PCI Standards Group Deflects, Assigns Blame for Melting GPU Power Connectors (arstechnica.com) 130
An anonymous reader shares a report: Nvidia's new RTX 4090 and 4080 GPUs both use a new connector called 12VHPWR to deliver power as a way to satisfy ever-more power-hungry graphics cards without needing to set aside the physical space required for three or four 8-pin power connectors. But that power connector and its specifications weren't created by Nvidia alone -- to ensure interoperability, the spec was developed jointly by the PCI Express Special Interest Group (PCI-SIG), a body that includes Nvidia, AMD, Intel, Arm, IBM, Qualcomm, and others.
But the overheating and melting issues experienced by some RTX 4090 owners recently have apparently prompted the PCI-SIG to clarify exactly which parts of the spec it is and is not responsible for. In a statement reported by Tom's Hardware, the group sent its members a reminder that they, not the PCI-SIG, were responsible for safety testing products using connector specs like 12VHPWR. "Members are reminded that PCI-SIG specifications provide necessary technical information for interoperability and do not attempt to address proper design, manufacturing methods, materials, safety testing, safety tolerances, or workmanship," the statement reads. "When implementing a PCI-SIG specification, Members are responsible for the design, manufacturing, and testing, including safety testing, of their products."
But the overheating and melting issues experienced by some RTX 4090 owners recently have apparently prompted the PCI-SIG to clarify exactly which parts of the spec it is and is not responsible for. In a statement reported by Tom's Hardware, the group sent its members a reminder that they, not the PCI-SIG, were responsible for safety testing products using connector specs like 12VHPWR. "Members are reminded that PCI-SIG specifications provide necessary technical information for interoperability and do not attempt to address proper design, manufacturing methods, materials, safety testing, safety tolerances, or workmanship," the statement reads. "When implementing a PCI-SIG specification, Members are responsible for the design, manufacturing, and testing, including safety testing, of their products."
Deflects? (Score:2)
Sounds like they, the PCI group, are in the right here. What a shady title, seems like yellow journalism. I wonder if nVidia paid for this titling so as to control the narrative.
Re: (Score:2)
Re: (Score:2)
More like if NEC specified that each outlet shall provide 20A and then suggested that it should use plug XYZ2a, where XYZ2a is a plug capable of handling 10A.
Shame on the electrician? Sure. Shame on NEC? You bet! Doubly so if they later deny any part in the fire hazard.
Re: (Score:2)
Let's assume I have an interior decorator who said I should paint my room red. If my paint supplier gives me toxic paint, is it the interior decorator's fault? The paint supplier should have either not used toxic ingredients, or they should have admitted they don't know how to make it safely.
Re: (Score:2)
Re: (Score:2)
WHOOOOOSH!!!
Yes, I know. I didn't start that analogy.
Re: (Score:2)
Yup. And it is not like their is a shortage of better designs out there that do exactly this. My favorite being PP15 as it is rock solid, cheap, and simple. Deans Ultra, XT90, and XT120 are some other common good ones.
It's frustrating that the computer industry is stuck in hold-my-beer mode when it comes to power connectors this. They got away from it with ATX and then came back with P4, EPS-12V, and finally this one.
Re: Deflects? (Score:2)
I'd argue against Deans Ultra as there is no shield of any kind for the male pins. (Yes, they're supposed to be unpowered; unless of course they are partially inserted.)
These days, there's no excuse for a contact that's exposed from the side when partially inserted.
Re: (Score:2)
They were holding it wrong?
There is a broad spectrum there from any idiot should have been able to plug it in properly all the way over to it's practically impossible to fully plug it in.
Apparently it's harder than average to get right.
Re: Deflects? (Score:5, Insightful)
Thre's barely, if any, physical and auditory feedback once the retention mechanism is engaged, any no visual difference at all. If not engaged, then it can start to pull loose while performing cable management, at which point you generally don't have a visual angle to easily notice the connector coming partially loose, while it still boots up just fine right up until it melts.
Re: (Score:2)
I haven't run across that connector yet, but it sure sounds like a design problem. Positive action matters in human factors!
Re: (Score:2)
It was also a very poor quality plug. A decent one may have had a better tolerance. And one that bad should not be legal.
Re: (Score:2)
But isn't Nvidia part of the PCI Standards Group? So either way Nvidia is at fault for faulty connectors.
Re: (Score:2)
Re: Deflects? (Score:2)
I would hope my $1500 graphics card came with actual Molex brand connector.
Re: Deflects? (Score:2)
NVidia is a member of PCI-SIG. AMD is also a member.
