goodbye, 3x8; new PCI-e power standard delivers 600 watts over a single connector

[BachChain]

 

Summary

With the rise in recent years of graphics cards that have max power draws in the hundreds of watts, a problem has arisen on how exactly to get all that electricity from the power supply to the card. Historically, one would use the classic 6 or 8 pin connectors, but with each being limited to 150 watts, it's necessary to use 2 or 3 or more. Nvidia tried to solve this in the 3000 series with their special 12 pin connector, but it's a closed standard that has seen very little adoption outside of their own founder editions.

The PCI-SIG has recognized the issue and designed a new power connector as part as the new PCI-e 5 standard. The 12VHPWR "High Power Connector" has 12 pins that can deliver up to 55 amps at 12 volts for a maximum of 600 watts of continuous power on a single connector.

 

Quotes

Quote

 The PCI SIG specifies a pin current capability (excluding sideband contacts) of 9.2 A per pin/position with a limit of 30 °C T-rise above ambient temperature at + 12 VDC with all twelve contacts energized. This results in 55.2 amps in one direction for the 12 volt power rail, or an even 662.4 watts. Of course, tolerances and safety deductions are still taken into account here in the total of just over 11 percent, so that in the end 600 watts can still be safely guaranteed

 

Quote

NVIDIA is the first to move forward here and will probably also push through the specifications for PCIe 5.0. However, it is still unclear whether the rumored RTX 3090 Ti will also support PCIe 5.0 on the data connector.

 

My thoughts

 This was long overdue. It's a good thing an actual standard was created before something like Nvidia's proprietary connectors saw widespread

 

Sources

 https://www.igorslab.de/en/up-to-600-watt-power-for-graphics-cards-with-the-new-pcie-5-0-adapter-nvidias-rtx-3090-ti-makes-the-start-exclusive/

[da na]

How long until single SATA power to one of these adapters start showing up on Wish huh? 

[Estiar]

Just now, Mel0nMan said:

How long until single SATA power to one of these adapters start showing up on Wish huh? 

Sounds like a recipe for a melted connector

[Zodiark1593]

4 minutes ago, estiar said:

Sounds like a recipe for a melted connector

Molex to 12 Pin anyone?

[Levent]

I am more curious about the 4 pin, what is that? Some sort of serial communication port?

[da na]

6 minutes ago, Levent said:

I am more curious about the 4 pin, what is that? Some sort of serial communication port?

Looks Molex-sized

[NumLock21]

that looks a lot like the connector on them new Nvidia reference cards.

[Lightwreather]

This is great and all, but it does concern me a little. Are they really expecting GPUs or some other PCIe device to consume upto 600w even in beyond the enar future?

[Mihle]

27 minutes ago, J-from-Nucleon said:

This is great and all, but it does concern me a little. Are they really expecting GPUs or some other PCIe device to consume upto 600w even in beyond the enar future?

Could possibly b used for servers?

 

It's better to have it capable of taking more than needed than finding out it's not enough later.

[Arika]

I read the title and got excited thinking it was going to be 600w over the PCI-e slot, which would remove the need for cables completely.

[Mihle]

3 minutes ago, Arika S said:

I read the title and got excited thinking it was going to be 600w over the PCI-e slot, which would remove the need for cables completely.

That would have made motherboards much thicker if it's even possible.

 

Would have been cool tho.

[leadeater]

2 hours ago, Levent said:

I am more curious about the 4 pin, what is that? Some sort of serial communication port?

 

Quote

and the four smaller contacts below are for the sideband signals

Please don't ask me what the hell that means lol

 

Edit:

Still not sure how it relates to the PSU though...

 

Quote

The Transaction layer also includes a Message Space, which PCI-E uses to handle all the sideband signals of the PCI bus. Sideband signals include interrupts, power-management requests, and reset commands. The Message space, in essence, provides “virtual wires” to replace these signals.

https://www.embedded.com/getting-aboard-the-pci-express/

 

http://openpowerfoundation.org/wp-content/uploads/resources/25Gbps-spec/content/sec_sideband.html

 

[porina]

4 hours ago, J-from-Nucleon said:

This is great and all, but it does concern me a little. Are they really expecting GPUs or some other PCIe device to consume upto 600w even in beyond the enar future?

It will be the maximum, and you don't have to run it on the maximum. A bit of headroom is nice.

 

Looking at what we have today, 3090 is rated at 350W. I don't know how much it takes with a mild overclock but if we assume 400W we're already 2/3 of the way into the new connector. Extreme overclockers probably will still do hardware modding as 600W wont be nearly enough.

[Dracarris]

4 hours ago, J-from-Nucleon said:

This is great and all, but it does concern me a little. Are they really expecting GPUs or some other PCIe device to consume upto 600w even in beyond the enar future?

