Bitcoin Forum

How does everybody feel about utilizing two (or three) PSUs?  Do you feel that connecting their ground is a good idea?

I traditionally thought that grounding them together was the way to go.  I built a molex adapter that simply connected ground to ground.  However, I went to disassemble my rig that had been running for months without issue, and I hadn't even hooked up my ground adapter. 

I recently tried expanding to 3x500W.  Something bad happened and I shorted the mobo.  I was grounding all three PSUs together. 

Discussion?

I think grounding them together is the best way to prevent power fluctuations and other such stuff. I'm no expert but that's what I heard.

When you shorted the mobo, did it die afterwards ?

It did.  I am not entirely confident I know what happened.  I think I was getting cocky about 3 PSUs = 1.

Grounding will be necessary... it's keeping the +ve rails apart that I would worry about.
You absolutely do not want to connect them up in parallel.

I thought I read the exact opposite; something about avoiding a ground loop?

FWIW, Ive run a rig on 2 PSUs for a few months, sharing nothing. Each PSU fed different cards, and it worked fine, but its not something I can guarantee or even recommend. Ive since upgraded the PSU to be safe.

I will point out there are VGA power supplies that provide extra 6/8 pin PCIE connectors, and afaik they are not connected to the main PSU - other than for powering up automatically.

I don't think internally sharing a ground is good idea. 

Each PSU is going to share the same external ground anyways.  You likely don't want to connect multiple PSU to different outlets where they could experience different resistance to ground but if you got 3 PSU connected to a power strip and it is plugged into the wall outlet I don't see what splicing ground wires together achieves beyond that.

As an example N+1 redundant power supplies for servers don't have any grounds tied together and there grounding can be more of a challenge because each supply should be on a different branch circuit connected to different UPS otherwise you compromise redundancy. 

Depends on what type of "grounding" you are talking about.  If you mean connecting to a common "earth ground" on the AC then they already share the same ground point when they are plugged into the wall (your ground wire in the house is connected to a common stake that is driven into the ground outside, or connected to a cold water pipe).  If you mean a common negative on the PSU's DC voltage then that is a different thing, I connected the PSU's together via DC negative on my setup.  You need to keep the connection between the 2 short to prevent ground loops though. 

Do a quick search as this was discussed in a much greater detail back around June/July...couldn't find a link to a writeup but I know I linked them in the past.

I don't mean to sound offensive here but I remember clearly you claiming to be an electrician.


If you do work as an electrician and are no expert at what you do, God save the Queen :)

Well there's always the disclaimer : if your GPUs / stuff blows up don't blame me ;)

That disclaimer should be on by default when trusting any information from the net  ;D

Does the answer to OP's question depend on the type of wiring at their premises?
Does it depend on whether their installation uses three wires (building has its own grounding) or just two (in which case the grounding is supposed done at the power station)?
I do remember reading an article about being supposed to connect ground to neutral if the building isn't grounded and the electrical installation is done with two wires.
...or am I just prattling nonsense?

I have just finished debugging my failed setup.  It turns out that the CPU and Motherboard were taken out in the fiasco. 

There are numerous good arguments here for not connecting the DC ground.  1) VGA PSUs and 2) Server PSUs (although, in this case the components are probably always powered by one or the other?) 

I think I am going to move away from my setup of connecting the DC ground together.  In the end, they are all connected to the same power strip (with ground). 

Thank you for everybody's advice.  I think I am still going to consider moving away from 3x500 to something simpler.  Maybe 2x750.  I am trying to power 6x5850.  Thoughts? 

I doubt you need 1500W to drive 6x 5850s.  I drive 3x5970 (6 total GPU) which are more powerful and AC wattage at the wall is 870W.  DC is probably closer to 800W.

The "internal" grounds are already pretty damn well connected.
1. On ATX PSUs, secondary GND is directly tied to safety ground, otherwise it'd be illegal to sell in most of the civilized world. Don't believe me? Just measure it.
2. Graphics cards have all PCIe power connector and card edge connector grounds tied into a massive ground plane. Again, if in doubt, measure it... use a 4-wire mOhm meter or you'll only be seeing 0.0 ;)

So, where does the "you have to tie grounds together" myth come from?
Simple.
Basically you need the following scenario:
1. Building with improper/unusual wiring and several V difference between safety ground potential between "close" outlets.
2. 2 PSUs that are *not* electrically connected via a common metal chassis (if both are screwed to the same steel chassis, that's usually a low enough resistance path to reduce the potential difference to safe levels of a few dozen mV).
3. said PSUs being plugged into said outlets. Now we have 2 PSUs where #1s GND is a few V different from #2s GND.
4. One connected to mainboard, another connected only to drives or similar. Now we have mainboard GND and drive GND differ by a few V...
5. Data cable from mainboard to drive has ground/shield wires (duh). Now you have grounds differing by several V connected with a few strands of AWG26, smell the PVC? Worse... if data lines make connection first, (relatively) large currents flow through the I/O pin protection diodes, usually resulting in dead controllers and/or drives.
In *that* scenario, connecting the grounds via a few AWG18 jumpers or similar reduces the potential difference between grounds to a safe level.

