Power Supply Looking to power two GTX 1080 Ti without a SFX PSU

[aquelito]

Hi there,

I'm looking to power two GTX 1080 Ti Mini and an i7 7700, if possible without a SFX PSU.

Indeed, I've got this Supermicro board than can be powered without the 24-Pin connector :
https://www.supermicro.com/products/motherboard/Core/Q170/X11SSV-Q.cfm
Would be too bad to have to add a 24-pin cable to the system...

I did not find 1000 solutions :

#1
Use a 12V 650W Artesyn PSU with a custom load switch for my two GPUs.
https://www.mouser.fr/new/artesyn/artesyn-CNS650-M-acdc-power-supplies/

Problem : it's pretty expensive. Ripple is OK at around 97.6 mV :
https://www.artesyn.com/power/assets/trn_ac-dc_cns650_trn_20170125_1485360518_techref.pdf (page 8).

#2
Use a 550W 24V EOS Power PSU with two Dynamo 360 but I'm not sure I can run both Dynamos in parallel.

#3
Use a 550W 12V EOS Power PSU with two One2 Distro board, but same question : not sure I can use two boards in parallel.
https://uk.rs-online.com/web/p/embedded-switch-mode-power-supplies-smps/1222485/

Thanks !

 

[aquelito]

@Thehack @Kmpkt

 

[Thehack]

Hey,

So in terms of using parallel supply, I think you can use it just fine, as each power supply would likely power separate components. The issue is how the common (ground) works, and if you encounter load balance issues (one psu working harder than the other).

I think the SFX 750W would be the most reliable and power densed option. If you need more power, then I recommend dismantling an ATX PSU (after draining the power) and modifying it to be a bit smaller. The Silverstone strider series is actually quite power dense and can be modified much more economically than using industrial types.

 

[aquelito]

Thanks !

Yes, after a thorough volume comparison, the SFX 750 is the best solution.
I can always jumper the 24-Pin if I want to.

Your ATX PSU modification idea seems pretty interesting.
Do you have any example ?

 

[Valantar]

Yes, after a thorough volume comparison, the SFX 750 is the best solution.
I can always jumper the 24-Pin if I want to.

I have no idea if this would work, but might it be possible to make a small 2-pin cable to connect the correct pins on the PSU to the PS_ON and a ground connector on the motherboard to turn the PSU on without requiring the entire 24-pin cable? I guess something like this depends on how the motherboard treats its power input (i.e. if it has embedded logic to disable one power connector if another is detected or something similar), but that sounds rather unlikely IMO. This way you could get proper PSU on/off control and the advantages of not running a bulky 24-pin cable.

 

[Thehack]

I have no idea if this would work, but might it be possible to make a small 2-pin cable to connect the correct pins on the PSU to the PS_ON and a ground connector on the motherboard to turn the PSU on without requiring the entire 24-pin cable? I guess something like this depends on how the motherboard treats its power input (i.e. if it has embedded logic to disable one power connector if another is detected or something similar), but that sounds rather unlikely IMO. This way you could get proper PSU on/off control and the advantages of not running a bulky 24-pin cable.

It may be possible. You need to provide 5VSB, PS_on, and COM to the board at minimum.

This is a weird board so it'd only work with this one.

 

[aquelito]

I think I already mentionned that but someone already made that PCB for me. But it doesn't work

 

[Valantar]

It may be possible. You need to provide 5VSB, PS_on, and COM to the board at minimum.

This is a weird board so it'd only work with this one.

If the board supports pure 12V input, wouldn't it then necessarily be able to convert these voltages on-board?

I think I already mentionned that but someone already made that PCB for me. But it doesn't work

That's not quite what I had in mind: your board (looks like it) intercepts the power button signals and passes these to the PSU. What I'm suggesting doesn't involve the power button, but a simple, two-wire connection between the motherboard 24-pin's PS_ON pin and its counterpart in the PSU's 24-pin receptacle, as well as a motherboard ground pin and matching PSU-side pin. After all, these pins are normally shorted out when you press the power button connected to the motherboard when using a 24-pin power connector, so why should they not also be shorted out when using the 12V connector for power instead?

If you have a multimeter you could test this: connect a 12V source to the 12V connector, set your multimeter to continuity and attach the probes to PS_ON and ground in the motherboard-side 24-pin connector, and press the power button. If the meter beeps when you press the button, this method should be able to switch your PSU on.

Of course, this won't work if the 12V connection expects the 12V PSU to always be switched on, which seems to be a possibility as it doesn't have any sort of integrated PS_ON in the 12V connector.

 

[Thehack]

If the board supports pure 12V input, wouldn't it then necessarily be able to convert these voltages on-board?

That's not quite what I had in mind: your board (looks like it) intercepts the power button signals and passes these to the PSU. What I'm suggesting doesn't involve the power button, but a simple, two-wire connection between the motherboard 24-pin's PS_ON pin and its counterpart in the PSU's 24-pin receptacle, as well as a motherboard ground pin and matching PSU-side pin. After all, these pins are normally shorted out when you press the power button connected to the motherboard when using a 24-pin power connector, so why should they not also be shorted out when using the 12V connector for power instead?

