When a computer (usually a server) has two power supplies, where the 2nd is fully redundant, there can be an advantage to running the 2nd PSU from a separate power source. Thus instead of being protected from a failure of a PSU itself, it is also protected from a failure in the power source, whether protected with a UPS or not.

The possible problem I see is when that 2nd power source is on a different phase. If the supply voltage is 240 volts AC on a three phase system, there is a 416 volt difference between phases. This would mean that between some parts inside one PSU and the same parts inside another PSU, there is 416 volts. Since they are separate, in theory that should be the end of it, since the frames between them should be grounded, and thus no circuit component would see that 416 volts.

But what about these PSU control boards that interconnect between hot swappable PSUs? Is there any chance the AC line voltage is available there?

I've seen no warnings on PSUs that say to avoid crossing dual PSU systems between phases. So it's plausible to believe there is no issue with that. But I live in the USA where 416/240 volt power is uncommon and maybe the warnings would just not be put out here where the similar risk is only half the voltage, well within the 100-240 volt range most PSUs operate at.

I've curious if anyone using voltage in the 380/220 to 400/230 to 416/240 volt ranges has ever encountered or hear of any issues like this (not counting showoffs playing with live wires).

The input voltage is immediately reduced in these. I doubt you can screw them up somehow so as to create an issue. They will come with complete documentation as to how to install and what requirements are make to meet code. You local code or state code may vary and you have to select the most harsh in most cases. Never less than NEC.

In the very most basic sense, server farms have been built to some standard. That standard would take into account the electrical code of the region and most simply create to US standards for a majority of the earth.

Yes, worldwide server farms, even some in USA, are built with these voltages. I'm just curious if there is any need to be sure both PSUs are on the same phase.

Yes, normally the PSUs reduce voltage. But perhaps some hot swap controller design could have wire leads on the AC side to directly switch off the mains leads under conditions it is designed to do that. If it has such leads, the potential for 416 volts between those leads on the different PSUs exists (up to 480 volts if the phase angle is 180 degrees). I would hope the engineer designing such a controller would be sure to consider the extreme possibility of the power phase angles.

At least one maker of UPSes for 208 volts single phase makes them in single ended, apparently ungrounded, mode. The possibility exists to get as much as 416 volts between PSUs if they are separated connected to different UPSes happens, assuming end-grounding of the inverter. Ungrounded leaves open that possibility even if both UPSes are on the same phase.

NEC does not address issues within computers. It is a standard for (mostly) building and premise wiring and power control, for safety, fire prevention, etc.

The documentation would dictate how to install the devices. Most of those non-home user stuff has words like contact professional or consult local codes.
NEC would provide how to balance your phases.
The device ought to be certified by some safety agency be it UL or your local agency.

I don't doubt it but there could be unsafe devices. I no case would I suspect any voltage above 50v to be outside the case of any PSU by design or failure. Saying that I have seen a 460V VFD fail with +1000VDC legs. Normally one tests for AC, but I tested for DC also on a failed module and was a believer in checking all voltage or lockout tagout before you put your finger in there.

There are only a few instances where you can work on this stuff live. OSHA has all the limits for each class of hazard and the required protection. You might be surprised how low of voltage it takes to require some expensive protection and controls and training.

Unless I severely misunderstood your question, that would be impossible, unless:

1) The electrical inspector dropped the ball and allowed a circuit to be wired incorrectly during the construction of the building, or

2) One modifies the plug to fit the outlet backwards (they are directional for a reason).

Even though the two units may be on different circuits in my home, those circuits are supplied by the same three wires that feed my entire neighborhood. However, the point is moot-- even if the two units were on different phases, the PSU converts the output to DC, so the input phase is irrelevant to the output.

As someone who earned his money inside these things, let me explain.

The smpsu has the mains isolated in the early stages, and that isn't an issue. There is an issue running on 416 volts, as you are now dependent on 2 phases (416V is phase to phase). Further, doing a phase to phase thing means there is a much greater potential between earth and "neutral." And you have to earth a pc. A neutral-earth fault would be as catastrophic as a live/earth fault.

I would wire in neutral (which exists) and use 220V(any phase) each. Keep neutral near earth, and you can sleep in your bed and not get fired if things blow up.

How would it be incorrect?

