The future of power supplies and GPUs is here (ATX 3.0 spec)

[Dogzilla07]

https://www.tomshardware.com/news/intel-atx-v3-psu-standard

 

Aris released the article detailing the future of power delivery and the new ways graphics and psus will communicate to enable extreme power consumptions (amongst many power-saving, and other features). The thing that stands the most to me is that his Cybenetics Certification project is now considered by intel to be of the same importance as the 80 plus certification body, and is in official Intel guidelines for the whole industry as an official standard. Additionally a lot of cool stuff (probable changes to every platform released so far so they can work with the most effectively with upcoming GPUs, requirement for the new PCIe gen5 cable on every PSU >450w, and many more. Like "PSUs should be able to turn on/off 175,200 times per year of their life without breaking!"

 

It's an amazing read.

 

There's a few things that look too artificial, but there could very well be reasons for that not immediately apparent. (Like assuming tertiary PC components need 300w, and possible hard locking that (1. GPU = 600w, 2. CPU = 300w 3. Rest = 300w). The only thing that comes to mind is 20,30,40W NVMe Gen5, Gen6, SSDs in the future, and bigger TDP DDR5/DDR6 RAM, and higher powered ARGB and so on. I can definitely see these things creeping up above 150w circa 2023-2027, so just overprovisioned future proofing.).

 

The second thing (and I'm biased a bit here), is locking GPUs above 450w to liquid cooling only (on one hand it's guideline, but there's also a lot of sensor lock-ing, knowing Asrock, them going MAD on a Air cooled design is probably already in the cards xD though). I get the point, but a revised 4.5 slot Asus Noctua 3070 can surely handle 500w, or more ? Though again liquid cooling helps immensely with VRAM and so on, so maybe Nvidia already set a precedent with chips on the back of a card, and that gets a bit tough even with 5 slot full sized 120x25mm fans. Either I can't wait to see what designs we get, I'm literally jumping from joy from all the potential and the new and awesome things these changes will bring xD

[jaslion]

This seems like something that just like atx 12V will be mostly ignored so as to not limit manufacturers too much.

[porina]

If I'm reading it right, 12VHPWR is NOT the connector nvidia came up with for 30 series? I wonder if there is any degree of compatibility between them. The problem with nvidia connector was you could end up in situations where the connector supported more than PSU can take. The signalling here would prevent that scenario.

 

That new connector is probably the main stumbling block for self builders who might piecewise upgrade systems, but if we keep in mind the much wider PC market it'll make little to no negative impact to new builds. Presumably PSUs would still include some legacy 6+2 connectors as we transition.

 

On the 300W for rest of system number, keep in mind it is a maximum and it doesn't follow that many if any people will be anywhere near that. This also applies to the 600W limit for 12VHPWR connector. There are use cases outside of gaming builds that might go that direction.

 

The allowance of power spikes, I wonder if this is related to nvidia 30 series? Cases were reported early on of PSUs tripping due to power spikes even though the average power should be fine.

[mariushm]

I'm not a fan of the 12v high power connector they decided to use .. it will be a pain in the ass for sleeving, using ribbon cables with different diameters (AWG16 for power, AWG28 for the 4 data pins)

Plus different crimping tools, most likely limited sources for these connectors... 

 

Also, the 4 data pins are stupidly used .. 2 wires are used just for the video card to indicate how much it consumes (see on the right).  Let's waste two wires for 4 possible values. Could have used one wire and a couple resistors as a voltage divider on the video card to output a voltage and use ADC in the psu to read the voltage (0..2v - 0 , 3v..5v - 1 , 6v..8v = 2 , 9v..11v = 3)

The other two wires are also poorly used, just off/on switches. 

FFS they could have used I2C or SPI, just two wires, and put the two wires into a AWG16 (coaxial cable like) diameter cable to look the same as the other AWG16 wires... you can have a 10-20 cent microcontroller or chip on the video card respond to some i2c/spi commands from the power supply or push a notification to the power supply and replicate the whole functionality of these 4 wires just as safe.. 

 

Honestly I hope it doesn't go anywhere. 