The problem here is a crappy connector design from PCI-SIG.
Re: (Score:2)
I have to agree. If the person building my house uses cheap 10 amp plugs on a 20 amp circuit it's not the NEC's fault for underspec-ing the 10 amp plugs.
Yup. It's the builder's fault for using incorrect material, then the building inspector's fault it that all passes inspection.
Re:Deflects? (Score:4, Insightful)
Nonsense. If the contractor puts in 20A outlets on a 20A circuit and then the user plug a 20A device just in enough that they barely get contact, the resulting fire is the user's fault. Although because this type of connection is aimed to even work when a complete idiot does it, there likely will not be a fire, the outlet will just get charred a bit and the plug may melt or not melt. It will be a mess, but the house should not burn down.
Seriously, what is this stupidity? Electrical connections inside a PC are not aimed at "every idiot can do it" and what we observe is what happens if some really big idiots do it. Expecting power connectors to work reliably when not plugged in correctly is just dumb.
Re: (Score:2)
If earlier connectors did not have this problem then why are we so quick to blame end users now? NV's own 3090Ti reference card used an early version of the 12VHPWR connector without issue. Most of those connectors were fed by 3 8-pin PCIe connectors for a peak of 450W with no circuitry to precisely regulate power per connector or anything like that. Now that we're trying to push 600W across the finalized PCI-SIG connector, and it's melting.
This connector is almost entirely of NV's own creation. It looks
Re: (Score:2)
I am not "quick" to blame the luser that messed it up. I have actual knowledge of how these things work. Also, most of these connectors as are _not_ melting. It is just some select idiots that make it melt and then try to blame somebody else for their incompetence. Obviously, there is not much redundancy when pusing 600W over a 600W connector. But the only real blame on NVIDIA is that they underestimated how incapable some people obviously are. Yes, they should have taken that into account, but the blame is
Re: (Score:2)
The problem here is an overestimation of one's own knowledge.
It's well known what the power draw of these cards are.
I would tread carefully while installing the power connections for it whether or no this whole mess had occurred.
Perhaps if you're unfamiliar with the intricacies of connecting fucking 40amp connectors, then perhaps you are unqualified for what you're doing.
PCI-SIG should perhaps be berated for trying to push that much current over that many connector
Re: (Score:2)
That would disqualify anyone from buying or using these cards. Which would be absolutely hilarious, by the way. Good job nVidia for selling a card that requires an electrician's license just to plug in the power connector.
Re: (Score:2)
That would disqualify anyone from buying or using these cards. Which would be absolutely hilarious, by the way. Good job nVidia for selling a card that requires an electrician's license just to plug in the power connector.
Complete blullshit. It just means you do the installation at your own risk.
Re: (Score:2)
Obviously not! Everyone having the problem gets a refund.
Re: (Score:2)
Not because they deserve one though. Also obviously.
Re: (Score:2)
That would disqualify anyone from buying or using these cards. Which would be absolutely hilarious, by the way. Good job nVidia for selling a card that requires an electrician's license just to plug in the power connector.
It does disqualify most people from installing these cards, if we use that word literally.
Of course lots of people do things they're not qualified to do quite successfully. And if they're unsuccessful, they often learn a valuable lesson on how to be more qualified to do it.
As for whether or not NVidia should have used a connector that pulls 40 fucking amps, I agree with you. Probably not wise on their part.
Of course, I get their reasoning.
12VHPWR is "the next thing" as far as standardized high power de
Re: (Score:2)
I can't help but to agree with you.
My apologies for causing that situation!
The problem here is an overestimation of one's own knowledge.
Indeed. And worse: Not a clue about it either and then blaming others. Classical Dunning-Kruger effect. I do tend to go regularly outside of what I reliably know and can do (otherwise, how can you learn?), but if I break things I am completely clear on it being my fault and own that. That is how adults behave.
But as you pointed out- ultimately, these guys are selling components. If you don't have the knowledge required to handle components that draw this much power, leave it to a professional.
And that really is the core of it. This is not an "anybody can install it" product and of you make basic mistakes like not plugging in a connector correctly, it
Re: (Score:2)
Well gee we went from almost no reports of PCIe 8-pins melting (except in cases where they were clearly overloaded) to 12VHPWR connectors melting in reportable numbers. The people experiencing these problems are getting refunds from the AIBs that sell 4090s, and it has been shown time-and-again that the users experiencing these problems are using tried-and-true cable management techniques that have been just fine for:
MOLEX
PCIe 8-pin
8/16-pin EPS
20 and 24-pin ATX
SATA power
etc.
cables. But for some reason thi
Re: (Score:2)
It is not the bends. It is people not plugging the connector all the way in. Try to keep up.