The idea is to bring in 416Y/240 three phase, and run single phase circuits to each PSU. So the PSU would be connected line-to-neutral at 240 volts. In order to gain the additional advantage of redundancy over a single phase loss, each machine that has dual redundant PSUs (can run the machine entirely from one PSU) would be wired from different phases. Consider 3 racks of machines. 6 single phase circuits would come from the panel. 1=A, 2=B, 3=C, 4=A, 5=B, 6=C. Rack 1 gets circuits 1 and 2. Rack 2 gets circuits 3 and 4. Rack 3 gets circuits 5 and 6. Each circuit feeds a separate UPS so there are 2 single phase UPSes per rack. Each UPS has its own PDU. Each machine is split between the PDUs.

Now, lose one phase. The UPSes on that phase can run down and shut off, and each computer still has power.

My point is, there can be an advantage I term "phase diversity". And this is not incorrect building/premise wiring.

The risk of a broken neutral might be a big issue, too. I might need to consider that.

Backwards will get more voltage? Interesting.

As long as that is all the PSU does, fine. Ordinary PSUs do. But those special PSUs that are hot-swappable in a box with some funny connections on a control board MIGHT also have an AC line sensor going to that controller. Maybe they don't or maybe they do and it's engineered for the worse nominal case of 480 volts (e.g. the highest voltage of the common 100-240 doubled due to 180 degree phase difference).

---------- Post added 2011-04-29 at 16:17 ----------

Actually, one of the ideas of dual redundant (where the whole machine can run from just one of them) power supplies is to be INdependent of phase loss. If the machine is connected to phase A via PSU 1, and phase B via PSU 2, and you have a single phase loss, you're still up.

Maybe you are thinking I'm running the PSU on a phase-to-phase or line-to-line connection (which would be 208 volts on a typical USA three phase service). I am doing that now with 208Y/120 service. But my idea is for a new data center to use 416Y/240 and each PSU to be connected phase-to-neutral to get 240 volts. The problem is that this does have a 416 volt difference between some wire in PSU 1 and that same wire in PSU 2 when different phases are used to get that redundancy coverage over a phase loss.

But I disagree about a phase to earth potential. It will always be 120 on a 208Y/120 system, and 240 on a 416Y/240 system.

Yep, it seems you thought I was wiring each PSU between phases and effectively feeding them with 416. The neutral was always going to be there, so each PSU would be fed with 240.

The issue was in the PSU design as to just how thoroughly they are mains isolated. In a normal PSU, I'd expect it to be fully so because there is only the DC wires coming out the other end. The hot swap PSUs are a different beast, and there was my concern. I could not determine if there wasn't something going to the controller that operates between them that would be at mains voltage.

Of course any engineer worth his degree would consider the maximum of the voltage range (e.g. 240 for 100-240) and if the design had the mains voltage from both anywhere together (for example a voltage sense wire from the controller to the mains input of each PSU), the design would have to be made to handle double that voltage for the worst case (180 degree phase difference).

I just wanted to be sure of this. If it were a real issue, and machines really were such that they could not handle more than 240 volts that way, I'd probably not routinely hear about that in the USA, because powering data centers at 416Y/240 is still an uncommon thing (though I have suggested doing so since 1996).

Two posts got merged. Not sure how that happened.

In each PSU the "high" voltage is isolated against :

1. The case, at earth potential. (there are some small capacitors between the AC and rectified DC to earth, but they are all rated at >>415 V)

2. The secondary side of the PSU, at earth potential (12 V is negligible vs the voltage on the primary side)

This means that any isolation in there can see only "high" voltage from one side to earth, and there is nothing that "sees" both live supplies together

AND this isolation can withstand alone 500 V no problem if it's intact (many standards specify that the isolation should be tested for breakdown at 1KV or more)



Since the AC is immediately rectified anyway, you dont have to bother about same frequency (for example online UPS with output frequency slightly off won't matter)



One thing you have to care about is the neutral wire : It must be reliable, and any plugs / switches / breakers in the 3 phase side must be made to connect the neutral first and disconnect it last (only second to earth)

Having 2 phases connected without neutral means 2 power supplies get 415 V in series. It'll easily divide to "almost 415" and "almost 0" and can kill both power supplies and connected hardware

I think I see the problem.

You never control with line voltage. All controls are at low voltage. If ever find one then remove it from service and destroy it.

No PSU in these servers would ever have line voltage beyond it's case. It should not fail. If it did fail then the built in grounding/bonding and proper circuit limits should prevent catastrophe. There is almost no chance that any voltage beyond 50 VDC would be outside of that case.

It wasn't too long ago that PSU's did have a on/off switch that had line voltage. That was dumb indeed. After a few disasters you don't see that.