 

Still feel like they should have added 20v to the ATX standard, which would allow to have USB ports in computers that can switch between 5v/12v/20v and potentially power monitors or other devices. And also, for laptops and ITX machines, would allow to power directly from a 20v +/- 5% laptop adapter brick (they make them for 18v or 19.5v, they could make them 20v easily)

And 20v is still low enough that you could make 20v -> 1v dc-dc conversion reasonably efficient.

[Dogzilla07]

1 hour ago, jaslion said:

This seems like something that just like atx 12V will be mostly ignored so as to not limit manufacturers too much.

My understanding is this was done together with the manufacturers (both PSU and GPU) and has a lot of their input, and many compromises made (because the GPUs will require a lot of these changes one way or another, sooner or late). Asus Loki, Cooler Master, Gigabyte, etc, ... all have 12+4 pin PSUs coming out with integrated connector. Not at all like the previous 12v situation.

 

1 hour ago, porina said:

If I'm reading it right, 12VHPWR is NOT the connector nvidia came up with for 30 series? I wonder if there is any degree of compatibility between them.

It is the exact same connector. just missing the 4 sense pins to enable 600w mode. cross-compatible and up to 450w delievery (in theory). But some limitations in the article above so we'll see.

1 hour ago, porina said:

On the 300W for rest of system number, keep in mind it is a maximum and it doesn't follow that many if any people will be anywhere near that. This also applies to the 600W limit for 12VHPWR connector. There are use cases outside of gaming builds that might go that direction.

Not exactly, PSU will get smarter now, and control this together with the GPU, unless explicitly disabled, there's gonna automatic limitations done by PSU and GPU. As for spiking exactly, there's rumors of 1600w spiking next-gen.

[Dogzilla07]

 

[Juular]

5 hours ago, Dogzilla07 said:

My understanding is this was done together with the manufacturers (both PSU and GPU) and has a lot of their input, and many compromises made (because the GPUs will require a lot of these changes one way or another, sooner or late).

PSU manufacturers were nowhere near this, there are no PSU manufacturers in the PCI-SIG, they were given the spec post factum, GPU designers - sure, both AMD and nVidia are there, but they're not really GPU manufacturers either, as in the whole product, for customer market at least. Regardless, i don't really see the use of this 4-pin appendix in the way it's implemented either, if you're going out your way to make a specification for something new then a true digital interface would've made way more sense IMO too, it's not like we're living in 1950's, it can be made reliably and server PSU already have it.

5 hours ago, Dogzilla07 said:

Asus Loki, Cooler Master, Gigabyte, etc, ... all have 12+4 pin PSUs coming out with integrated connector. Not at all like the previous 12v situation.

Yeah, thing is, you don't need this connector PSU-side. A cable with 2/3x8-pin on one side and 12VHPWR on another will do just fine. Neither you need the 4-pin part because it's a simple short to ground.

5 hours ago, Dogzilla07 said:

Not exactly, PSU will get smarter now, and control this together with the GPU

The PSU would've had control over this if there would've been an interface, there isn't. It's a passive in-cable physical config merely telling that the GPU is 'good to go'.

[Dogzilla07]

29 minutes ago, Juular said:

Yeah, thing is, you don't need this connector PSU-side. A cable with 2/3x8-pin on one side and 12VHPWR on another will do just fine. Neither you need the 4-pin part because it's a simple short to ground.

for 450w yeah, but for 600w ?. Wouldn't the GPUs limit themselves (unless tricked of course). Someone also mentioned on the comments in the video that In the case of 600w adapters and I quote "you want an actual module with additional components. For modern PSUs at least a few extra capacitors are required, maybe coils too, if you want true 600w support on the 12vhpwr output connector". To me it sounds that leads to extremely janky cables, even more than the in-cable capacitor ones we previously had.

 

Aris also mentioned all (or almost all) current PSUs would probably have problems with the 10ms spike requirements, and some possibly even the 1ms.

29 minutes ago, Juular said:

PSU manufacturers were nowhere near this, there are no PSU manufacturers in the PCI-SIG, they were given the spec post factum, GPU designers - sure, both AMD and nVidia are there, but they're not really GPU manufacturers either, as in the whole product,

That's really unfortunate, I guess i got the wrong impression.

[Guest]

28 minutes ago, Dogzilla07 said:

for 450w yeah, but for 600w ?. Wouldn't the GPUs limit themselves (unless tricked of course).