Re: (Score:2)
Is *is* the cable bend! It's unseating the connectors by just enough to change resistance on one of the pins from the connector tilting in the socket, leading to thermal runaway. This phenomenon has been shown already from multiple sources, and it is now widely-recommended that you must be VERY careful when routing the power cable to a 4090Ti.
Re: (Score:2)
Nope. That explanation about "one pin" does not work. The numbers do not add up. You only get thermal runaway if a rather large number of pins is not seated correctly and you only get that when the connector is not pushed in all the way.
Re: (Score:2)
Nonsense. If the contractor puts in 20A outlets on a 20A circuit and then the user plug a 20A device just in enough that they barely get contact, the resulting fire is the user's fault.
One could also argue that it could also be the electrician's fault for not using an arc-fault breaker, or the device manufacturer for not detecting an unreasonable voltage drop when load is applied.
Companies that build things like plug-in car chargers do a lot of work to reduce the risk of these sorts of failures precisely because that sort of thing happens in the real world with alarming regularity. In spite of that, fires still happen sometimes, but way less often than they would if the chargers didn't h
Re: (Score:2)
Nope. An arc-fault breaker would not help as there is no arcing. If you really want all device manufacturers (i.e. _ever_ plugin in electrical appliance) to detect an "unreasonably high voltage drop", you are out of your mind. As mains voltage varies and a small drop can be quite enough to cause heating up, this would mean some sort of sensor in the electrical outlet and a communications network. So if you do not mind that your hair-dryer suddenly costs $500 more, sure. You have no clue how things work in t
Re: (Score:2)
Nope. An arc-fault breaker would not help as there is no arcing. If you really want all device manufacturers (i.e. _ever_ plugin in electrical appliance) to detect an "unreasonably high voltage drop", you are out of your mind.
The way I see it is this: If a device is designed to pull continuous power for an extended period of time that is more than a small fraction of the outlet's capacity and is designed to be plugged in and unplugged for an extended period of time and is designed to be used unattended, it should do those sorts of safety checks.
But those limits rule out darn near every appliance you could reasonably think of except for your car charger.
As mains voltage varies and a small drop can be quite enough to cause heating up, this would mean some sort of sensor in the electrical outlet and a communications network.
Not necessarily. Take the running average of the voltage over thirty seconds
Re: (Score:2)
Nope. An arc-fault breaker would not help as there is no arcing. If you really want all device manufacturers (i.e. _ever_ plugin in electrical appliance) to detect an "unreasonably high voltage drop", you are out of your mind.
The way I see it is this: If a device is designed to pull continuous power for an extended period of time that is more than a small fraction of the outlet's capacity and is designed to be plugged in and unplugged for an extended period of time and is designed to be used unattended, it should do those sorts of safety checks.
Nope. You clearly have absolutely no clue what you are talking about and what can actually be done without driving cost though the roof.
Re: (Score:2)
Nope. An arc-fault breaker would not help as there is no arcing. If you really want all device manufacturers (i.e. _ever_ plugin in electrical appliance) to detect an "unreasonably high voltage drop", you are out of your mind.
The way I see it is this: If a device is designed to pull continuous power for an extended period of time that is more than a small fraction of the outlet's capacity and is designed to be plugged in and unplugged for an extended period of time and is designed to be used unattended, it should do those sorts of safety checks.
Nope. You clearly have absolutely no clue what you are talking about and what can actually be done without driving cost though the roof.
Maybe it is more challenging than I think it would be, at least for mains power. That said, lots of companies are doing things like this, so it obviously can't be *that* hard.
But in a tightly controlled setup like a computer power supply and GPU, it's mind-blowingly easy. The simplest approach requires just two extra return wires on the bus and one communication wire, a precise high-impedance voltage meter and four transistors to switch its source, and an 8-bit microcontroller to drive it all. Measure th
Re: (Score:2)
Nope. An arc-fault breaker would not help as there is no arcing.
Forgot to reply to this. Every time I've had a bad contact situation with an AC outlet, I've heard sizzling, i.e. micro-arcing. A high-resistance contact with low-resistance surfaces separated by a short air gap is going to arc. Whether it is enough of an arc for an arc-fault breaker to detect or not, I couldn't say, but preventing electrical fires is quite literally what they are designed for.
Re: (Score:2)
You are seriously incompetent. A partially-plugged in connector does not arc. A connector with bad surface or just about not plugged in will arc. In addition, "sizzling" can have a number of causes and arcing is only one of them.