Better make it future proof. Today we might only need 300-400W but it's still a major advantage to have only a single connector.

3 hours ago, leadeater said:

Please don't ask me what the hell that means lol

Sideband signals is a general concept which often gets used to handshake or signal supported features which then influences what happens on the "main" channel. USB-C has those side channel signal which are e.g. used to exchange info about supported transfer speeds and, crucially, what type of PD a device supports. Classic Louis Rossmann e.g., "It starts with 5V but then the Macbook says to the charger, hey I am a Macbook I am not a phone I don't want any of that 5V rubbish, give me 20V". (Yes if sth goes wrong there in the wrong direction you can fry your device).

 

Given that we're talking about 0.6kW here I guess the PSU will initially limit the power draw from this connector until the GPU tells it that it actually needs that much power. Otherwise the PSU will probably disable power on that connector for protection. And hopefully that feature also makes those idiotic SATA-to-this adapters much harder if not impossible to make.

[leadeater]

21 minutes ago, Dracarris said:

Sideband signals is a general concept which often gets used to handshake or signal supported features which then influences what happens on the "main" channel. USB-C has those side channel signal which are e.g. used to exchange info about supported transfer speeds and, crucially, what type of PD a device supports. Classic Louis Rossmann e.g., "It starts with 5V but then the Macbook says to the charger, hey I am a Macbook I am not a phone I don't want any of that 5V rubbish, give me 20V". (Yes if sth goes wrong there in the wrong direction you can fry your device).

 

Given that we're talking about 0.6kW here I guess the PSU will initially limit the power draw from this connector until the GPU tells it that it actually needs that much power. Otherwise the PSU will probably disable power on that connector for protection. And hopefully that feature also makes those idiotic SATA-to-this adapters much harder if not impossible to make.

The problem is more PSU support, sure this is a PCIe spec but currently "desktop" PSU's don't have any of these sideband communications at all so where does that 4 PIN sideband connections actually go? To new PSUs, to a connector on the motherboard?

 

I've seen sideband functions on dual PSU server configurations from HPE, this is used along with iLO to configure power profiles as well as power capping if you want to cap the peak power of the server safely without causing an OS crash (as it'll change the CPU turbo settings and voltages as required).

 

 

I have never seen this kind of functionality on a regular ATX PSU, it's also not a feature on non-redundant single PSU server configurations either. I also suspect these are different sideband configurations and usages as well.

 

If this is really going to require PSUs with logic in it to support this then we're talking way more than a mere PCIe spec update and a new connector that can carry more current. It means to total and complete cut of old PSU support, no PCIe 8 pin to PCIe 5.0 12 pin cable converters even if it were safe to do. This sounds like a hard cutover with zero backward compatibility options.

 

Also now we're going to have to worry about PSU PCIe 5 spec implementation quality and reliability as well. Sounds like cheap PSUs are going to be even more recopies for disaster.

Edited October 11, 2021 by leadeater

[Dracarris]

10 minutes ago, leadeater said:

The problem is more PSU support, sure this is a PCIe spec but currently "desktop" PSU's don't have any of these sideband communications at all so where does that 4 PIN sideband connections actually go? To new PSUs, to a connector on the motherboard?

You're right. My guess was it connects to the PSU but that would require to buy new. It could be specified in a way that older PSUs can simply supply all the juice without any communication taking place. I mean that's anyway the only way if you don't buy a new PSU, strapping together multiple 6- or 8-pin connectors that are always supplied.

[leadeater]

1 minute ago, Dracarris said:

I mean that's anyway the only way if you don't buy a new PSU, strapping together multiple 6- or 8-pin connectors that are always supplied.

Strapping together 4 PCIe 8 pins isn't exactly the smartest idea anyway lol.

 

I sense much pain for AIBs, now they are going to have to think about two variants of cards for who knows how long, legacy PCIe power connector and PCIe 5 power connector models.

 

RIP if you buy the wrong one.

[Doobeedoo]

Well interesting to see new PCI-E power connectors as well as new mobo ones for new platforms. Definitely great to see less cable clutter.

[Kisai]

7 hours ago, Levent said:

I am more curious about the 4 pin, what is that? Some sort of serial communication port?

Honestly, this looks like two 2x6 connectors married to a 4-pin molex. Probably a backwards compatibility consideration IMO.

https://www.amphenol-icc.com/product-series/minitek-pwr-cem-5-pcie.html#

[Mihle]

15 minutes ago, Kisai said:

Honestly, this looks like two 2x6 connectors married to a 4-pin molex. Probably a backwards compatibility consideration IMO.

https://www.amphenol-icc.com/product-series/minitek-pwr-cem-5-pcie.html#

Article says it's 3mm instead of 4,2mm pin spacing, so no, it's not backwards compatible the way you are thinking of. (The whole plug is shorter than what 2x6 connectors would be)

The pin shapes is also not the same shape pattern I think but that wouldn't matter because different size.