That's what the sense wires are for.  Read the article.  The table is in there.  When you ground certain pins in a certain combination on the +4, it "tells" the GPU that the PSU is capable of supporting XXX wattage.

 

[Juular]

30 minutes ago, Dogzilla07 said:

you want an actual module with additional components. For modern PSUs at least a few extra capacitors are required, maybe coils too, if you want true 600w support on the 12vhpwr output connector". To me it sounds that leads to extremely janky cables, even more than the in-cable capacitor ones we previously had.

For higher transient capability the GPU that would actually need 600W - yes, probably, since 'conventional' pre-ATX3.0 PSUs don't have nearly enough capacitance to deal with transient like that with the probable exception of some 1.3-1.6kW designs. But it's too early to say anything concrete, we'll see when those GPU actually come.

[Vishera]

Molex to PCI-E adapters all over again...

[Guest]

6 minutes ago, Vishera said:

Molex to PCI-E adapters all over again...

Let it happen.  Let there be fires.  And then maybe they'll start teaching basic EE including Ohm's law in high school again. 

 

[Dogzilla07]

1 hour ago, jonnyGURU said:

That's what the sense wires are for.  Read the article.  The table is in there.  When you ground certain pins in a certain combination on the +4, it "tells" the GPU that the PSU is capable of supporting XXX wattage.

No, I got that, I meant when using an adapter from 3x8pins (or 4) to one 12-pin, and intending to force 600w. In the context @Juularanswered.

 

1. with no sense pins

2. with faked sense pins on one end to trick the gpu (but nothing else on the cables/adapter)

3. 4x8pins adapter with faked sense pins but additional parts on the cables like the youtube comment I quoted (extra capacitors, coils to make it as close to the real thing, etc)

[Guest]

55 minutes ago, Dogzilla07 said:

No, I got that, I meant when using an adapter from 3x8pins (or 4) to one 12-pin, and intending to force 600w. In the context @Juularanswered.

 

1. with no sense pins

2. with faked sense pins on one end to trick the gpu (but nothing else on the cables/adapter)

3. 4x8pins adapter with faked sense pins but additional parts on the cables like the youtube comment I quoted (extra capacitors, coils to make it as close to the real thing, etc)

There is no "faked" sense pins.  I think that's what's killing me here.  There is no signal.  It's just a ground.  The graphics card is doing the sensing.  It senses that a pin is ground and it knows the cable supports X wattage.  Those PSUs that you see renders of with 12+4-pin connectors on them do nothing special.  They simply terminate ground wires that the graphics card "senses".

 

[Dogzilla07]

Not even next-gen, 3090 Ti coming with the 600w connector:

 

https://videocardz.com/newz/galax-geforce-rtx-3090-ti-boomstar-pictured-with-pcie-gen5-power-connector

 

 

https://rog.asus.com/articles/psus/power-your-high-octane-small-form-factor-pc-with-rogs-new-loki-psus/

 

https://videocardz.com/newz/full-spec-pcie-gen5-124pin-msi-power-supplies-for-next-gen-gpus-have-been-pictured

 

And the Gigabyte and the Cooler Masters on the (EU's version of Pcpartpicker) below:

https://geizhals.eu/?cmp=2685195&cmp=2679049&cmp=2677802&cmp=2677804&active=0

 

However it pans out, at least there's only a single stock cable now to make cable management easier. On another side, a person on a different forum mentioned the new Gen5 PCIe connector looks quite harder to custom-sleeve (without different/new tools ?). On a third side, another person suggested it might only be tougher to pull out/push-in pins, and that the 4 small 28AWG cables could easily be encased to be similary sized to 16/18AWG.

[Guest]

4 hours ago, Dogzilla07 said:

Not even next-gen, 3090 Ti coming with the 600w connector:

 

https://videocardz.com/newz/galax-geforce-rtx-3090-ti-boomstar-pictured-with-pcie-gen5-power-connector

 

 

Yeah.  You use the standard FE 12-pin on that card.  They just used the 12+4-pin so in a year or two if/when PSUs come with the cable and someone buys this card and a new PSU, they would already have a "native" cable.  I guess you can say they're being "forward compatible".