Well. You are just another person with no clue but a deep conviction that it must be the vendor's fault and that users are never to blame, no matter how stupid. All your claims do is clearly mark you one of the stupid.
Re: (Score:2)
You are seriously incompetent. A partially-plugged in connector does not arc. A connector with bad surface or just about not plugged in will arc.
A partially plugged-in connector also isn't particularly high-resistance. You start to get high resistance when the contact surface area drops significantly below the cross-sectional area of the wire (I'm ignoring skin depth here), which shouldn't happen until you get pretty close to being unplugged, at least for most connectors I've encountered.
In addition, "sizzling" can have a number of causes and arcing is only one of them.
Water incursion, and actual overheating insulators, sure. But one of those is pretty darn unlikely, and the other one doesn't happen instantly when the load is ap
Re: (Score:2)
Except that the spec was bad from day 1. So bad it should never have seen the light of day. EPS-12V and P4 are similarly bad designs. Only in the computer industry do you see this hold-my-beer school of bad design when it comes to power connectors with adding more pins for more amps.
And it is not like there is a shortage of better connectors that do handle this level of DC amps out there to choose from. PP15, Deans Ultra, XT90, and XT120 are some of the first that come to mind off hand.
Re: (Score:2)
Re: (Score:2)
You will get this when disconnect/connect any high ampage or high voltage connector while it is live. Just a question of how visible it is.
Re: (Score:2)
Indeed. This disaster is a combination of NVIDIA loading the connector up to the limit, but really users not plugging it in correctly. To be fair, the only thing NVIDIA is guilty of here is not expecting the level of user incompetence that can be observed. If this thing is plugged in correctly and the maximum cycle number is respected, the connector works just fine.
Re: (Score:2)
Indeed. This disaster is a combination of NVIDIA loading the connector up to the limit, but really users not plugging it in correctly. To be fair, the only thing NVIDIA is guilty of here is not expecting the level of user incompetence that can be observed. If this thing is plugged in correctly and the maximum cycle number is respected, the connector works just fine.
You forgot C. nobody on either end of the connection detecting the voltage drop and shutting down, D. the spec not requiring such detection....
I would argue that a proper standard for a high-current power connector should specify a means for the power supply and the device to communicate with one another, providing enough data so that the card can recognize a dangerous voltage drop and go into a limp mode until the problem can be fixed. This can be done in multiple ways:
Re: (Score:2)
Certainly there could be, if the connection was bad enough, and the PSU lacked the current being demanded with the high resistance, but the connector already runs *really* hot, and sometimes these things are plugged into 1000W rails on the PSU side.
That's more than enough current to melt the connector to plastic slag long before the voltage drops even a mV.
Really, this connector just runs too damn close to the limit to account for any kind of error.
Either
Re: (Score:2)
It's also just a shitty connector for this job. It's the same connector as what's used to bring power to the motherboard, but in that context you don't have to make awkward bends to attach the cable. In most cases, and with most power supplies, you do have to do that to attach it to the video card. The more conductors it has, the more of a problem it is. The longer the card is, the more of a problem it is. The user has in many cases already demonstrated their ability to plug in a mini-fit jr, when they plug
Re: (Score:2)
I wonder if nVidia paid for this titling so as to control the narrative.
Control what narrative?
Standards group: "Use this connector"
NVIDIA: "Okay"
NVIDIA: "It blew up!"
Standards group: "We are not responsible for testing the safety of the connector we told you to use."
Re: (Score:2)
Dude, nVidia chose to use that connector in their product. They should have made sure the build quality and specifications met the PRODUCT requirements. That's really like an SUV maker choosing to use bicycle tires and then upon failure blaming whoever made bicycle tire specifications. They really should have determined what kind of tire to put in their SUV. They should have looked at the specification to see if it meets their capabilities.
Re:Deflects? (Score:5, Insightful)
Re: (Score:2)
Well, if somebody is too dumb to plug in a connector, who is at fault? The designer of the connector? The people that selected the connector? Or the person that plugged it in wrongly? I say the person that plugged it in wrongly is the guilty party here. If somebody, say, puts their CPU in backwards, nobody sane would blame the CPU manufacturer or the mainboard manufacturer for that, after all unless it was extremely easy to do. That is clearly not the case here of we would see a lot more of these incidents.
Re: (Score:3)
Well, if somebody is too dumb to plug in a connector, who is at fault? The designer of the connector? The people that selected the connector? Or the person that plugged it in wrongly? I say the person that plugged it in wrongly is the guilty party here.
Indeed, user error is a thing. However one needs to question: a) if there's a significantly higher rate of user error with a new plug than one used previously, then it seems nonsense to lay all the blame at the user.
