don't even have decent strain relief on those cables is just bonkers.
@@alouisschafer7212 no,
a) 2 8-pins are not rated for 600W so don't do that.
b) 8-pin connectors still have only 3 12V pins so that's 6 pins total so still 8~10A per pin at 600W
If iam not wrong, the "old" 4.2mm pins of pcie 6 and 8 pin connectors should be rated for up to 11.2 amps but are only used with up to 150w/3pins/12v = 4,167 amps per pin, thats what headroom means.
I mean just the connectors, not the whole cable and psu circuit.
The new 3.0 mm micro fit pins are only rated for up to 8.5 amps.
GPU manufacturers are going to innovate with military grade power connector guards next gen. Thanks Nvidia for pushing innovation
Military grade? So, something designed in the early 60s and contracted to 20 different companies?
@@Kios_Antares Andbuilt to the minimum requirements to pass the test
@@Kios_Antares I love that no matter the country you served in, this is always true 🤣
you need to be careful with the words " military grade" that does not equate to the best....by any means...
@@Kios_Antares I think military grade in this context means a plastic shroud that looks like powder-coated metal, preferably earth-tones in color, and fewer RGB leds.
Why dont we just use the EPS connector (CPU one) with 4 12V pairs? Im pretty sure they re used in enterprise and data center GPUs but not the consumer ones
would probably very confusing because the EPS and PCI-e look very similar and it's actually possible to force one into the other if you push really really hard.
@@ActuallyHardcoreOverclocking Im not talking about shoving wrong connectors into eachother. I meant why dont we just use the same connector for both cpus and gpus by design and spec? IMO that would be less confusing.
Are sense pins really that necessary if gpus like NV A100 use the EPS cpu connector
@@ThunderingRoar oh I agree that the 8pin EPS would be way better GPUs. I meant that the transition over to 8pin EPS would be very confusing.
For other connector systems there's plastic retention clips you can get that go in the back of the plug and hold the crimp terminals in. I wonder if that sort of mechanism might have helped in this case.
I do like the idea of tidier cabling with smaller everything, especially when we know the wire gauge can safely handle it but those crimp connectors were always going to be a problem especially with how they deform after repeated insertions. Looking and the diagram, there appeared to be 2 punched pins/domes (the 2 round things) on either side of the receptacle terminal that make contact with the solid pins on the plug terminal. On a larger 8-pin connector, it's probably fine if only one of these is touching the plug pin as it's physically larger and there's less current to push through it.
On the server side of things, I've seen power supplies that plug directly into the motherboard via PCB gold fingers and those can push quite a few amps, though it's a completely different kind of connection. On the Apple side, they use screws to deliver 12V to their motherboards/GPUs which always seemed janky AF to me but I also see this for older style power supplies, namely still used in 3D printing power supplies.
It would be neat if the power delivery standard for PCs also moved to 24V, 48V, etc to cut down on the current needed while still working fine with most connectors without having to worry about arcing, kind of like what we are seeing with USB-PD, but dumber and fixed to 1 voltage for one type of connector.
There's also the other kind of Nvidia solution which is to push for SXM form factors so the GPUs get all their power from the motherboard and cooling rigidity from the case.
I wonder if the 7000 Series GPU's from AMD will use the same connector as this or will stick with good old 3x 8 pin?
Oh, I think if it's at all possible for them to stick with/go back to the 8-[in connectors, they will. I can't see them going with this new one after all the crap that's been coming out on them since they launched.
I think the latter
It is the new ATX PSU standard. Make your own conclusion from that.
@@henryvaneyk3769 New standard doesn't mean automatic switchover the second it comes out. Look at Windows. People are still running Windows 7 and 8 machines, even though 10 is now EOL and 11 is the "new standard". Given the issues this new connector has shown so far, AMD will likely hold off on adopting it till these have been resolved and we're not getting situations where customers are having these things catch on fire/melt. Intel opted not to use them, because they were unnecessary.
@@_Clivey Where has it been "confirmed"? I'm not seeing anything definitive one way or the other, and that included some reports coming out just last month. If you've got a link that definitively says that AMD will be using 12VHPWR on 7000 GPUs, cool. I'd love to see it.
Had dual 8 pins for a 3080 10G undervolted in an ITX case and they burned up, even though I suspect the 3rd party cables might have been a little underspecced this just means that the 8 pin spec is absolutely there for a reason
Meanwhile the Corsair 8 pins are rated for 340W. Their 600W 12VHPWR only has two 8 pins at the PSU end.
that should never happen...
I pushed 330W trough two 8 pins on my Vega and it was fine that's 1.7 amps per pin...
Well yeah 3rd party cables are no joke, you really need to make sure they use the correct wire gauges or you are in for a serious trouble. Standard 8pin is able to safely provide around 200W without melting the insulation, even though technically it is out of spec. But also depends on the ambient temperature etc... How did you push undervolted 3080 over 475 watts?
@@1337Ox he didnt, he claims it was undervolted so 300W tops which is barely 100W per 8 pin. Either he got garbage no name cables, a garbage PSU or his card had a catastrophic failure.
It's not a non-problem, the current PCI-E 8 pin connector is VERY inelegant if you need more than 150w and we badly need a new connector.... but this connector is a complete fail given that every build needs cable management which means running cables at a minimum 90 degree and often even 180 degree angle.
I mentioned that in reddit before the launch just to receive "Doooh, is designed like that" answers.