What wattage does an atx motherboard draw?

[JonJ678]

I've been reading through formfactors and can't get very far with this, so I thought I'd ask ocuk. I'm told I phrase things bizarrely, so I'll summarise the question: Could I run an overclocked i7 motherboard using a "pico psu", with the 8 pin atx and pci-e cables coming from a different (not pico) atx psu?

A common atx board has an 8 pin atx cable and a 24 pin one. It also supplies current to expansion cards, iirc a pci-e x16 slot is rated for 75W, and to peripherals. An overclocked i7 processor is generally considered to draw around 200W. However some fraction of this is through the 8 pin cable, and it's unlikely that a graphics card will draw 100% of 75W from the board when it has its own cables as well. What current does the 24 pin cable need to supply (estimates or a function of installed cards would be good)? My guess is 100W + 70W per pci card but I base this on nothing whatsoever.

The vast majority of current drawn by a computer is at 12V, however some is at various lower voltages. A pico psu takes 12V dc, has a stab at cleaning it up, and connects directly to the 20/24 pin socket which is the only place lower voltages are applied. I'm pretty certain, though I haven't checked, that the pico psu will run from any 12V source with enough current behind it. Say, a spare pci-e line.

It may then be possible to run an entire desktop computer from a single (high current) 12V line rather than from mains AC. This could be useful. At least, I can imagine uses for it.

I suspect I need to buy a multimeter and measure the currents running through each wire myself, but I also suspect this will be tricky as the only approach I know of is connecting an ammeter in series, so cutting a wire, measuring, attaching it back together. Any and all thoughts welcome.

Cheers

 

[weird_dave]

Why would you want to use a pico psu and an atx supply?
It might work if your gfx has it's own power cables, I just can't see the point.
There's also the pain of wiring up the atx supply to turn on when the pico does.

 

[JonJ678]

What I have in mind is an atx psu in one box and motherboard/cpu/ram/gfx/hdd in a second. The objective is to seal the motherboard box, leaving me with an air tight box with water cooling pipes running in and out and a single 12V cable running in. That the 12V source in my case is an atx psu isn't the point really, this should allow a meanwell to be used instead, or any 12V source. I believe boats run off 12V for example.

I also think it might work, the people who make pico psu will probably get back to me. I'm fairly sure I can combine the 8 pin and peg cables into a single line then split it up again later, it's just 12V after all.

Cheers

edit: in case it's not clear, an airtight box (with two water pipes, a 12V line and an ethernet cable running out) can be put anywhere, hidden in a cupboard say. It can also be run below ambient temps with no fear of condensation, once you've bothered to fill it with CO2. Moving the psu to a different box simplifies this.

 

[weird_dave]

You could run a single thick pair of wires to the sealed box then split from there, but you can't reliably join 12V sources together on the outside, one source will tend to try supplying all the current.
I wouldn't try running a PC from a set of batteries being charged via an engine, you'll get all sorts of nasty spikes appearing which may harm some components. Also, it's only 12V in the discharged state, fuilly charged it's more like 13.8V. I suspect that would work fine for most system components though.

As a side note, I really wish 'they' would come up with a new spec for components so that they ran off a single voltage (possibly 24V or more to keep the current down) and just generated what they need locally. Cabling would be simpler, psus would be cheaper and more reliable. The mobo and gfx do this anyway these days so it's no big leap.

 

[hooFlungDung]

 

[JonJ678]

The intention is the 12V line comes in from some source, goes direct to pci-e cards and the 8 pin atx plug(s), and is converted by the pico psu to supply the motherboard and a hard drive or two. Water cooling (ambient or chilled) keeps temperatures under control, and the system should run without any need for maintenance/cleaning indefinitely.

Combining 12V supplies into one is beyond my knowledge although I believe it can be done. My atx one will supply 64A at 12V continuous, 70A peak so I can't imagine needing more unless running pelts in which case one supply running tecs, the other running system won't cause any grounding issues. I much appreciate the warning though.

A car battery kicks out anything from 5 to 15V or so iirc, so doesn't really count as regulated dc. I'm sure there's at least a few sources of steady 12V supply available, but I admit I'm not especially interested in them myself.

Regarding the voltage, which components do you have in mind? The motherboard needs to be supplied with 12V, 5V and 3.3V (& -12, perhaps not -5) at present, so running everything off 12V would mean circuitry on most components to step it down. This is more expensive than generating the voltages in the psu, as then you only need to do it once. Or am I missing something?

Useful post Dung, do come back soon.

Pico psu spec isn't looking good.

Pico psu specifications
120W version
3.3V 6A load
5V 6A load
12V 7A load (I think this is passed straight through, only change for 150W version is 8A here)

It also states

Precautions for operating this DC-DC converter:
-For fanless operation de-rate the output of the 3.3 and 5V rails by ~35% or
ensure PSU surface temperature should not exceed 65C, whichever comes first.
-Combined and sustained output should not exceed 65% of total power or or
ensure PSU surface temperature should not exceed 65C, whichever comes first.

which sounds suspiciously like "our psu can't actually output its rated spec".