It is the job of engineers to account for users. This is why buildings have fuses, arc fault current interrupters, why power outlets are shielded, and why companies exist to certify that something won't burn your house down. If a product is shitty enough that many users can trivially cause a pro
Re: (Score:3)
My guess would be the rate of error is pretty standard, but the standard error behavior is just the GPU not working. The only thing that makes this "news" is "OMG! It _melted_!!! IT could have set the house on fire!!!!!", when no, it could not. The plastic used has a far higher ignition temperature than melting temperature and it will self-extinguish or rather not even get there because it will short out before and the PSU will switch off. The whole thing is stupid.
I do agree that engineers need to take use
Re: (Score:2)
why does the device still function with an improperly inserted connector?
They added extra pins for sensing
They should USE THEM
This is just badly engineered. There are quite a few high-current connectors in a modern computer. How often do we hear about them melting or otherwise?
So, what gives?
Re: (Score:2)
This is just badly engineered. There are quite a few high-current connectors in a modern computer. How often do we hear about them melting or otherwise?
The other connectors do have these very same problems. They have for decades. This isn't a new/unique issue. There are a few factors here for why it's in the news. First, Nvidia is already in the crosshairs over card availability the last couple years. Second, the price of the 4090 being as high as it is. Third, it's a new product and people pay more attention to new things. Over time people adjust to the nuance of the new thing, or stop caring as much because it's not new. Now, I don't have numbers or the
Re: (Score:2)
Re: (Score:2)
This is not a question of build quality here. But bad spec. Does not make nVidia blameless as they designed something using said bad spec and have been doing so for awhile. The EPS-12V is another spec that was fucked up in the exact same manner as this one.
100% manufacturing issue (Score:3, Insightful)
It's sad they felt the need to defend themselves on this.
No matter what the spec said, the products should have been tested. Especially for over heating as Nvidia is perfectly aware normal people game their asses off all weekend, crypto bros are wasting power 24x7 and the benchmark kids are always pushing the limits for funsies.
If I had a melting video card the last folks I'd think of to blame would be the industry spec writers. The article headline is bullshit. There is no blame to deflect as they had nothing to do with the cards melting.
Re:100% manufacturing issue (Score:4, Insightful)
If only it were that simple. Everything I've read/heard is that it comes down to improperly seated connectors, and subsequent arcing. Has that changed? That makes the whole thing very fuzzy. I think NVidia has handled it right so far... they'll honour the RMA on the card regardless. Finger-pointing aside, they're doing the right thing by the customer.
If you're implying they didn't test the design, I'd call bull. The actual number of meltdowns is not high. it's even reasonable to conclude they tested improperly seated connectors and simply didn't experience it.
Re: (Score:2)
If the connector is able to be seated in such a way that it gets hot enough to melt itself, I'd say it's a duff connector design. Sure the user should check, but any time anything potentially dangerous like that is involved we have to design it to be fool-proof.
The locking mechanism on the connector and the tactile feedback when it is plugged in are poor. It can appear to be inserted correctly when it isn't. They appear to have designed it that way primarily to fit into small form factors like Intel NUC and
Re: (Score:2)
I have to agree; the spec needs to be inherently safe in order to actually be interoperable. The manufacturer needs to test it for safety against a reference design.
It still sounds like a duff design. With hindsight, it seems like the connector should be low insertion force with a latching mechanism rather than friction-fit.
Re: (Score:2)
If the connector is able to be seated in such a way that it gets hot enough to melt itself, I'd say it's a duff connector design.
Agree. In that case I would call 100% of connectors used in computers a duff design. It is after all possible to actually melt any of the connectors in the computer by not seating them correctly. Which should be obvious considering this is not a new connector design, the pins and housing are the same Minifit-Jr connectors we've been using for over 2 decades now.
They appear to have designed it that way primarily to fit into small form factors like Intel NUC
What's laughable is how you're just making shit up now. Actually this connector has the *opposite* requirement. Specifically one without sharp bends
Re: (Score:2)
Adding a locking mechanism makes the connector deeper.
And yes all the connectors used in computers are crap. Molex, Berg, the 2.54mm pin headers, SATA, all of them.
Re: (Score:2)
Everything I've read/heard is that it comes down to improperly seated connectors, and subsequent arcing
That is a design/manufacturing problem. It should not be possible to create an unsafe situation like that. Either the design has too tight of tolerances, or the manufacture was out of spec.
This is a consumer product. Manufacturers are responsible for preventing unsafe operation. Where not possible to prevent, they are required (by law) to warn consumers of potential hazards and advise on avoiding same.