 

Can also take more wattage than what 2x6 or 2x8 can take.

[Moonzy]

Can they not name it pcie 5? It's gonna be confusing when people say pcie 5.0, is it the power connector or the lanes???

 

Bruh where do they even pull that name out of?

[Dracarris]

37 minutes ago, Kisai said:

Honestly, this looks like two 2x6 connectors married to a 4-pin molex.

Ignoring that molex is a company, I guess you refer to the old connectors which powered HDD drives before SATA was introduced? Those are definitely very different from the 4-pin part of this new connector (and hopefully finally go to way of the dodo).

 

A ton of the power connectors found in PCs, what we refer to as "Molex", is just a tiny part of their product portfolio called Mini-Fit.

 

The 4 pins below are clearly designated as "signal" lines. Other than that there is absolutely zero info out what they are supposed to carry, maybe the PCIe spec states something but I don't have access (yet another spec that is locked behind a "partners-only" wall, smh).

[leadeater]

26 minutes ago, Moonzy said:

Can they not name it pcie 5? It's gonna be confusing when people say pcie 5.0, is it the power connector or the lanes???

 

Bruh where do they even pull that name out of?

Well it's actually called 12VHPWR, support for it has been added in to the PCIe 5.0 spec. As long as we don't common name it as 'PCIe 5 Power Connector' and just use say '12V High Power Connector' then it should be fine, "we" of course won't heh.

 

Or just short name it to 12VHP, or really the most obvious '12Pin Power' in keeping with what we use now anyway.

[Distinctly Average]

8 hours ago, BachChain said:

 

Summary

With the rise in recent years of graphics cards that have max power draws in the hundreds of watts, a problem has arisen on how exactly to get all that electricity from the power supply to the card. Historically, one would use the classic 6 or 8 pin connectors, but with each being limited to 150 watts, it's necessary to use 2 or 3 or more. Nvidia tried to solve this in the 3000 series with their special 12 pin connector, but it's a closed standard that has seen very little adoption outside of their own founder editions.

The PCI-SIG has recognized the issue and designed a new power connector as part as the new PCI-e 5 standard. The 12VHPWR "High Power Connector" has 12 pins that can deliver up to 55 amps at 12 volts for a maximum of 600 watts of continuous power on a single connector.

 

Quotes

 

 

My thoughts

 This was long overdue. It's a good thing an actual standard was created before something like Nvidia's proprietary connectors saw widespread

 

Sources

 https://www.igorslab.de/en/up-to-600-watt-power-for-graphics-cards-with-the-new-pcie-5-0-adapter-nvidias-rtx-3090-ti-makes-the-start-exclusive/

Personally I am not sure if we are moving the right way here. We are in a world where we are struggling to keep the lights on, yet we are demanding an exponential growth in electricity. The resulting carbon footprint is huge. We have to begin thinking about this and not just giving PC manufacturers an easy way to use more power instead of optimising. It can be done I am sure, many laptops, Apple M1, and more have shown that we can have performance without huge current drain. It needs to be taken a step further. Just a couple of years back people were moaning here in Europe about vaccums being limited by law to 800w (or something around that mark). People were upset that their carpets would no longer be as dust free as they would have been with their 2.2Kw model. The manufacturers innovated and have models now that are more powerful but use less electricity. It sometimes takes regulation to force change and I feel the PC market is heading that way rapidly. We have already seen some US states legislating and I think it is only a matter of time before more do as well as the EU etc.

[Moonzy]

16 minutes ago, Distinctly Average said:

Personally I am not sure if we are moving the right way here. We are in a world where we are struggling to keep the lights on, yet we are demanding an exponential growth in electricity. The resulting carbon footprint is huge. We have to begin thinking about this and not just giving PC manufacturers an easy way to use more power instead of optimising. It can be done I am sure, many laptops, Apple M1, and more have shown that we can have performance without huge current drain. It needs to be taken a step further. Just a couple of years back people were moaning here in Europe about vaccums being limited by law to 800w (or something around that mark). People were upset that their carpets would no longer be as dust free as they would have been with their 2.2Kw model. The manufacturers innovated and have models now that are more powerful but use less electricity. It sometimes takes regulation to force change and I feel the PC market is heading that way rapidly. We have already seen some US states legislating and I think it is only a matter of time before more do as well as the EU etc.

no, my game needs that 10% more fps, so i can have intercourse with my opponent's mother