1000w 5v psu for powering multi mini pcs?

[MuntyScruntfundle]

Hi there.

I've been looking for a while for a power solution to run 128 x 6w (Initial load) mini board computers.

I've looked at USB charging blocks that can provide 100w over 8 ports, that means I need 8 of them and splitting each port is a bugger for cabling.
I've looked at led power supplies but mostly they run to 30w until you go up to 12v.
I've looked at 1000/1200 PC psus, but I can't find one that supplies more than 30w on the 5v rail.

So, the need is to find, a 5v psu that can provide 1000w on a single rail. I'm not sure this exists.

All ideas and suggestions are welcome, but please bear in mind I've been searching for this solution for quite a few months now....

Many thanks.

 

[MuntyScruntfundle]

Sorry, I should have mentioned, I need 2.5a for each unit.

 

[alasdairp]

1000W @ 5V is 200A; 128 x 6W = 768W; 128 x 2.5A =320A=1600W
All those are different.
Even if we take your initial 1000W at 5V (200A) - that is an enormous current, especially at low voltage where you can't afford to drop more than about 0.2V - that would need remote sensing of the final terminal voltage, but you can't do that for 128 separated supply connections. You would need to run really thick copper or brass busbars and tap off for each ot the USB power feeds. It would not be easy to arrange the thin USB wires to connect properly and neatly to the busbars

If you really have to use USB connections for the power (a naff way of connecting so many miniPCs - that is why you can't find suitable off the shelf PSUs), then the 100W 8 port units look a possible solution. You should not split those USB ports to supply more than 8 as USB port slpitters will not allow (at all or reliably) those sort of currents to be split and pass through - you should get 16 of the PSUs and just run 8 PCs off each.

There are many other issues with running so many close together - not the least being how you communicate with them. USB will fall over. WiFi with each will give to a lot of data conflicts, slowness and unreliability as they are so close.

They really need to driven from a 128 way (or 2 x 64 way) powered managed data switch which would sort both the power and the data. However, such items are expensive (for a reason!). I wonder if you have ever done such a project before - it does not sound like it to me. Best of luck!

 

[Bob Geldoff1234]

Is this any good?
HRP-450-5 - http://www.bravoelectro.com/hrp-450-5.html

 

[MuntyScruntfundle]

I have run the entire system for some time from some fairly inferior usb power blocks, it's fine until I push the processors, then they start to overpower the blocks. I have purchased some new blocks which I've tested and work fine running 16 pis each, the internal quality is superb. However, this is a cabling mess.

The idea of taking the power from a single rail, large gauge copper wire to the underside of proto boards and dupont connections on top. This also works perfectly. and is partially employed at the moment, but it's a rats nest!

As for the network, 4x Netgear GS752STB managed stacked switches handle the ethernet and the wifi actually also works perfectly. These are good little machines, the Raspberry Pi.

Thanks Bob, that's an interesting unit, I can't find any for sale in the UK, I might have to bite the import tax bullet and get a couple, it would save an awful lot of mess.

Many thanks for the replies.

 

[Richard Burns]

Mouser seem to have several possibilities, however you will need to take great care in designing your distribution system to function effectively and safely.
Presumably these are ones you have already assessed and discounted, but I expect that would be the case for most standard solutions offered on this post.

 

[shaun1]

If you do go down the route of a single 1000w PSU then your main power rails will probably need to be somewhere around 50mm2 to handle that kind of current. You will need to calcukate this properly to make sure you dont get voltage drop (as mentioned earlier). You will then need to fuse down the individual tap offs to protect the thinner cable and the RPis, otherwise a fault could quite easily cause a fire. (200A DC is more current than you would get out yoir average welder, and it will only be limited by the volt drop on the cables)

 

[alasdairp]

I have run the entire system for some time from some fairly inferior usb power blocks, it's fine until I push the processors, then they start to overpower the blocks. I have purchased some new blocks which I've tested and work fine running 16 pis each, the internal quality is superb. However, this is a cabling mess.

The idea of taking the power from a single rail, large gauge copper wire to the underside of proto boards and dupont connections on top. This also works perfectly. and is partially employed at the moment, but it's a rats nest!

As for the network, 4x Netgear GS752STB managed stacked switches handle the ethernet and the wifi actually also works perfectly. These are good little machines, the Raspberry Pi.

Thanks Bob, that's an interesting unit, I can't find any for sale in the UK, I might have to bite the import tax bullet and get a couple, it would save an awful lot of mess.

Many thanks for the replies.