[QUOTE=Skaperen;4341136]Yep, it seems you thought I was wiring each PSU between phases and effectively feeding them with 416. The neutral was always going to be there, so each PSU would be fed with 240.[QUOTE]

I think it's terminology. Mentioning 416V _implies_ no neutral, as that is a phase to phase voltage.I wasn't worried about voltage - you could have had 416/220 transformers or 416v power supplies.

Well, the internals are all the same. there's inductors, supression, a bridge rectifier, and smoothing caps; then you have a tx and mosfet on the live side. There is a 'creepage' gap of 7mm between mains and low voltage on the pcb. Just look at any psu board. If the supression stuff goes, but not the fuse, the board can be carbonised, lowering resistance. There was a range of PSUs here which were responsible for burning down 2 houses, because the vdrs caught fire.

Mentioning a "416V electrical service" might imply no neutral. But I have seen many a reference to electrical services that I would refer to as 208Y/120, 220Y/127, or 240Y/139, or 380Y/220, or 400Y/230, or 416Y/240, or 480Y/277, or 600Y/346, or 1000Y/577, being referred to as just 208, 220, 240, 380, 400, 416, 480, 600, or 1000, respectively. The implication of no neutral might also be in the way things are connected.

Even if you do have a neutral wired into some equipment, if it has 2 or more phases coming in, there are points where the line-to-line voltage can be found.

In the case of dual PSUs, either they are connected line-to-line (practical up to 240Y/139) or line-to-neutral (practical up to 416Y/240). The latter is where the concern is.

In many cases, there are identifiable places where the line-to-line voltage can be considered as present. For example, two power cords plugged into different phases can have a 416 volt potential between them as they cross paths somewhere in the rack. Such cords are at least rated for 300 volts, and there being two layers of insulation, that should be effectively 600 volts, and so it would be plenty. This is not a worry.

In the case of the PSU enclosure, it should be grounded (and the whole PC as well). So any electrostatic fields between wires within the PSUs would be negated by the grounded frames between them. As long as those potential points are fully isolated within these frames, there's no concern. But I could not confirm this.

Unfortunately I don't have a spare PSU to dig into ... of the kind that slip into these dual PSU frames ... to see how it is wired. A regular standard PSU for ATX I have taken apart. But these ONLY have low voltage DC wires coming out, so unless there is a short circuit within, these are not a worry.

But you could do voltage sensing at line voltage.

But "the case" might have been considered to be including the circuitry where the two PSU frames plug into the power controller between them. That controller circuit might well have made sure no line voltage can reach the DC lines. But if it did have two line voltages there for sensing, did it design it so they can be at worst case 180 degrees apart? If so, no worry.

And, of course, if they have no line voltage running to a controller for sensing, there's no issue.

Are you referring to the front or near front main on/off switch like the original IBM PC had? Or are you referring to the mains switch physically on the back of the PSU? The latter I see no problem with. The former ... yeah, one place I worked at a junior sysadmin once wired that switch wrong and ended up causing a UPS shutdown that took out several servers and network routers.

BTW those AC line switches on the backs of PSUs do need to be double pole. One reason is that in North America, they can be wired to 208/240 line-to-line. In other places like parts of Europe, outlets/plugs are not polarized, and either line can be at 230 volt potential. If you ever see a PSU with a single pole switch, and with an IEC inlet, it's wrong. If it has a NEMA 5-15P inlet, and is wired properly, in theory it is OK. I can't remember if the old original IBM PCs were that way or not.

I meant the original pc's and clones like dells.

I have worked on a lot of power supplies in my decades. They all can fail. The load design ought to help prevent a major issue.

About two months ago we lost a 50HP VFD amplifier when it was moving at high speed. Took out the entire building circuit breaker. That should have never happened.


The sense lines that I tend to see are results of some downstream testing. I know some phase to phase tests that do provide a sense that are directly connected but not by the large current cables the system uses. They are connected by lines that couldn't take much more than 10 amps if that. I'd suspect the wiring or connectors would burn off. I suspect that is the situation in any power supply. Any good design would have to follow professional design standards.

Agreed. There is a concept of breaker coordination to prevent that. There are also current limiting breakers to assist. I wonder if that VFD went to a lower ... or negative (e.g. feeding power) ... power factor during its troubles.

We should never see this from a PC.


The sensing I might expect in a dual PSU hot-swap system would be to manage how power is taken from the PSUs. For example, if one has a rather low AC voltage coming in, it can be triggered to stop, or its DC lines opened to keep it from stress. There could also be temperature sensing. While a PSU can be designed to shut itself down for certain limits exceeded, the common controller can do a soft shutdown when conditions ... though not meeting the safety cutoffs ... can still help protect better protect the PSUs when only one of them has the bad conditions.