Re: (Score:2)
The product works as advertised. If plugged in correctly, it does not overheat. Testing all the ways it can be plugged in wrongly is both impossible and out of spec. What NVIDIA should probably _have_ tested is how likely the user is to mess it up.
Re: (Score:2)
This is not a consumer product. A consumer product is a pre-packaged PC.
Re: (Score:2)
When I was young and... less wise... I setup install a couple of 1200W amplifiers in my car. Each with their own power run back to the capacitor/battery under the hood.
Now, if you've ever done something that stupid, you've almost certainly learned the hard way that shit which is designed to handle 100A aren't toys.
When I bought correctly rated terminals, and they melted down to slag... it never actually occurred to me to blame the terminal manufacturer.
I immediately knew w
Re: (Score:2)
Your car audio is intended to be professionally installed. It's not a video card expected to be installed by dads at Christmas.
Says who? You?
Your video card is a component that goes inside of a computer.
Car Stereos are installed professionally, and by amateurs.
The same is true of computers.
That you melted it says you weren't a professional.
Precisely. I was not.
Very few car owners mod their cars. PC assembly and mod has been an industry since very early days. I love the car analogy but not a good fit on this story. Nvidia didn't properly design or test their product for the expected/common users and use cases.
I disagree, entirely.
Comparatively few people mod their cars.
Also, comparatively few people mod their PCs.
There have been system integrators since the dawn of time as well.
For me, I've found that both generally followed age.
When I was young, I did both.
Now? I'm too fucking busy. I pay others to do it for me.
But
Re: (Score:2)
I mean, you are the shit-for-brains here saying the "card" melted, so there is that.
Re: (Score:2)
the cards didnt melt, the power connector did and then its most likely because users can't access the full port to make sure the thing is seated cause of vanity of nividia to try and hide the connector inside the heatsink
its important to hide because nvidia could not figure out where to put connectors on a 1 foot long, 5 inch wide, 2 and a half inch thick 5lb, 1600$ graphics card
Re: (Score:2)
not exactly, your making it out to be like "oh I shorted an outlet in my house, thus the house burned down" but its ok, I get it now, you are a dipshit moron
Re: (Score:2)
I dont disagree with you except the cards melting part THE CONNECTOR which is only one part of the card melted, and heyo it will still work without those pins just at a lower power consumption and performance level (that's the entire point of having the sense lines, if you only plug in half the connectors it knows)
god, you are dense
0.02% failure rate (Score:2)
Gamers Nexus reported a failure rate of 0.02%, due to the cable not being plugged in all the way.
Re: (Score:2)
Re: (Score:2)
Funny how that wasn't a problem for the 3090Ti with the three-prong-to-early-12VHPWR connecor.
Re: (Score:2)
I suspect it absolutely is a problem for the 38 people that have one of them. (I'm one of them)
My 3090Ti was my first exposure to a 12VHPWR connector, and when I first plugged it in, I was amazed by how fragile and dainty it was for the kind of current this thing could pull, and this isn't even a 4090 (though it's not far in terms of current). Fully plugged in, it gets hot. Burn you hot.
I expected people would have been melting these on accident long before the 4090 snafu ha
I blame nVidia (Score:2)
Idiots selling overpowered GPUs that require hundreds of watts and sell those for over a thousand dollars? Fuck them.
We need affordable, low-power GPUS. Make older GPUs with new node processes. It will allow them to make a lot more GPUs per wafer and people won't need 600+ watts power supplies in their computers.
Enough with the snob rich gamers demands already.
Re: (Score:2)
You tell 'em, Steve Dave!!
Affordable, lower power GPUs? (Score:2)
The world is absolutely awash in them. Are you having trouble finding one?
Re: (Score:2)
Affordable lower power GPUs that are not four or five generations behind.
Re: (Score:2)
Oh, so all you want is better (same performance, lower power needs) and cheaper.
I'd like that, but free, please. As long as we're wishing upon a star.
Re: (Score:2)
I bought a 3090Ti because I have exactly no need for an affordable low-power GPU.
You're more than welcome to buy a budget card.
Re: (Score:2)
I replaced my RX560 with a Ryzen 5600G for $129 and it's faster with much lower power.
I put one in my NAS for transcoding too. On-die latency.
Why would I buy a low-power PCIe GPU?
Re: (Score:2)
I can't wait for the "7600G", whatever name they end up using.
Re: (Score:2)
I replaced my old gtx 260 with a ryzen 2600g and it's even more extreme... Faster graphics and hugely less power. 10 years of gaming on that thing and I was still playing older games... but so much better on the Ryzen integrated graphics.