Some testing by spcr suggests an oldish 775 system will draw 3A+ on each of the 3.3, 5V rails. It also shows a sharp increase in load on the 12V rail (from the 20 pin atx) when the cpu is under load.

Not feeling optimistic right now, but perhaps the makers of pico psu will have good news.

 

[SMAndy]

I'd say don't do this. That Pico PSU is designed to be used to power a very low power PC, not as an addition to an ATX power supply. The problem here comes from separate power coming from the ATX, as mentioned above, one will try to supply all the current.

There is also the added problem that you'd have two different Earths on this. One from the Pico, and one from the ATX. The reason we normally use just one power supply is to keep the earth at a steady level for everything, which gives a single reference point to go from everything else. - Also, there is case contact between the PSU and Case, meaning any stray current that gets sent through the case gets earthed to the PSU.

If we say the ATX PSU gives an earth at actually 0v (a comparison point) you can't guarantee that the pico will do the same. There could be 5v differences either way (an example. Could be as little as 0.001v, with no maximum.) If the atx power source ends up going to a pico earth, you could gain or lose 5v on either. Yes, this does mean a +3.3v input could end up as -1.7. It's highly inadvisable to mix two power supplies because of this.

 

[weird_dave]

The intention is the 12V line comes in from some source, goes direct to pci-e cards and the 8 pin atx plug(s), and is converted by the pico psu to supply the motherboard and a hard drive or two. Water cooling (ambient or chilled) keeps temperatures under control, and the system should run without any need for maintenance/cleaning indefinitely.

What about the warm motherboard components and the pico? That heat still needs to escape. Farnell sell heatpipes, you could use the outer case as a heatsink and connect via the heatpipes. Could all get a bit fiddley.
Don't most gfx cards exhaust air out, this may make it tricky to make air tight unless you're watercooling this as well.

Combining 12V supplies into one is beyond my knowledge although I believe it can be done. My atx one will supply 64A at 12V continuous, 70A peak so I can't imagine needing more unless running pelts in which case one supply running tecs, the other running system won't cause any grounding issues. I much appreciate the warning though.

That won't be 64A from one supply rail, most likely 4.

Regarding the voltage, which components do you have in mind? The motherboard needs to be supplied with 12V, 5V and 3.3V (& -12, perhaps not -5) at present, so running everything off 12V would mean circuitry on most components to step it down. This is more expensive than generating the voltages in the psu, as then you only need to do it once. Or am I missing something?

Most of the devices on the motherboard don't run from any of the standard PSU outputs, they are derived, mostly from 12V. For the few items that really do use 3.3V and 5V on the mobo, the cost of adding a small switcher is pence (same for drives). PSUs currently have much more 5V and 3.3V capacity than is needed, by eliminating these supplies and just having one, the PSU becomes much simpler to design. But, I don't think 12V is the answer, 24V would be better since a 500W system would only require 21A, making the PSU cheaper yet again. But this is a complete digression from your original post

Pico psu spec isn't looking good.

If you can get the heat away it might work. Might be simpler to find a fanless PSU and build an airtight case with lots of heatsinks on the outside.....

Don't worry about different earths, the pico 0V will be at the same potential as the PSU 0V as they are physically joined together. Plenty of people use multiple supplies with no problems (not just for PCs either).

 

[JonJ678]

@Andy I don't believe these are problems with this idea. The pico psu may be incapable of supplying the currents needed to the board, but the majority of the power used comes from the 8 pin atx line and the peg cables. So it *might* cope, need a clamp meter really to be sure. Common earth is present, the pico passes the 12V, 0V lines unchanged as far as I can tell and derives the 5,3.3 as a potential upwards from this 0 (easily checked once I have a multimeter and a pico). I'm pretty certain the pico is just a couple of step down voltage regulators, so it's not really the same as wiring two atx psu's together.

Specifically, the 12V line isn't going to supply current directly on 3.3 or 5V, and the pico only passes 12V through so the current source is the same as if it was connected directly, so the two shouldn't fight. *fingers crossed*.

@Dave I don't trust or understand heatpipes, but I should be able to water cool most things. The plan would be to put cpu/gpu/mosfets/chipset(s)/ram under water, and if need be mount a radiator in the air tight box with a fan. This last will keep ambient temps in the box equal to that of the water +/- a bit, sinking heat into the radiator instead of the normal approach of radiating heat using it.

It definitely is 12V 64A on one rail, and to be honest I believe it'll output this for half a decade without complaint. Spec here. I need to open it up to see how the cables are soldered, if the 8 pin & peg clearly come from one mass of solder I'll be a lot more confident combining them into a single cable. However you're spot on in saying that it should be across four rails, as the atx spec says 20A limits per rail are "safer" so a lot of supplies take this approach.

