Proposed new Power Connector

[qxp]

With all the attention on melting graphics card connectors, this might be an opportunity to switch manufacturers to a connector that we like. Here is my proposal - please comment whether you like the idea or not, any changes and if we get enough people behind it we'll try to make them listen.

The power connector and cable shall:

• have only two power delivery wires, ground and positive voltage
• the voltage should be increased to 48V (same as maximum for USB-C)
• the pins for power delivery should be designed to handle large current - paddle shaped or nub, according with the best state or the art.
• there should be four auxiliary wires.
• two of the auxiliary wires can be used for communicating between GPU (or other accessory) board and the power supply.
• two of the auxiliary wires should be tied on the GPU (or other accessory) board to power ground and power voltage. They will be used by the power supply to determine actual voltage drop, resistance and power dissipation of power delivery wires via Kelvin measurement. Excessive power dissipation in the cable should result in power supply disconnecting this power port.
• There should be a standard data connection from power supply to the motherboard to transfer real time data on current consumption and voltage. Ideally per port.

What do you think ?

 

[Macro Device]

have only two power delivery wires, ground and positive voltage

Simplicity is good.

the voltage should be increased to 48V (same as maximum for USB-C)

This will require a complete PSU overhaul. Every existing PSU will go obsolete. We are not ready for this.

the pins for power delivery should be designed to handle large current - paddle shaped or nub, according with the best state or the art.

Okay...

two of the auxiliary wires can be used for communicating between GPU (or other accessory) board and the power supply.

Not okay. All we need is two sensor pins that are designed the way it's impossible* to plug it incorrectly.

*for a person not actively trying to botch their setup

• They will be used by the power supply to determine actual voltage drop, resistance and power dissipation of power delivery wires via Kelvin measurement. Excessive power dissipation in the cable should result in power supply disconnecting this power port.
• There should be a standard data connection from power supply to the motherboard to transfer real time data on current consumption and voltage. Ideally per port.

"Smart" thing in electronics is only good on paper. In fact, anything might go wrong: incorrect data reading resulting in a completely fine connection going offline and vice-versa. This overcomplicated design isn't what we actually need.

Extending standard 8-pin to, say, 10 pins (by adding one +12V and one GND) and forcing it to be AWG14 is beyond enough. This way it can theoretically handle up to ~1 kW. Zero reason to be fancy about it.

 

[sbacc]

If PC users accept having to deal again with more cable in their pristine glass PC case, just using several EPS12v (8pins) or the ATX12v (4pin) or a combination of both would be the best solution.

 

[BoggledBeagle]

Reposting from the melting thread:

Looking at the 12 tiny pins, why do not we have just 6 massive tabs instead? The overall size of the connector would be the same.

There could be some cutouts on the plug connectors sleeves, and in the end of the plug travel, some protrusions in the socket would through these cutouts push on the metal reeds of these connectors, so that they would contact the tabs securely, even if they lost some of their original springiness.


Some locking mechanism, which would be usable only when the plug is fully inserted and which would actuate some contact enhancing mechanism surely would not be that hard to develop.


Adding to the points mentioned here:

48V to cut the current running in the cable is an interesting idea, but it would make sense only if the rest of the computer would be converted to run on 48V DC. An electronics engineer should comment on converting 12V to 1V vs 48V to 1V needed to run the chip.

 

[Solaris17]

What do you think ?

We already do this in the data center, only at 240, step down to 48. They are called bus bars.

Edit:: a fun aside, they also like occasionally melting. Sometimes spectacularly.

 

[maxfly]

Nvidia apparently had the solution ready to go but out thought themselves

NVIDIA GeForce RTX 4090/5090 Prototypes Tested with Quad 16-Pin 12V-2x6 Connectors

Recent engineering prototype leaks reveal NVIDIA's power delivery design for its RTX 5090 flagship GPU featured a jaw-dropping quad 12V-2x6 connector configuration. The prototype board, while not publicly photographed, is significantly different from the final version we have today. The...

www.techpowerup.com

 

[N/A]

I propose using 6AWG car battery jumper cables

 

[BoggledBeagle]

I propose using 6AWG car battery jumper cables

Why not, just fasten them in some massive screw terminals and the problem is solved... Almost, because people would fasten them the opposite way.