 

4 hours ago, Dogzilla07 said:

 

 

And the Gigabyte and the Cooler Masters on the (EU's version of Pcpartpicker) below:

https://geizhals.eu/?cmp=2685195&cmp=2679049&cmp=2677802&cmp=2677804&active=0

 

However it pans out, at least there's only a single stock cable now to make cable management easier. On another side, a person on a different forum mentioned the new Gen5 PCIe connector looks quite harder to custom-sleeve (without different/new tools ?). On a third side, another person suggested it might only be tougher to pull out/push-in pins, and that the 4 small 28AWG cables could easily be encased to be similary sized to 16/18AWG.

Those 12+4-pin connectors on the PSU make me laugh.  It's like, as soon as they saw the new connector, they freaked out and put one on the PSU itself just make sure their butts are covered.

 

Lotes and Astron are selling cables only, not the parts (except for some rare low volume exceptions) that have the 12+4-pin on either side and they're stupid expensive.  Guess who I'm not planning on using as a connector supplier? 

[AnonymousGuy]

On 3/4/2022 at 1:04 PM, jonnyGURU said:

There is no "faked" sense pins.  I think that's what's killing me here.  There is no signal.  It's just a ground.  The graphics card is doing the sensing.  It senses that a pin is ground and it knows the cable supports X wattage.  Those PSUs that you see renders of with 12+4-pin connectors on them do nothing special.  They simply terminate ground wires that the graphics card "senses".

 

I presume I'm just going to be able to buy a little ribbon cable that grounds the sense pins off the GPU then so I'll get the full 600W with 8 pin cables + adapters. 

 

I want to buy the Asus Thor 1600W (cause the AX1500i has a stupid fan profile that is noisy all the time)  but I don't think it supports 600W and it's not even been released yet. 

[Guest]

52 minutes ago, AnonymousGuy said:

I want to buy the Asus Thor 1600W (cause the AX1500i has a stupid fan profile that is noisy all the time)  but I don't think it supports 600W and it's not even been released yet. 

I think you mean AX1600i because AX1500i has been discontinued since 08/2019.

 

Where did you see this "stupid fan profile" comparison with the Asus Thor?  Because I don't believe the Thor is any better.  In fact, tree canopy confirms likely worse.

 

Tom's review calls AX1600i "silent":  https://www.tomshardware.com/reviews/corsair-ax1600i-psu,5406.html

 

Fan graphed:  https://www.tomshardware.com/reviews/corsair-ax1600i-psu,5406-5.html

 

Fan noise versus Wentai 1616W, Kolink 1500W and Enermax 1800W:  https://www.techpowerup.com/review/wentai-aidan-t1616-1616-w/6.html

 

 

 

[AnonymousGuy]

28 minutes ago, jonnyGURU said:

I think you mean AX1600i because AX1500i has been discontinued since 08/2019.

 

Where did you see this "stupid fan profile" comparison with the Asus Thor?  Because I don't believe the Thor is any better.

 

Tom's review calls it "silent":  https://www.tomshardware.com/reviews/corsair-ax1600i-psu,5406.html

 

Fan graphed:  https://www.tomshardware.com/reviews/corsair-ax1600i-psu,5406-5.html

 

Fan noise versus Wentai 1616W, Kolink 1500W and Enermax 1800W:  https://www.techpowerup.com/review/wentai-aidan-t1616-1616-w/6.html

 

 

 

No, I've had the AX1500i in service since 2015(?).  Warrantied once a couple years ago when it died after a shutdown and refused to restart.  It's just that I sit 24/7 in the 600-700W range so I'm always sitting in the 17-25dB range here (Cybenetics):

 

 

The Thor 1600W allegedly is silent until about 1000W and wouldn't hit equivalent noise until >1200W.  I'm assuming <6dBA means the fan isn't running at all.

 

 

 

I know I'm not a mainstream audience however.  It's just that the AX1500i is the loudest and only thing I can hear around my case now, because the rest of it is encased in 3" of foam and the PSU happens to be in the same place I have to have a gap in the foam for cables and hoses to come out.  I wish I could let the thing run hotter and not have the fan run at all:

 

 

Judging from what I see even up at 1100W it's targeting <40C with the fan RPM increasing to hit that.

