coercitiv
Diamond Member
- Jan 24, 2014
- 7,441
- 17,725
- 136
Just disassemble and bend the connector you're not supposed to disassemble and bend.
DrMrLordX
Lifer
- Apr 27, 2000
- 23,042
- 13,140
- 136
- Aug 22, 2001
- 32,128
- 32,706
- 146
Are there any non-problematic right angle connectors?
I'm using right angle connectors so this is relevant to my house's continued existence.
I'm using right angle connectors so this is relevant to my house's continued existence.
DrMrLordX
Lifer
- Apr 27, 2000
- 23,042
- 13,140
- 136
So long as the connector actually fits your card's power socket, it's not any more problematic than any other 12VHPWR/12v2x6 connector/cable.
Fallen Kell
Diamond Member
- Oct 9, 1999
- 6,230
- 543
- 126
If you have a 4080, 4090, 5080, or 5090 tier card, I would make sure you have separated the strands of the 12VHPWR/12v2x6 cable somwhere and used a clamp meter (voltimeter) to measure the power draw down each of the 6 power pins and ensure none of them are overloaded (i.e. 9amps +). Part of this problem isn't the cable itself, but there seem to be at least three variants of the pins being used, one of the variants is only rated for 8.5amps and seem to be the most widely used (another is 9.5amps, and the third known variant is 13amps). But you won't know what you have for the most part, so if you see 9 amps on any of the lines, it is possible you are overloading it which will slowly cause it to melt over time (and get higher and higher resistance as it melts, causing increased temps and thus increasing the pace it is melting things).
The real issue here is that for a full 600W card, it will take an almost perfectly balanced spread of load across the 6 cables to provide that 600W without exceeding the amp limit of 8.5 amps (12*6*8.5 = 612W total! There is literally only a 2% margin of safety on the cable/connectors with the 8.5amp rated pins, which means everything has to be absolutely perfect with the physical connections and cable construction). This whole situation has been one of the stupidest design failures in consumer electronics in recent decades. I don't know who in the right mind signed off on anything that has a 2% margin of safety. Power related items like this should have a minimum 20% margin of safety, if not even having 100% margin of safety. Your home's electrical wiring typically has a 80-100% margin of safety in the wires/connectors over the circuit breaker's rating for a 15amp or 20amp service loop.
Last edited:
DrMrLordX
Lifer
- Apr 27, 2000
- 23,042
- 13,140
- 136
That's a really good point, thanks for bringing that up. Would be nice to know which pin variant is used in any given cable.
- Aug 22, 2001
- 32,128
- 32,706
- 146
This is where things are. Can't just buy a card and use it anymore.
DrMrLordX
Lifer
- Apr 27, 2000
- 23,042
- 13,140
- 136
- Aug 22, 2001
- 32,128
- 32,706
- 146
I think it will be superior to the Ampinel. The wireview pro 2 isn't going to have load balancing. Roman explains why. However, if it fails, RMA and they will fix the card and send you a new WP2. If the card cannot be fixed they will send you a card with the same performance.
poke01
Diamond Member
- Mar 8, 2022
- 4,511
- 5,818
- 106
I'll be going with a GPU that has a 3x 8pin, if AMD offers that I'll be going with them next gen.
This connector is not good at all
DrMrLordX
Lifer
- Apr 27, 2000
- 23,042
- 13,140
- 136
Yeah, AMD cards aren't gonna be much better using this thing. Best to avoid any cards using 12v2x6/12VHPWR.
I guess the 9070 XT is far less likely to get anywhere near to drawing 600W though - but then why did Sapphire decide to try it? Makes zero sense!
Still this whole thing is a problem waiting for a solution... or is that the wrong way around?
"A solution waiting for a problem"?
"A solution bound to cause a problem"?
In any case just avoid it altogether - it is just Nvidia are being ultra stuborn on the designs they allow so anyone wanting x90 performance has little choice than to play the lottery!
coercitiv
Diamond Member
- Jan 24, 2014
- 7,441
- 17,725
- 136
New PSUs come with HPWR cables and the build can look nicer, especially with the way Sapphire placed the connector on the card.
So they did it because they thought it would sell better.
DrMrLordX
Lifer
- Apr 27, 2000
- 23,042
- 13,140
- 136
coercitiv
Diamond Member
- Jan 24, 2014
- 7,441
- 17,725
- 136
DrMrLordX
Lifer
- Apr 27, 2000
- 23,042
- 13,140
- 136
It saves in PCB space, which is pennies or dollars per card. Yes they really do optimize around stuff like that. Plus NV in particular seems to have cut corners on power delivery components on everything after the 3090Ti.For what, a fire extinguisher?
Fallen Kell
Diamond Member
- Oct 9, 1999
- 6,230
- 543
- 126
After thinking this through over and over again, there is no good solution to fix this with a 12V system. The GPUs have simply exceeded the capabilities of a 12V power delivery solution and need to look at 24V or realistically a 48V solution. This also means having entirely new power supplies and no middle solution where-in you can somehow use the older PSUs with the newer cables/GPUs, and a new PCIE socket/motherboard.
The amperage down the cable and connectors is what defines the thickness and heat dissipation requirements. By changing to a higher voltage level system you will cut down on the amperage needed for the similar overall wattage. This is much like we are seeing right now with LED string lights. Most of these were originally 5V, then 12V, 24V, and now even 48V due to the overall length of the string and/or the number of physical LEDs in the string getting denser and denser. The power cables were simply needed to get too thick to continue using the same voltage, so they simply increased the voltage so they could maintain the same cables.