If you have a multimeter you could test this: connect a 12V source to the 12V connector, set your multimeter to continuity and attach the probes to PS_ON and ground in the motherboard-side 24-pin connector, and press the power button. If the meter beeps when you press the button, this method should be able to switch your PSU on.

Of course, this won't work if the 12V connection expects the 12V PSU to always be switched on, which seems to be a possibility as it doesn't have any sort of integrated PS_ON in the 12V connector.

You still need 5VSB to run power on circuit.

The other voltages should be on board as you've said.

The PS_on is just a signal. When you press the power button, you are turning on a transistor to pull PS_on to ground. So it likely needs to 5VSB to run this logic (and other board stuff)

 

[aquelito]

Yes, the board has an on-board cricuit to convert 12V to 5V and other voltages.
It has a Molex connector to power HDDs, etc. Same as a thin mini-itx board.

 

[Valantar]

You still need 5VSB to run power on circuit.

The other voltages should be on board as you've said.

The PS_on is just a signal. When you press the power button, you are turning on a transistor to pull PS_on to ground. So it likely needs to 5VSB to run this logic (and other board stuff)

Ah, I see. I guess the question then becomes: how does the board switch on at all if it doesn't also generate 5VSB by itself? And if it does this on-board, doesn't this also mean that any 12V PSU connected needs to be always-on?

 

[Thehack]

Ah, I see. I guess the question then becomes: how does the board switch on at all if it doesn't also generate 5VSB by itself? And if it does this on-board, doesn't this also mean that any 12V PSU connected needs to be always-on?

If you are using a regular 12V psu, the 12V power is always applied and it'd just have a 5vsb off of that.

However, the issue is we are using pcie power devices that must be turned on and off, which is much different than its intended use as a server board.

Though I'm not sure how all this interacts.

It is also possible that all rails are done off the 12V and the motherboard ignores 5V and 3.3V on the 24pin atx.

It is also possible that the motherboard turns off its internal 5v and 3.3v rail and just uses the psu if it senses 5vsb being present.

Hard to tell without a theory of operation or block diagram.

 

[Valantar]

If you are using a regular 12V psu, the 12V power is always applied and it'd just have a 5vsb off of that.

However, the issue is we are using pcie power devices that must be turned on and off, which is much different than its intended use as a server board.

Though I'm not sure how all this interacts.

It is also possible that all rails are done off the 12V and the motherboard ignores 5V and 3.3V on the 24pin atx.

It is also possible that the motherboard turns off its internal 5v and 3.3v rail and just uses the psu if it senses 5vsb being present.

Hard to tell without a theory of operation or block diagram.

So, long story short, if @aquelito wants to use an SFX PSU but not the 24-pin, the best route would be to run a custom 12V cable to the 12V input and keep the PS_ON pin on the PSU shorted to keep the PSU on all the time?

 

[Thehack]

So, long story short, if @aquelito wants to use an SFX PSU but not the 24-pin, the best route would be to run a custom 12V cable to the 12V input and keep the PS_ON pin on the PSU shorted to keep the PSU on all the time?

That's the simplest way but it creates the problem of your gpu being stuck on.

 

[Valantar]

That's the simplest way but it creates the problem of your gpu being stuck on.

Of course. I guess being a server board, this likely isn't really designed to be turned off for any significant period of time anyhow, so this wouldn't be a problem in those cases.

 

[aquelito]

That's what that lil' PCB was designed for : start any PSU with a PS_ON function with your power switch.
No luck so far. I'll try with a SF750.

Otherwise I'll just tuck away that damn 24-pin cable, not an issue either.

 

[Valantar]

That's what that lil' PCB was designed for : start any PSU with a PS_ON function with your power switch.
No luck so far. I'll try with a SF750.

Otherwise I'll just tuck away that damn 24-pin cable, not an issue either.

I have two hypotheses as to why that didn't work:

1 (most likely IMO): The motherboard doesn't generate on-board 5VSB, and as such doesn't trigger the logic to short the PS_ON pins when using that power source.

2 (still possible, and at least testable): the power button is a momentary switch, so it will only short out the PSU pins for a very short time, after which the PSU will switch off again. The motherboard, not getting an external 5VSB signal, will only have power for a fraction of a second while the button is pressed - if that, as the PSU is just turning on. This might not be enough time to properly initialize the power circuitry on the board to maintain the shorting of the PS_ON pin after the button is released.

Of course this is all guesswork. Did you try pressing and holding the power button with that board installed? If you have a multimeter, you try powering the board with the 12V input and checking for continuity between the PS_ON pin in the motherboard 24-pin and motherboard ground. If there's continuity, the motherboard does initialize the PS_ON signal, so your board should work, but might require a longer press to initialize fully.

 

[Thehack]

That's what that lil' PCB was designed for : start any PSU with a PS_ON function with your power switch.
No luck so far. I'll try with a SF750.

Otherwise I'll just tuck away that damn 24-pin cable, not an issue either.

Can I see the schematic for it?

 

[aquelito]

Yes sure I'll share it tomorrow.
I think it might even be visible in our MPs

 

[reifnotreef]

You could look into something like this: https://www.hdplex.com/hdplex-800w-dc-atx-with-16v-63vdc-input.html