 

I see the AX1600i seems to have gone much quieter, but $610 doesn't sound appetizing, especially when it's now a 4-year-old design.

[Guest]

2 hours ago, AnonymousGuy said:

I see the AX1600i seems to have gone much quieter, but $610 doesn't sound appetizing, especially when it's now a 4-year-old design.

LOL!  "Four year old design" doesn't mean much when everything else on the market uses an "even older design". 

 

So, are you using a bunch of RGB fans? Maybe with white LEDs?

 

[AnonymousGuy]

50 minutes ago, jonnyGURU said:

LOL!  "Four year old design" doesn't mean much when everything else on the market uses an "even older design". 

 

That's why I haven't been interested in changing out the PSU until now: the 1600 T2 and AX1600i have been the top dogs since 2018 and that's been about it.  So when Seasonic wants to build a new PSU for Asus on the same GaN digital platform blah blah as the AX1600i but with an emphasis on making it quiet, I take notice. 

 

1 hour ago, jonnyGURU said:

So, are you using a bunch of RGB fans? Maybe with white LEDs?

 

Yes, about 1000 of them.  Boy is my case packed!

 

[Guest]

15 minutes ago, AnonymousGuy said:

That's why I haven't been interested in changing out the PSU until now: the 1600 T2 and AX1600i have been the top dogs since 2018 and that's been about it.  So when Seasonic wants to build a new PSU for Asus on the same GaN digital platform blah blah as the AX1600i but with an emphasis on making it quiet, I take notice. 

LOL!  It's not Seasonic. It's Wentai.  HUGE difference..... 

 

15 minutes ago, AnonymousGuy said:

Yes, about 1000 of them.  Boy is my case packed!

So you don't know what the actual sound level will be.  Even with the Wentai.  Because the +5V (and +3.3V for that matter) aren't even passively cooled, when you have a bunch of RGB going, the fan tends to spin up really high to keep those MOSFETs cool.  

 

Have you ever bothered to do a comparison with your own PC with RBG on versus RGB off?

 

 

[SavageNeo]

is this man calling the best psu in the market bad, because ti has "four years old desings"??

 

Asus rog thor, is pretty bad, atleast the Seasonic rebranded thor units are. They are just Seasonic prime, with worse fan and heatsinks.

[Dogzilla07]

@AnonymousGuyHave you looked at Be Quiet! Dark Power Pro 12 ? (If you're very sensitive to noise, the stock fan + custom air inlet + Lian Li/Phanteks meshfilter instead of a grill makes it unbeatable for noise performance in theory). Additionally it's the only one with unchanging 6-10dB noise at up to 150w on the 5v&3.3v rail (pic below). In comparison the Asus doesn't even show testing over 110w, and the Corsair AX1600i is louder in the 117-150w range. the Pro 12 is also a fully digital platform (from CWT) like your current AX1500i.

 

Though an absolute silence user on another forum did end up returning it, as there was "switching" coil whine noise when the PSU wasn't loaded. (With the insane amount of RGB you might not experience this, but depending what exactly it is, who knows, maybe it shows up for you as well).

 

The 4000 series supposedly has Gen 5 dongles that are gonna have 4x8pin on one side

 

 

 

[Juular]

6 hours ago, jonnyGURU said:

So you don't know what the actual sound level will be.  Even with the Wentai.  Because the +5V (and +3.3V for that matter) aren't even passively cooled, when you have a bunch of RGB going, the fan tends to spin up really high to keep those MOSFETs cool. 

I mean, looking at the graph (at 40°C ambient) it looks like it is quiet, but whether Wentai/ASUS actually made sure that DC-DC doesn't blow being under full load but passively cooled for prolonged periods is the another question.

 

Back to the question about ASUS THOR II being able to work with ATX 3.0 GPUs, it's not about the 12/12+4 connector, because that's not a problem, you get a new cable with 3x8-pin on PSU side and 12+4-pin on load side with appropriate 'signal' pins shorted to ground to tell the GPU that it's good to go. Instead, it's about the design itself, it needs to be able to work with 1.6kW transients associated with the 600W continuous power draw by ATX 3.0 spec, whether it would able to no one knows.