We have reached that issue now in GPUs. To have cables thick enough to handle the amperage, while also being supple enough to route within a PC case, and have safety margins so that they do not melt, we have to go to a higher voltage system. With a 48V system, you can use a single pair of wires 14GA wires and deliver 720W on a wire that will stay under 60C/140F. The single pair of wires is obviously the ideal solution, since it removes any issues of load balancing. You will still want monitoring/sensing because even with a single pair you can have poor cable construction or connection, however it is exponentially easier to have protection circuits on a single pair with a fuse or inline breaker.
The amperage down the cable and connectors is what defines the thickness and heat dissipation requirements. By changing to a higher voltage level system you will cut down on the amperage needed for the similar overall wattage. This is much like we are seeing right now with LED string lights. Most of these were originally 5V, then 12V, 24V, and now even 48V due to the overall length of the string and/or the number of physical LEDs in the string getting denser and denser. The power cables were simply needed to get too thick to continue using the same voltage, so they simply increased the voltage so they could maintain the same cables.
We have reached that issue now in GPUs. To have cables thick enough to handle the amperage, while also being supple enough to route within a PC case, and have safety margins so that they do not melt, we have to go to a higher voltage system. With a 48V system, you can use a single pair of wires 14GA wires and deliver 720W on a wire that will stay under 60C/140F. The single pair of wires is obviously the ideal solution, since it removes any issues of load balancing. You will still want monitoring/sensing because even with a single pair you can have poor cable construction or connection, however it is exponentially easier to have protection circuits on a single pair with a fuse or inline breaker.
Heartbreaker
Diamond Member
- Apr 3, 2006
- 5,226
- 6,844
- 136
Cable isn't the problem it can handle the specified load. It's the connector creating large imbalances.
Switching to 48V isn't practical. Even datacenter equipment with 48V to the cabinet using a PSU to step down to 12V for boards.
Having 48V to the board means including another PSU stage on each board to step down to 12V for standard components, or a more esoteric (expensive) 48V to 1V converter on the board, and more heat dissipated on the boards for that.
Fallen Kell
Diamond Member
- Oct 9, 1999
- 6,230
- 543
- 126
But the cable still is a problem. Once you go to multiple cables, load balancing across the different cables is surprisingly hard to do while being completely fail-safe. Cable defects are more common than you think and even minor defects in a multi-cable system that is already pushing those cables close to their design limit becomes a problem. The safety overhead on an individual cable is too small in the design, when the system of cables itself (in this case 6) when just one failed cable overloads all the others. They can't spread that additional 8.5 amps of load across the remaining 5 cables without overloading them, and that assumes that they had proper load balancing design, which they do not. Right now, the power simply goes down the lowest resistance paths, and in stranded cables, you easily will have some cables that may have one or more individual strands that might have a break in them, but the cable will pass continuity testing just fine. It won't show up without a resistance test, and even then it might not show up on the initial resistance test, but only after the cable was bent and pushed around in the case with the ends of the cable possibly straining due to a 90 degree bend, pushing some of the strands while pulling on other strands of the stranded cable. This results in cables of difference resistances, which drives more amperage down the cables with the least resistance, and exceeding the limits of the connectors and/or the cable itself.
A single cable pair or a series of cable pairs which has plenty of overhead is what is needed, not something that is operating with just a few percent of safety margin like we currently have... And yes, it would be expensive, and would require a lot of engineering work to change things from 12V to 24V or 48V. If they want to stick with the same motherboards and socket, it would require a 48V to 12V converter on the cards. With a new motherboard and card socket, you wouldn't necessarily need to convert down to 12V but the GPU and memory chips would need to be redesigned to work at 48V and all the components changed out to ones compatible for 48V operation, but these are mostly larger physical devices, so it may not be possible to use similar board designs even when scaling up to fit the larger components, and even then some components like the memory chips or GPU chip would either still need a step down circuit to obtain proper power voltage levels or need a completely new design.
I didn't say this would be easy, but what I said is that it is what is needed to continue down this path of higher wattage GPU cards and maintain safety without a high risk of fire or melting/burning of components.
EDIT: Also to top all this off, I am surprised that home insurance companies have not flagged the current designs yet as something that will void their fire insurance coverage at this point with so many examples of melting cables and connectors.
Last edited:
bro its 2025 and nothing ever happens anymore, its a pirates world and laws basically don't exist unless you step on the wrong toes
from an engineering standpoint yes but all of the failures we've been seeing are due to terminals not having enough contact, and being so small that their thermal capacity is IMO too low. The cables themselves are fine for the power delivery, but the shrinking of the connector has clearly lead to knock on thermal effects.
You do got me thinking though, any card with a faulty connector/bad connection should be pulling more power from the PSU than it actually needs, thats the only way the connector can get hot after all. If smart PSUs were a thing and if they could monitor power draw over a rail suddenly going up 200w higher than it used to, that would stop this from happening. Makes me wish more people did power measurements from the wall instead of relying on software.
coercitiv
Diamond Member
- Jan 24, 2014
- 7,441
- 17,725
- 136
Dance, NITRO, dance or feel the heat!
Black screen of connector death.
Black screen of connector death.
Last edited:
Heartbreaker
Diamond Member
- Apr 3, 2006
- 5,226
- 6,844
- 136
Heartbreaker
Diamond Member
- Apr 3, 2006
- 5,226
- 6,844
- 136
Maybe we need a variation on this:
That's the 500 Amp version which is probably too much overkill. But you could probably scale that appropriately for 100 Amps, then use it for 60 Amps max.
That's the 500 Amp version which is probably too much overkill. But you could probably scale that appropriately for 100 Amps, then use it for 60 Amps max.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter