Precautions before using Dual PSU in my PC?

[Carsomyr]

I've used a Dual PSU configuration back when I was mining crypto so I know some of the basics.
I intend to use my two PSUs from my mining rig which doesn't serve anymore for my main PC. The reason being that my current PC has plenty of power with the current config but lacks modular cables and just has less of everything. I run many accessories and need more specific cables to be used (sata, molex etc...)
I have a weird PC case... so long story short, a dual psu config would solve a lot of my problems.

I assume the smart thing to do would be to power the CPU/mobo and some components on one power supply and the GPU (rtx 3080) on the other (along with whatever other small components I need to use (fans, leds, rgb etc...)
Are there other things to plan for when using dual psu config?

 

[eidairaman1]

I've used a Dual PSU configuration back when I was mining crypto so I know some of the basics.
I intend to use my two PSUs from my mining rig which doesn't serve anymore for my main PC. The reason being that my current PC has plenty of power with the current config but lacks modular cables and just has less of everything. I run many accessories and need more specific cables to be used (sata, molex etc...)
I have a weird PC case... so long story short, a dual psu config would solve a lot of my problems.

I assume the smart thing to do would be to power the CPU/mobo and some components on one power supply and the GPU (rtx 3080) on the other (along with whatever other small components I need to use (fans, leds, rgb etc...)
Are there other things to plan for when using dual psu config?

So its current capability is enough? The only thing I suggest is a atx 24 initiator so when you hit power it starts both at the same time, and a means of protecting the other atx 24 pin so it doesnt short on bare metal or a pcb. Also maintain proper airflow.

 

[Harlow]

I assume the smart thing to do would be to power the CPU/mobo and some components on one power supply and the GPU (rtx 3080) on the other

You'll probably get away with this, but it's not what I call good engineering practice. I've done it in the past, but on a really basic PC and only for a couple of hours.

If your GPU requires additional power from the PSU, e.g. a 6 or 8-pin cable, you'll end up feeding the 3080 from two +12V sources. PSU #1 will supply 12V at up to 75W via the PCIe x16 socket on the mobo. PSU #2 will supply the remaining power via the +12V cable.

Although it's probable the two different +12V supplies (mobo and cable) will feed different voltage converters on the GPU card, there's no guarantee the two +12V rails will remain isolated from each other. You may end up connecting the nominal +12V rails on each PSU together. Both +12V supply rails will be dynamic, because output voltage varies continously depending on load.

There might be a difference of several tenths of a volt between the two nominal +12V rails, leading to "interesting" load sharing current flows. How well each PSU copes with external voltage from another PSU being "stuffed up" its own output remains to be seen.

What you're trying to do is common practice in servers, where multiple (often redundant) PSUs are employed. In these situations, the tendency is to "diode OR" the PSU rails together, using special high power (low voltage drop) Schottky diodes, bolted to heatsinks.

https://www.eetimes.com/Redundant-power-techniques-for-servers-explained/

The PSU designers take into account the 0.2 to 0.3V volt drop across the Schottky power diodes and probably increase the output voltage on each PSU to +12.3V, to compensate for the diode drop. The result is the Blade server receives a nominal +12.0V after the diodes, but each PSU is isolated from the others using one way diodes. N.B. Normal Silicon diodes drop 0.7 to 0.8V.

You're not going to be able to tweak ordinary PSUs to provide +12.3V and you probably don't want to invest in expensive high power Schottky diodes, capable of handling at least 50A (plus big heatsinks). If you did fit diodes, your GPU (and mobo) would receive only +11.7V, which would increase current draw. Probably not a good idea.

By all means go ahead with your plan, but bear in mind there's a slight chance things might go wrong. Ideally, you'd invest in a single more powerful PSU that copes with everything, but I appreciate your dilemma.

I've used a Dual PSU configuration back when I was mining crypto so I know some of the basics.

Were you using standard GPUs in the mining rigs or ASICs? You may already have enough experience with other GPUs to be reasonably confident of success. Best of luck.

 

[Aquinus]

You might not like this take, but if I were in the same position, I would get a PSU with a proper number of outputs for your needs and sell off or give away the other two. A single PSU solution outside of specifically designed redundant power supplies is going to be the safest option. That isn't to say that you can't make it work, I just think the amount of risk being introduced by doing this is likely not worth frying hardware over should you get something wrong.

 

[GerKNG]

Sell both and buy a proper new one.

 

[Carsomyr]

So its current capability is enough? The only thing I suggest is a atx 24 initiator so when you hit power it starts both at the same time, and a means of protecting the other atx 24 pin so it doesnt short on bare metal or a pcb. Also maintain proper airflow.

I've got the initiator you're talking about. As that's how I was running my dual psu mining rig before.

Yeah the two PSUs are 850w each (seasonic prime gx-850).

About your second thing you said "some way to protect the other 24 pin so it doesnt short on metal or pcb". Here I'm not sure I understand what you mean.

Could you elaborate?

You'll probably get away with this, but it's not what I call good engineering practice. I've done it in the past, but on a really basic PC and only for a couple of hours.

If your GPU requires additional power from the PSU, e.g. a 6 or 8-pin cable, you'll end up feeding the 3080 from two +12V sources. PSU #1 will supply 12V at up to 75W via the PCIe x16 socket on the mobo. PSU #2 will supply the remaining power via the +12V cable.

Although it's probable the two different +12V supplies (mobo and cable) will feed different voltage converters on the GPU card, there's no guarantee the two +12V rails will remain isolated from each other. You may end up connecting the nominal +12V rails on each PSU together. Both +12V supply rails will be dynamic, because output voltage varies continously depending on load.

There might be a difference of several tenths of a volt between the two nominal +12V rails, leading to "interesting" load sharing current flows. How well each PSU copes with external voltage from another PSU being "stuffed up" its own output remains to be seen.

What you're trying to do is common practice in servers, where multiple (often redundant) PSUs are employed. In these situations, the tendency is to "diode OR" the PSU rails together, using special high power (low voltage drop) Schottky diodes, bolted to heatsinks.

https://www.eetimes.com/Redundant-power-techniques-for-servers-explained/

The PSU designers take into account the 0.2 to 0.3V volt drop across the Schottky power diodes and probably increase the output voltage on each PSU to +12.3V, to compensate for the diode drop. The result is the Blade server receives a nominal +12.0V after the diodes, but each PSU is isolated from the others using one way diodes. N.B. Normal Silicon diodes drop 0.7 to 0.8V.

You're not going to be able to tweak ordinary PSUs to provide +12.3V and you probably don't want to invest in expensive high power Schottky diodes, capable of handling at least 50A (plus big heatsinks). If you did fit diodes, your GPU (and mobo) would receive only +11.7V, which would increase current draw. Probably not a good idea.

By all means go ahead with your plan, but bear in mind there's a slight chance things might go wrong. Ideally, you'd invest in a single more powerful PSU that copes with everything, but I appreciate your dilemma.


Were you using standard GPUs in the mining rigs or ASICs? You may already have enough experience with other GPUs to be reasonably confident of success. Best of luck.

I'm not sure I fully understand. Psu 1 would not be powering the gpu, it would power the motherboard, cpu and small components. Psu 2 would be in charge of the gpu solely no?

It's been a while since I used my mining rig some of those notions are far behind in my memory. I'll re read your comments and do some research to make sure I know what I need to before doing the dual psu config

And yes in my mining rig I was using all rtx GPUs. Although they were all connected to pcie risers with usb cable so in some ways they were better isolated from the psu

I should also mention my old crypto rig where I had two PSUs, I kept using it as a secondary gaming pc for my son. It's been running on that dual psu setup for years now with no issues.

(rtx 2080 ti)

 

[Harlow]

I'm not sure I fully understand. Psu 1 would not be powering the gpu, it would power the motherboard, cpu and small components. Psu 2 would be in charge of the gpu solely no?