 

[qxp]

A few explanations in response to posting above:

• With regard to PSU changes, they were made anyway to support the melting connectors. Why not change again and make it right ?
• You would not need any drastic redesign for 48V because the chips drawing high currents do not run on 12V anyway. This is a holdover from old CPUs. So as long as you have VRMs, they could just as well convert from higher voltage. As mentioned above by @Solaris17 this is done in data centers anyway.
• The analog sensing wires are useful to make sure the high-power cable is operating correctly. They actually reduce complexity compared to what we have now - several shunt resistors on good GPUs each sensing current and voltage. Instead, power supply will monitor the cable and insure that it does not have too much voltage drop and power dissipation to cause problems. This would solve the case of damaged wires and bad pin contacts. The power supply would need the sensing circuitry once and can provide several such high-power connections for different cards.

 

[FoulOnWhite]

Have the cables fixed to the GPU with a cows tail, with a connector on the end. Less stress on the GPU connector(mechanically, not electrically)

 

[qxp]

With regard to 6AWG wires - wires this massive would not be necessary for two reasons: the increase in voltage to 48V reduces required current by a factor of 4. Secondly, having only two current carrying wires would allow to carry more current over thinner wires. If you lookup how much a wire of given gauge can carry on Google you will find that different web pages list different ampacities. The reason is that how much the wire can carry depends on how much it will heat up and how much its insulation can withstand.

How much the wire is cooling depends on difference between wire temperature and ambient. So teflon insulated wire can carry more current than the same gauge wire insulated with PVC because it will heat up more and will lose the heat effectively before melting.

Similarly, a wire in the case can carry more current than a wire in a house wall inside thermal insulation.

Take a look at this link and the difference between max amps for chassis wiring (i.e. inside case) and for power transmission (i.e. walls)

American Wire Gauge Chart and AWG Electrical Current Load Limits table with ampacities, wire sizes, skin depth frequencies and wire breaking strength

AWG Wire size chart and ampacity table for design engineers including skin depth frequencies and tensile strength data; electrical cable size

www.powerstream.com

So 14 gauge wire inside the case by itself - not in a bundle can carry 32A. We have two wires in a bundle (+ and ground), so lets cut this in half, we get 16A. At 48V this gives 768W of power.

Have the cables fixed to the GPU with a cows tail, with a connector on the end. Less stress on the GPU connector(mechanically, not electrically)

I am not sure I like this. If the cable or its connector gets worn out or damaged you would effectively lose an expensive card. And what happens when you need a longer cable for a larger case ?

 

[BoggledBeagle]

I think we should decide if this thread is about fixing the current problem and is aimed in the short future, or if it is a visionary piece about how things should look in 10 years.

In the first case the discussion about 48V makes no sense.

 

[DemonicRyzen666]

I'm just sitting here with two RTX 2080 ti's that can eat up to 585 watts.
All while using 32 pins to connect the power to them.
Aesthetics aside, it just works.

 

[qxp]

I think we should decide if this thread is about fixing the current problem and is aimed in the short future, of it is a visionary piece about how things should look in 10 years.

In the first case the discussion about 48V makes no sense.

The problem we have now is also an opportunity to get industry to change to a better connector. I think the reality is that current crop of 5090 cards require a dedicated PSU with 12-pin connectors that we now know melt because of design issue with the cards and connectors. So a fix would require a redesign of both PSUs and the cards anyway.

One could also consider allowing a range of voltages from 12V to 48V in a manner similar to USB-C. That's what the data lines are for.

 

[eidairaman1]

Interesting ideas but do some believe nv even pays attention to these ideas?

 

[FoulOnWhite]

I am not sure I like this. If the cable or its connector gets worn out or damaged you would effectively lose an expensive card. And what happens when you need a longer cable for a larger case ?

Instead of the connector from GPU to PSU being on the GPU it is on the end of the cows tail, so you just plug the 12pin cable from this to PSU, with much less stress on it. Also the connections on the GPU could be pretty strong, so not very likely to break.

 

[qxp]

Interesting ideas but do some believe nv even pays attention to these ideas?

If enough people support it, then sure. Also the connector is not just NVidia, but also Intel, power supply manufacturers, probably AMD.

Instead of the connector from GPU to PSU being on the GPU it is on the end of the cows tail, so you just plug the 12pin cable from this to PSU, with much less stress on it. Also the connections on the GPU could be pretty strong, so not very likely to break.

In order not to have bending issues the cable would have be clamped to the GPU, but then you could provide means of clamping for a cable with a connector as well.

But with a fixed cable, you would have to choose card based on cable length to fit into your full-size case vs small form factor case. And then if you close the door and accidentally pinch the cable there goes $2000 GPU. I also have not seen a lot of equipment having fixed cables to the motherboard. It might be expensive to do that as automated tools might not be available.