DC power to a computer

In article , snipped-for-privacy@ipal.net (known to some as snipped-for-privacy@ipal.net) scribed...

Kind of depends on the environment. For mobile (vehicle) usage, it needs to handle a range, and have some pretty serious spike and overvoltage protection on the input.

For a fixed environment, such as a telecomm shelter, where you already have a well-regulated source of either -48 or -24V, it's not as much of a factor.

A mainboard engineer or manufacturer could answer that better than I could. I will say that point-of-load regulation, especially with modern IC's and components, can reach efficiencies in the very high

90's.

The DC/DC supply I installed in my mobile computer is a single PC board about seven inches long by about three or so wide, and it needs less than an inch of clearance all the way around for ventilation. Considering that DC/DC supplies don't need the big input caps and inductors found on the AC/DC supplies, the answer is "Yes," the DC/DC side can be quite a bit smaller.

Been going on for years. Decades, even. Google for "PIC-MG" and "single-board computer" and you'll get more hits than you'll know what to do with. Heck, I've got a single-board in my mobile system. It plugs into a PCI/ISA combo passive backplane.

Happy searching.

There are already many computers that have 12 volt DC power supplies for vehicle operation. There are also much pricier -48VDC power supplies intended for the telecom market. Pricier because they are generally built to higher standards, and sometimes incorporate N+1 redundancies for better uptime. These both are niche markets, and so higher costs and they will not be stocked at Best Buy.

--Dale

On 17 Dec 2006 18:48:29 GMT, snipped-for-privacy@ipal.net Gave us:

The best way is to buy a PC power supply that has multiple source inputs on it, which already exist.

There is a supply that accepts 28V DC, 95 - 265 VAC 50 - 60 Hz, as well as an internal case battery. It autoswitches between the three and all three are always hooked up. I think it is a 450 Watt supply, but you would have to buy a shitload of them before they'd even think about waking that line up.

It also has one rail that remains up long enough for a dump to occur on certain systems.

On Mon, 18 Dec 2006 00:23:01 GMT, Dale Farmer Gave us:

Typically, such specialized supplies are not stocked at all. One must place an order for many and await the production cycle completion to get any kind of decent price. Begging a single unit from anywhere is asking for a reaming on price. The N+1 is popular though, so some iterations of that are on shelves, but as you say his requisite is specialized.

The intent of the study was to improve the power utilization in the server room. A good part of power usage is for air conditioning. It's not only power supply efficiency that must be considered, but the ease and cost of removing heat from where it's generated.

The standard PC supply's switching supply works at ~340 volts, (240 x

1.414213562...), because that is what you get rectifying 240V or doubling from 120V.

If one were to produce the DC at a common point, and route it to all the individual servers, there is no reason why that voltage should be sacrosanct. A higher voltage would be slightly more efficient, and it could be regulated at the common origin.

On the other hand, standard PC supplies could be modified fairly easily to accept 340V DC input from a common source.

It now becomes an economic study which must include installation ROI, compliance with existing electrical codes, as well as overall power/cooling trade offs.

It is getting to be a big enough consideration for people like Google, that some solution will be forthcoming.

On Mon, 18 Dec 2006 01:18:17 GMT VWWall wrote: | snipped-for-privacy@ipal.net wrote: |> In another thread the topic shifted to DC input power to a computer. And |> this is not a new topic, either. I thought I'd just launch this directly |> as a new thread to see what others think about this. |> |> Here is the original paper from Google: |>

| | The intent of the study was to improve the power utilization in the | server room. A good part of power usage is for air conditioning. It's | not only power supply efficiency that must be considered, but the ease | and cost of removing heat from where it's generated. | | The standard PC supply's switching supply works at ~340 volts, (240 x | 1.414213562...), because that is what you get rectifying 240V or | doubling from 120V. | | If one were to produce the DC at a common point, and route it to all the | individual servers, there is no reason why that voltage should be | sacrosanct. A higher voltage would be slightly more efficient, and it | could be regulated at the common origin. | | On the other hand, standard PC supplies could be modified fairly easily | to accept 340V DC input from a common source.