Re: (Score:2)
We need affordable, low-power GPUS. Make older GPUs with new node processes.
No.
That doesn't make sense for many reasons.
The first reason is that the old design would have to be revised for the new process. It's not just "print this out smaller".
The second reason is that the new GPUs are faster at the same clock rate. You would be leaving performance on the table.
The third reason is that they wouldn't be cheaper. In fact, making more GPUs on the same new architecture spreads the development costs across more products, making each new device cheaper. If they had to both revise the ol
Microfit (Score:5, Interesting)
The connector comes from the 20+ year old range of the Molex ‘fit’ series which has the mini fit, micro fit and mega fit.
Molex derate the per pin current by number of pins carrying high current. For example they may allow 10A for a 2pin micro fit, but 7A for a 6pin micro fit. This is purely due to the connector assembly being able to dissipate the heat from 10ish milliOhms of contact resistance.
It’s generally a reliable industrial grade connector, but we have had instances of burned out pins even when operated below spec. Remember that power dissipation is I^2.R, so the difference between 4A and 8A increases dissipated power by 400%.
Thus we recommend a 50% derating from the data sheets to cover cases where there is imperfection in the fabrication or mating of the connector.
Also, I’ve never been a great fan of //ing lots of smaller current pins together. It has a bad engineering smell for many reasons. Better to use single connector pins rated for the application, or increase your derating.
Re:Microfit (Score:4, Insightful)
Exactly. The trouble with parallleing 6x 10-amp pins and assuming that you can now shove 60 amps through them is thermal runaway. If one pin has slightly higher resistance than the others, which it will, it will heat up, raising its resistance further. Meanwhile, the other 5 pins have to shoulder the extra load. Out of those, one will have a slightly higher contact resistance.
Lather, rinse, repeat, call the fire department early and avoid the Christmas rush.
Re:Microfit (Score:4, Insightful)
Molex derate the per pin current by number of pins carrying high current. For example they may allow 10A for a 2pin micro fit, but 7A for a 6pin micro fit.
According to the product specification [rs-online.com] Molex does not commit to a current carrying capacity at all. They literally only provide a maximum current rating, that rating is provided in table form, and they also literally say it is only a guideline:
IOW they don't "allow" (or in any other way encourage) any number of amps on their connectors, it is entirely up to the implementing party to decide how much current a Mini-Fit Jr. connector can carry. And that's of February of this year, which seems pretty current.
So... the PCI-SIG doesn't take any responsibility, and Molex doesn't take any responsibility, which leaves the manufacturer of the card absorbing it all. nvidia has committed to replacing the cards sold under their name, and I tend to assume the other vendors of 4090s are going to do the same.
Thus we recommend a 50% derating from the data sheets to cover cases where there is imperfection in the fabrication or mating of the connector.
That seems wise. Too bad nvidia didn't figure that out.
Also, Iâ(TM)ve never been a great fan of //ing lots of smaller current pins together. It has a bad engineering smell for many reasons.
Amen to that. There are plenty of connectors which could have been used there, many of them also made by Molex :) I also think that it might just be wise to use a higher voltage, like the 48V which is already commonly implemented in non-PC power supplies due to telephony. Sure, that means another step-down regulator, but it would still be preferable to this BS.
I was already prejudiced against the Mini-Fit Jr. connector before this happened, though, because it is such a fiddly little PITA to work on, and it has such delicate contacts. They really are easy to mangle.
Re: (Score:2)
I agree on both: To get good reserves do not load connectors more than 50% and use connectors that can handle the whole current in one contact. These do exist, even for the 50A we are talking here. The problem is likely that nobody wanted to get the blame for introducing a completely new connector. Also note that a new connector requires an adapter on the other side and that will have the micro-fit jr. connector in any case as the PSU comes with it. I wonder whether anybody had the other side of the adapter
Re: (Score:2)
You propose a connector with 100A-capable pins?
Have you ever fucking worked with 2 gauge wire before?
There is absolutely no problem with splitting the current over multiple conductors.
Re: (Score:2)
No, I propose 50A capable pins. And yes, I am aware that requires AWG8 cables (at least Molex wants them) and yes, I know how thick they are because I also do some stick-welding. The problem is not the conductors because they reliably have the same resistance, the problem, is the connector pins. Splitting current over several pins in connectors is always asking for trouble, unless you load them lightly, say 50% max.
Re: (Score:2)
50% loading capacity and single-conductor.
For a 50A load, that'd be 100A. Either way- fine 50A. 8 gauge is still pretty horrible.