So short version... if it makes you feel good, tie secondary grounds together all you want, but it's flat out pointless if you're powering GPUs.

Good info, makes perfect sense. Thanks Art.

Excellent analysis.  What if the two PSU are *not* electrically connected?  Ie, an open rig without a chassis. 

The "common ground" between them might be slightly less common than expected.
As Art already pointed out, the voltage at "ground" might differ between these not interconnected PSUs.
If they power different components of the same PC, current will run where it's not supposed to (from one "ground" to another).
You can fry your components that way. That's not a very likely scenario though.

Mind you, there's nothing to worry about if you're running those PSUs from the same power extender - the power strip will provide common ground anyhow.

If each PSU is properly grounded then they are electrically connected via ground wire.  If you for some insane reason broke off the ground pin on the power cable then you likely would have an issue.

Awesome, sounds great!

The big concern I have with running two power supplies into one rig is that even if the grounds are externally linked, the output voltages can be different between the supplies. Lets say one supply puts out +12.1V and another puts out +12.4V, then there is a 0.3V potential between them, and current will flow between the two power supplies. The amount of current is only limited by the design of the power supply regulation and the resistance in the path. If one power supply is connected to your motherboard and supplies a video card +12v through PCIe slot (up to 75W), and the other is connected to the video card power connectors, then you seem to have made a path through the card and motherboard where this current can flow with little limitation until a trace goes up in smoke. While I have run such a system with two power supplies and it worked, frankly I am surprised it did.

No graphics card I know of does that, it's also a massive no-no according to PCIe spec.
Simple reason, defeats per-rail overcurrent protection -> fire hazard.
Not to mention the havoc that could cause with PSUs that have 12V rails coming from 2 independently regulated 12V supplies (not too uncommon in >1kW units nowadays).
Too lazy to dig up the exact text, but basically you're forbidden from creating a galvanic connection between each 12V aux connector and 12V from slot.
Also cards have to accept any power sequencing of 12V for any amount of time without sustaining damage.
They're not required to function if time from first to last 12V input coming up is > 100ms or if any input drops out after the initial 100ms.
But so far all cards I've seen seem perfectly happy with aux 12V coming up long before slot and staying on while slot 12V goes away and comes back (= aux supplied from 2nd always-on PSU).

I have successfully gotten 6 cards working, on 3 500W PSUs.  I am no longer connecting their grounds (explicitly).  I appreciate the detailed discussion on this issue.  Hopefully everything remains stable. 

Are you running anything on the 5V lines of the slave psus?

If not then tell us how long they did last ...

I have a couple of rigs running since september on 2 seasonic 750's  4x5970.   KISS   just short the plug and use the pci cables, thats it.

So the general consensus is that :

-no need for common ground as they have common ground from the power strip anyway

-do not mix and match cables into one card ( e.g. do not have one GPU powered by 2 PSUs and just stick to GPUs powered by individial PSUs )

-still need a 5V load in order to put less pressure on the 12V line ?

Thanks and can someone confirm I got it right ?

First the reason for putting load on 5V rail had nothing to do with the 12V rail.

Older PSU (due to higher 3.3V and 5.0V requirements) converted 120V AC to 12V, 5V and 3.3V.  Not having a minimum load was bad to the method used in the AC to multiple DC conversions.

Today honestly there is no reason entire computer couldn't run only on 12VDC.  70%, 80%, sometimes 95%+ of the load is on the 12V rail.  To improve efficiency most modern PSU are rail to rail conversion.  The PSU converts 120VAC to ONLY 12VDC.  Then based on load converts some of the power on the 12VDC -> 3.3V DC and/or 5 VDC.  There is no reason to put any load any of the rails.  

Artfortz claims there is no issue powering 1 GPU from multiple PSUs. Im not sure its something I would do if I could avoid it, but at least there is no such consensus.

OK. Now I really am confused.

So basically all that is needed is just a mechanism of turning both PSU on at same time ???

Like this :

http://www.frozencpu.com/images/products/main/cpa-167.jpg

Unless your PSU is an old group-regulated design, yup, that should be all you need.

Loading the 5V rails is a way to avoid what's called a cross-load, a situation where some rails aren't used at all and other are heavily loaded.
Group regulated PSUs could easily go out of spec - even dangerously so - in this scenario.

Keep in mind that if your PSU has a MINIMAL load defined for 3.3V or 5V, you should make sure that load is applied.
Otherwise you're running the PSU out of manufacturer's spec.

I have Corsair AX1200 and Seasonics ( X-1250 ) only so are they fine without that dummy load on the 5V rail then ?

Thanks !

Yes.  Both of them are rail to rail designs.  I ran a similar (older version) of the X-1250 without anything but 12V PCIe connectors loaded for 6 months and it ran cool and quiet even at 900W+ load.

you guys are still over thinking this.

use a paperclip to short out the plug.  leave the psu on unless your moving cards around (or need it off for some other reason)

mine!