Mouser stock the HRP-450-5 for the UK - however, the PSU do require an external final terminal voltage sensor connection in oder to meet their spec. I do not see how you could effectively arrnange that for 64
individual connections. Good uck if you can. The separate response about fire risk is also very important. Your liability!

 

[Edtwozeronine]

I'm more interested in why you want to make a Pi network , are you going to remap the human genome?

 

[Lucien Nunes]

This is the sort of problem I often have to solve and I would strongly recommend using multiple, smaller PSUs. With many devices connected to the same DC rails, if there is any other grounded connection to the outside world, a fraction of your total PSU current tends to divert along those external grounds. One slip-up, one loose connection on your common rails that makes the main ground high-resistance and you might get a hefty chunk of that 200A flowing through the shields of HDMI and STP LAN cables etc., and you can kiss good bye to half the tracks on any associated PCB. Fuses won't save you here. I have seen expensive gear blown up by large common power supplies, and when working, problems with analogue interconnections (not sure if you have any of those) due to circulating ground currents.

It's not impossible to use a single PSU but you must watch these parallel ground paths carefully, many people overlook the more subtle implications, even commercial manufacturers.

 

[alasdairp]

Absolutely, Lucien, well explained! The replies above show who has done this sort of thing successfully and who has just theorised about it. I have seen many of such instances of blown pcb tracks and damaged connectors.

 

[Lucien Nunes]

A recent example I had was an audio mixer powered by its own linear regulated AC PSU. Between signal ground and mains earth is a typical ground-loop eliminator network, consisting of a pair of anti-parallel diodes, a 10R fusible resistor and a 100nF capacitor. As found, the diodes were S/C and the resistor was O/C.

The damage was not done by the mixer's own power, but from the nature of the damage I enquired whether it had been connected to any other equipment that had DC power distributed from a large central PSU, which was confirmed. Apparently there had been a brief S/C to one of its loads, from which the current returning to the PSU must have passed along a signal ground, the mixer, its mains earth and the 12V PSU mains earth, frying the mixer isolating network in the process.

 

[MuntyScruntfundle]

Thanks for all your feedback guys. Might sound silly, but I hadn't thought this would be so difficult to do.
Anyway, the solution has been to find 16 usb Charger units, each 10 port, each putting out 5v at 2.4a capable of pushing through 105w if needed, so I have a decent overhead. Each output on these is regulated so I can't burn out pis, in normal circumstances anyway. These are distributed along good quality rack power blocks, (the kettle lead type) then out to the mains. It works well. A benefit to the blocks being well regulated is I can ditch the usb connector and send the power directly to the gpio, this saves a load of space and crimping duponts is a lot easier than soldering a load of tiny usb plugs. When it's complete and flashing away I'll post a pi. I know we all like a flashing light!

The reason for doing this? Curiosity, turning to complexity, turning to insanity!! Seriously, to learn MPI on a reasonable scale without having to rent out super computers at huge cost. And when it's not crunching numbers it's going to churn out cgi renders.

As a side note, does anyone know if it's possible to get 26awg red/black (figure 8) silicone cable anywhere in the UK? Or anywhere? I can find 22awg figure 8, and I can find single strand 26awg, not both. Is handling the silicone at this smaller gauge too difficult? If you do know please pass on, I needs loads!!

Thanks again.

 

[MuntyScruntfundle]

There are other questions I haven't raised yet, which are, how many wall sockets should this be spread over? Should it have a separate ring from the fuze box? What rating should the trip be?

 

[UKMeterman]

A better idea would be mains to 24V over a couple of units and then some 24V to 5V DC DC converters.

 

[MuntyScruntfundle]

The problem with that approach is finding converters that can push through the required watts. A unit with an output of 60w will only supply 10 pis. So I still end up with 16 24-5 converters (with headroom), and they're not in a case or finished particularly nicely. So I'd actually end up with more spaghetti.

 

[RicardoF1RST]

Interesting thread. With the above in mind, would the below work?

This 1kw ATX PSU:

https://docs.rs-online.com/2c59/0900766b81697f03.pdf

Has a 5v rail supplying 25A, and a 12v rail supplying 83A. The 12v rail is 996w, so would it possible to drop the 12v down to 5v with DC to DC convertors to supply 5v 83A max 415w. Which would then provide a total of x32 5v/3A lines? (8 from the 5v rail and 24 from the 12v rail)?

But then I am guessing, 3A per 5v means 24 DC to DC convertors which defeats the simplicity objective.

Unless these are suitable:

20A, so 6x 3A each means 4 of these for x24 3A outputs...