I'm very interested in your next point, I think it's exactly what this thread is about and not much of a digression. Do you have any sources for what sort of currents the motherboard draws on the 3.3,5V lines? My particular psu is rated for over 20A on each of them which is surely overkill, but the pico is good for at most 6A which may not be, e.g. ram seems to draw from 3.3V and I'm using six sticks of it.

Cooling the pico would be important it seems. It will also be sufficiently difficult that I'm inclined to look for a separate dc-dc board, preferably with higher current capacity on the 3.3,5V rails, as this should be easier to cool. I'm not impressed that it states for 24/7 use it can only output 65% of its rated 6A, unless I'm misreading this it's a deliberate lie on the specifications.

Cheers

 

[weird_dave]

@Dave I don't trust or understand heatpipes, but I should be able to water cool most things. The plan would be to put cpu/gpu/mosfets/chipset(s)/ram under water, and if need be mount a radiator in the air tight box with a fan. This last will keep ambient temps in the box equal to that of the water +/- a bit, sinking heat into the radiator instead of the normal approach of radiating heat using it.

A radiator in a sealed box just isn't going to work, you're going to need airflow somewhere in the system.

It definitely is 12V 64A on one rail, and to be honest I believe it'll output this for half a decade without complaint.

One rail makes life easier
The danger with having a 64A capable rail is that the cables can't take it if there's a short, unless they've added protection for each set of outputs.

I'm very interested in your next point, I think it's exactly what this thread is about and not much of a digression. Do you have any sources for what sort of currents the motherboard draws on the 3.3,5V lines? My particular psu is rated for over 20A on each of them which is surely overkill, but the pico is good for at most 6A which may not be, e.g. ram seems to draw from 3.3V and I'm using six sticks of it.

This is why I wished they'd run everything off one rail, RAM uses a user definable voltage anyway, why use the 3.3V to step down from? Every setup will be different so it's a case of getting hold of a current clamp.

Cooling the pico would be important it seems. It will also be sufficiently difficult that I'm inclined to look for a separate dc-dc board, preferably with higher current capacity on the 3.3,5V rails, as this should be easier to cool. I'm not impressed that it states for 24/7 use it can only output 65% of its rated 6A, unless I'm misreading this it's a deliberate lie on the specifications.
Cheers

These would take care of your 5V and 3.3V:
http://uk.farnell.com/murata-power-.../converter-dc-dc-10a-12v-variablel/dp/1694771
You'd just have to sprinkle a few cheap components around them and mount them safely.

One other issue with all this, the PSU power on pin will need to be fiddled with. Not entirely sure how everything will behave with all the supplies on before you push the power button on the PC. You could use a bunch of relays or FETs to isolate all the supplies apart from the +5V standby.
Might be easier to put a connector on the box for the power on and psu OK signals and pass them through.
In fact, why not put connectors on the box for all the supplies and just use the ATX as normal?

 

[JonJ678]

I suspect I described the radiator poorly. Imagine a cube, about 10" to a side, with a 120mm radiator with fan hanging vaguely in the middle. Pump water at 30 centigrade through the radiator from outside the box. If the box is under 30, this will warm the box up to 30. However if the box is over 30, heat will move into the radiator bringing the temperature back down to 30. Airflow is provided by the fan, the air in the sealed box moves around in vaguely circular fashion. If this seems pointless, consider pumping water at -30 through the radiator.

Good points on the other two, and I think the link is a good idea. Cheaper than the pico, will definitely output the required current. The last paragraph has given me pause. I can't remember the reason I didn't want to use heavy guage extension cables. If passing the power on cable (shielded) up alongside the big 12V/0V lines, and presumably the -12V and standby cables as well, then it feels like I may as well just use a very long 24 pin atx extension cable with sufficient gauge that voltage drop isn't much of a concern. I think this is what you're getting at in the last line.

If instead I take a load of relatively thick copper cables, bundle them together and feed them down 1/2" tygon sealing will still be simple enough and the extra dc boards aren't needed. Need to go back a few steps and work out what I was originally trying to achieve I think.

 

[weird_dave]

I'm still not understanding your cooling solution, any chance of a sketch?

Yes, I was thinking of some form of ATX extension cable, you could make one from an old mobo and PSU. It would only need to be long enough to get to the side of your sealed PC case, then just plug your PSU into it.

I forgot to find a solution for the -5V and -12V, but since they are barely used, these will be cheap. I suspect the PC would actually run fine without them, possibly the sound may not function correctly depending on the implementation.

Going back to your original idea of running a PC off any 12V source, I can see problems. The only capable source I can find that's stable is an ATX PSU! Batteries would need regulation which could get expensive and ideally you'd need 36V minimum for most capable DC-DC converters.

Unless of course, someone runs their mobo/gfx at 14-15V and see if it behaves, then you'd get away with a battery as long as it wasn't being charged by an engine. I strongly suspect you can run a mobo at higher than 12V, but what the limit is would be down to the components on the mobo. Without circuit diagrams showing what connects to 12V, this would be a nightmare to work out by hand.