Do you propose, perhaps, a fat ass pin that still has enough capacity in multiple lower gauge wires? (say, bonded at the connector)
Because I can say this for certain: Nobody is ever going to agree to stick even even 8 gauge in their computer.
Because it sucks to deal with, and because multiple conducts is a perfectly reasonable way to do t
Re: (Score:2)
Silicone-jacketed 8ga super-fine-strand wiring is not at all horrible. It is in fact a hell of a lot easier to handle and manage than these stupid ATX cables with a shitload of conductors. It is readily available and not very expensive. It is commonly used (in a variety of weights) for hobby battery pack leads. 6s and 8s high-current lipo packs are now moderately common, not for casuals but for drag cars and fifth scale. For instance, here's an 8ga XT90 extension cable [aliexpress.com] for $6.99, that's a 90A connector too.
Re: (Score:2)
Silicone-jacketed 8ga super-fine-strand wiring is not at all horrible. It is in fact a hell of a lot easier to handle and manage than these stupid ATX cables with a shitload of conductors. It is readily available and not very expensive. It is commonly used (in a variety of weights) for hobby battery pack leads. 6s and 8s high-current lipo packs are now moderately common, not for casuals but for drag cars and fifth scale. For instance, here's an 8ga XT90 extension cable [aliexpress.com] for $6.99, that's a 90A connector too. I'm seeing mostly 10ga pigtails right now, or I would have linked one of those. They make panel mount XT90 connectors too, and I speak from experience when I say that XT connectors are ultra reliable. They take quite a bit of oomph to [dis]connect, however.
Oh god, I couldn't agree less.
They're fucking terrible.
That's a matter of opinion though, so you're free to have yours.
They will if they know what they're talking about.
Nope. That's why it isn't happening.
There are several reasons why it's not a great idea.
What a pointless statement.
There are pros and cons to single and multi-conductor power delivery. There are several reasons why single conductor is not a great idea, too.
What we have, is the one that most people are happy with- and that is not fat fucking wires with shit flexibility.
Re: (Score:2)
What we have, is the one that most people are happy with- and that is not fat fucking wires with shit flexibility.
We already established you didn't know what you were talking about, why'd you have to go and prove it twice?
Re: (Score:2)
You spoke out of your ass, and demanded people take it for true.
Current division is day one in EE.
There's a reason you don't have a fucking triple-ought hunk of fucking copper on the bottom of your CPU.
Re: (Score:2)
Thus we recommend a 50% derating from the data sheets
The existing datasheet spec is already derated, firstly for full load on all pins, secondly the maximum current rating of the connector is specified at a temperature rise well short of the power rating.
Additionally worth noting is that for this 600W (derated by PCI-SIG from the already derated 660W from the manufacturer) people have been burning them using cards that ... only draw 350W from the connector as it is (75W comes from the slot).
Your 50% derating factor is in fact already implied in this case.
Also, I’ve never been a great fan of //ing lots of smaller current pins together. It has a bad engineering smell for many reasons.
Whet
Re: (Score:2)
Also, I’ve never been a great fan of //ing lots of smaller current pins together. It has a bad engineering smell for many reasons. Better to use single connector pins rated for the application, or increase your derating.
Would you be opposed to using multiple 4-pin format plugs, perhaps with slightly separated jacks on the card & different connector format (rounds & squares), like the plug used on the CPU power plug?
Somehow people are not able to flub up that one.
Yeah, you can't make a fancy looking card out of multiple 4-pin jacks, but heat dissapation-wise it might be a bit safer.
It is worse than I imagined (Score:5, Informative)
The cable should have never existed. Earlier, I read the issues were caused by half-inserted cables. Turns out there are more ways to get this wrong:
https://cablemod.com/12vhpwr/ [cablemod.com]
"Don't bend vertically"
"Don't bend horizontally"
"Have minimum 35mm distance from connector before a bend"
This is pretty much impossible given the gigantic sizes of 4090 (all of us have seen how an ITX motherboard mounts onto the card, rather than vice versa). And then there is also people's existing computer cases. Especially more so with the recent movement to smaller cases... (My NR200P will definitely require a very shard bend at the tip of the connector, if I can find a card that would fit in the first place).
This connector has the bare minimum design to carry 600W, and nothing more. In practice, the design has failed, and to be frank, should be scrapped altogether.
I am not an expert (Score:2)
...but from an outsiders view, that statement reads like "we will set the standards, but can't be arsed to test whether they are functional, safe, nor fit for purpose".
Then why have technical experts in such an organization? I mean, anyone can say "here's how the pins should be" and call it "a standard" if function and safety aren't part of the discussion.