Sorry if I didn't make myself crystal clear. It can be difficult when explaining electronics and electrical engineering concepts to a lay person.

If PSU1's 24-pin lead is connected to the mobo, the +12V supply (yellow wires) will feed the PCIe slot containing your discrete GPU card.

If you fit a low powered GPU, e.g. GTX 1650, GT 730, without an additional +12V input (from a PCIe 6/8-pin cable or a 12VHPWR cable) then the GPU will draw all its 12V power from the PCIe slot on the motherboard, up to the 75W per slot limit.

If you fit a bigger more powerful GPU, e.g. GTX 3060, GTX 5080, requiring additional +12V power (from PCIe 6/8-pin or !2VHPWR cables) AND you use PSU2 to provide this supply, you'll end up powering the GPU from BOTH power supplies, i.e. up to 75W via the PCIe x16 slot on the motherboard from PSU1, PLUS the remainder of the 12V power from the PCIe/12VHPWR leads on PSU2.

If these 12V feeds from both PSUs are connected together inside the GPU, you will have tied the 12V rails on both PSUs together. As I already mentioned, it is not good practice to connect two power supplies together in this manner. The preferred solution in servers and similar systems is to "Diode OR" the PSUs together and boost the voltage in front of the diodes slightly to compensate for the volt drop across the diodes.

Despite having two "identical" 850W PSUs, the two +12V rails will not be at exactly the same voltage. At one instant in time PSU1's 12V output might be higher than PSU2's 12V output. At other times, the situation might be reversed. If you tie both PSUs together inside the GPU, the two 12V rails will adopt a common voltage, roughly half way between the instantaneous output of each PSU, if they were working in isolation.

There is no telling if the PSU with the higher voltage 12V rail will force current into the PSU with the lower voltage rail. Hence my "don't like it up 'em" cartoon. We're probably talking only a few hundreds of milliVolts difference here, but given the high current capability of an 850W power supply, there could be 71A current capability on each PSU (850W/12V=70.8A). In practice you might not see more than a few hundred milliAmps, maybe an Amp or two, flowing the "wrong" way into the 12V rail of the PSU with the lowest voltage, but you might "reverse bias" components inside this PSU causing it to fail. Either that or a protective trip might operate shutting down one PSU.

If you've ever tried to jump start a car from another car using heavy duty leads, you'll see a fat spark bridge the gap as you make the final connection. The spark occurs because the 12V batteries in each car at different voltages. When you tie the two cars together, the car with the flat battery will see the voltage rise slightly. Conversely, the car with the healthy battery will experience a slight volt drop.

That's why we're suggesting a better approach would be to ditch the two PSU concept and fit a single PSU with sufficient power for the whole system.

As I said before, you'll probably get away with it, but if you see "magic smoke" rising out of your GPU, you've only got yourself to blame.

If this still doesn't make sense and you proceed with your solution, keep a fire extinguisher handy (CO2 or dry powder, not liquid or foam).

 

[Carsomyr]

Sorry if I didn't make myself crystal clear. It can be difficult when explaining electronics and electrical engineering concepts to a lay person.

If PSU1's 24-pin lead is connected to the mobo, the +12V supply (yellow wires) will feed the PCIe slot containing your discrete GPU card.

If you fit a low powered GPU, e.g. GTX 1650, GT 730, without an additional +12V input (from a PCIe 6/8-pin cable or a 12VHPWR cable) then the GPU will draw all its 12V power from the PCIe slot on the motherboard, up to the 75W per slot limit.

If you fit a bigger more powerful GPU, e.g. GTX 3060, GTX 5080, requiring additional +12V power (from PCIe 6/8-pin or !2VHPWR cables) AND you use PSU2 to provide this supply, you'll end up powering the GPU from BOTH power supplies, i.e. up to 75W via the PCIe x16 slot on the motherboard from PSU1, PLUS the remainder of the 12V power from the PCIe/12VHPWR leads on PSU2.