What happens if the paper clip falls out? </troll>

Google is your (http://www.jonnyguru.com/modules.php?name=NDReviews&op=Story3&reid=189) friend (http://www.tweaktown.com/reviews/4508/seasonic_x_1050_1050w_power_supply_review/index5.html)

I'm linking to a Seasonic X-1050 review but it's built on the same platform as the X-1250. No minimum load required.
As to the AX1200, see the CrossLoad2 test results. Anything weird going on with 0 load at 3.3 and 5V? The answer is a resounding NO. You're good to go without messing around with resistors.

Just be sure to use a "thicker" paper clip and some electrical tape to hold it into place.  Kinda sucks when you come home from work only to find out that somehow (fan vibration?) the paper clip fell out on not one but 2 rigs.  :(

Nothing bad happened other than crashing and corrupting one of the usb drives but it kinda sucks to lose a day of mining that way.

A dab of hot glue and that paper clip ain't going anywhere any time soon.

I like the paperclip method but this makes the GPUs powered all the time.

What about when they are not mining and the rig is shutoff but the PSU still is feeding them power ( or maybe not ? ).

Most PSUs have an on/off switch, that overrides the paperclip. If not, you can always use an extension cable plug thingy (whatever you call that in english) with a switch.

a mining rig not mining?  lol

rear view of a 1250 and a 650,  there is another 1250 in the lower rack, the 650 is powering one card on each rig  10x7970's

(see the paperclip?)
other pics at the photobucket
http://i1027.photobucket.com/albums/y335/jjiimm_64/10x7970%20rig/IMG_0536.jpg

The GPU are only powered when the motherboard is powered on (among other things).

Just because a cable is connected doesn't mean power if flowing (think of light with a switch).

There is ONLY one potential issue, with relation to "plug-neutral"...

This "situation" only exists if you have one PSU plugged-in to one half of the circuit-hot-A, and the other PSU plugged-in to the other half of the circuit-hot-B. The circuit I am talking about is the two circuits which US power is split into. Usually, each room has multiple plugs, but all are either on circuit-half-A or B, not both in one room. However, if you had an "extra" leg of power added to a room, and that leg is the opposite half of the hot-circuit... and you plug one into one, and one into the other, and attempt to join them... there is a 50% chance that the unrectified connection will share a 220-240v potential between the PSU's, through the hot-legs. Since the "common" is grounded to "neutral", which is shared for both, since they are both mounted in the same case, and grounds for 12v are shared "dc common", throughout the motherboard.

This is NOT an issue if you are using one power-strip for both PSU's. But I imagine someone my hijack an additional outlet, for use in bitcoin mining, after having a second-leg circuit installed into a new room, along side of other equipment.

You can join any grounds, since they are all already joined by the case, the MOBO, and every other component. It is "better" to have the cases themselves grounded, as that stops the "noise" from traveling through the ground-plane in the MOBO, where the cases are not joined/grounded. If they were isolated, the cards would not function. Only optoisolators could be used to separate the two sources, which is NOT how any computer components function. (Only old parallel ports did that, where voltage-potential "leaking back in" would destroy the sensitive LPT chips.)

Voltage will never "FLOW" back from one PSU, into another, due to all the circuit designs which protect against that. You will simply get the average voltage potential out, and the combined amps out, for that average voltage.

12V + 10V = 11V out (On a starved PSU that is outputting 10V for whatever reason.) The combined amps would be 10A and 8A = 18A @ 11V. Thus, more stable than one PSU, which would be that one starved at 10V, which would drive-up the amps, and further kill the PSU or attached component. 2V will not be "flowing in" to the other PSU, unless there was no load attached, and unless the circuits had reached the reverse-voltage break-down limit, which would not happen with 10V on the other side and 2V "trying" to potential-in.

This is NOT against ATX V1, or V2 specs... this IS ATX specs. It is called tandem and redundant operation supplies of external sources.

http://www.procooling.com/index.php?func=articles&disp=52&pg=2 (http://www.procooling.com/index.php?func=articles&disp=52&pg=2)

ISAWHIM -Just to expand a bit, first a clarification - A and B are phases, not circuits. Phases represent the different angles that AC electricity is generated. Circuits refer to wiring segments with some kind of Over Current Protection like a breaker.  Otherwise you're explanation is correct.

 I'm only clarifying because as I understand it, you can connect different phase neutrals together but you're not supposed to connect different circuit neutrals if on the same phase.  In case anyone is wondering, flipping a breaker and checking what is off and on only indicates what is part of the same circuit, it tells nothing about the phases, which you would need a meter like a multimeter to check.  As you said most receptacles will be on the same circuit which means the same phase but on the off chance you have receptacles on different circuits (different breakers) there are 2 possible situations, neither of which are good.  One is the same phase thing I mentioned which is not good for the AC neutrals, the other is multiple phases where you now have a 220-240v potential between the AC hots as ISAWHIM indicated. Anyway that's where my knowledge stops!

As for the link, somewhere in there the author talks about the "grounding" within power supplies and what needs to be done in order to run them together although I don't remember what exactly was said, something about isolating the internals of the PSU from the PSU case.  Hope it helps.