If these 12V feeds from both PSUs are connected together inside the GPU, you will have tied the 12V rails on both PSUs together. As I already mentioned, it is not good practice to connect two power supplies together in this manner. The preferred solution in servers and similar systems is to "Diode OR" the PSUs together and boost the voltage in front of the diodes slightly to compensate for the volt drop across the diodes.

Despite having two "identical" 850W PSUs, the two +12V rails will not be at exactly the same voltage. At one instant in time PSU1's 12V output might be higher than PSU2's 12V output. At other times, the situation might be reversed. If you tie both PSUs together inside the GPU, the two 12V rails will adopt a common voltage, roughly half way between the instantaneous output of each PSU, if they were working in isolation.

There is no telling if the PSU with the higher voltage 12V rail will force current into the PSU with the lower voltage rail. Hence my "don't like it up 'em" cartoon. We're probably talking only a few hundreds of milliVolts difference here, but given the high current capability of an 850W power supply, there could be 71A current capability on each PSU (850W/12V=70.8A). In practice you might not see more than a few hundred milliAmps, maybe an Amp or two, flowing the "wrong" way into the 12V rail of the PSU with the lowest voltage, but you might "reverse bias" components inside this PSU causing it to fail. Either that or a protective trip might operate shutting down one PSU.

If you've ever tried to jump start a car from another car using heavy duty leads, you'll see a fat spark bridge the gap as you make the final connection. The spark occurs because the 12V batteries in each car at different voltages. When you tie the two cars together, the car with the flat battery will see the voltage rise slightly. Conversely, the car with the healthy battery will experience a slight volt drop.

That's why we're suggesting a better approach would be to ditch the two PSU concept and fit a single PSU with sufficient power for the whole system.

As I said before, you'll probably get away with it, but if you see "magic smoke" rising out of your GPU, you've only got yourself to blame.

If this still doesn't make sense and you proceed with your solution, keep a fire extinguisher handy (CO2 or dry powder, not liquid or foam).

Hahaha thanks. It helped me understand better. I did some research myself and i found that in a mining rig, apparently somehow the pcie riser kits, separate the gpu from the PSUs ensuring no conflict between the two or janky behaviour. That situation is null and void if the gpu is directly connected to the motherboard (if I understood correctly). Making the dual psu solution bad.

Right now, I'm looking at different PSUS trying to make sense of which one would be best for a rtx 5090 upgrade. I'm finding mixed answers. Some say ATX. 3.0 is fine and some say you need 3.1...

Then some say 3.0 is fine IF you have a special adapter OR cable. So right now I'm trying to clear that question mark up before I choose my new Single PSU

 

[Harlow]

somehow the pcie riser kits, separate the gpu from the PSUs ensuring no conflict

It's not difficult. All they need to do is omit the +12V connections on the riser card. The important thing is you can't be sure if a GPU will work without +12V on the PCIe edge connector without testing, or checking with the card manufacturer. I guess some GPUs might work fine without mobo power, but others might not.

I'm finding mixed answers. Some say ATX. 3.0 is fine and some say you need 3.1...

Probably best to buy something with the new 12V-2x6 connector and not a 12VHPWR connector.
https://www.tomshardware.com/news/1...eeded-revision-meet-the-new-12v-2x6-connector

If you get an old-style PSU, you could use an adapter plus a bunch of 8-pin PCIe leads (4 for 600W) but each plug/socket combination adds extra contact resistance.

Check out the pros and cons of any PSU by reading reviews before buying.

 

[rockit00]

I used to run dual PSU's in my rigs. https://www.newegg.com/visiontek-ju...MCbHO-ogjf9PbXR2jOAd8pSkXyzwbGrTJJItTPIb5Jkb9. I was using MGE Vortec 500W main PSU with the Visiontek Juicebox for 950W total. I think Fortron Source Power makes a similar AUX power supply also.

 

[Shrek]

Got these two references from Lazar Rozenblat, author of Switching Power Supply Design: A Concise Practical Handbook

• https://www.smps.us/Powering_cPCI_systems.pdf
• https://www.smps.us/Specifying_Power_Supplies.pdf