Is that the voltage you would suggest be going into the mainboards on-board at-the-load regulators? Seems rather high for that role, especially for machines with exteranl power supplies (e.g. wall warts for single PCs, and a common power supply feeding a whole rack cabinet full of rows of blades).

The idea I have in mind is one single standardized voltage for all parts inside a computer case, especially the mainboard. Some machines would have a PSU to convert the mains AC voltage to this standard DC voltage. Other machines would have a direct DC connection for integrating into an array of machines served by a common PSU. Both could have the same mainboard.

| It now becomes an economic study which must include installation ROI, | compliance with existing electrical codes, as well as overall | power/cooling trade offs.

Building power would stay the same (large data centers might be doing this with a higher AC voltage like maybe 480Y/277 in the USA). The direct-to-board DC power would usuaully span no more than a rack cabinet or three; not a ceoncern of building electrical codes (though OSHA may matter for safety reasons if this is a high voltage or has large fault currents).

If 48VDC turns out to be a practical voltage for direct-to-mainboard, which could be made relatively safe in small home use (think wall wart cable), it would certainly be a boon to telco, and the subset of IT that is already using telco style power. It would not be a hard shift for the industry to do. The big question is: is this the right voltage to make mainboards work with directly? I think that choice needs to be made with concern for various things like cost and efficiency. I could personally deal with anything from 12VDC to 48VDC. But something other than 48VDC would add cost to the telco aspects. And something other than 12VDC with a wide range would add costs to the consumer market. If there is to be just one DC voltage, where the other has to convert, which should it be? How costly is a 12VDC to 48VDC converter at 100,

200, or 300 watts? Or what about the reverse (48VDC to 12VDC)?

| It is getting to be a big enough consideration for people like Google, | that some solution will be forthcoming.

With 450,000 machines as of some time ago, they have a huge interest, and hopefully a huge influence. They can get such boards now, but they have to depend on their own buying power to get a decent price. But if the whole industry moved in that direction, they would benefit even more because all the basic components to do this would become more massively produced, better integrated, and have an even lower price ... the difference between a 1,000,000 unit order Google could potentially place, and a 1,000,000,000+ unit market looking forward.

Phil, they call those laptops. Virtially every mobile PC I have seen, from the old IBM convertible to the latest shiny P4 uses a single ended DC supply. Any further slicing and dicing of the voltages is internal.

On Mon, 18 Dec 2006 01:18:17 GMT, VWWall Gave us:

You mean from where it gets manipulated and dissipated. It gets generated at the power station.

Most of our switchers, including our PC and Server supplies were using 400 Volt rails. They also did so from 90 to 265 volts in, so the input voltage has nothing to do with the switcher rail voltage designed into a supply.

Heavy current low voltage DC distribution systems do have their problems.

That would qualify as a high voltage distribution system. Not a very easy thing pull off. All of the supplies are custom at that point.

I think you need to rethink where the heat gets created at. Efficient power supplies do not make much heat. The heat is made in the CPUs and other power consuming devices in each server.

Their server rooms are working fine. going from open racks to closed racks will solve any thermal issues that might be at hand today.

The *heat* is generated in the server room. The power which produces the heat is generated at the power station.

Some are considering local power generation where a cheap source of a fuel like natural gas is available. This solves dependence on local utility company power. Google picked their sites where cheap power was available.

The switching PS then has to accept the mains voltage variation as well as regulate the voltage over the load variation.

The input section of the supply rectifies and converts the mains voltage to the switcher rail voltage.

You'll find that the power lost varies with the input voltage. A few watts from each server PS can add up. They would be more efficient running directly from a regulated 400V DC input.

Having to accept that input range, they picked the most efficient switching voltage. I'm not surprised that it came out somewhat above the common mains derived, unregulated, ~340V.

Precisely, which is why I made the following points:

Having the power conditioning in a central location,would increase the efficiency v.s. having each PS take whatever is thrown at it. Every watt counts! Although "custom", the power supplies would be cheaper to build. An interim modification to existing units adds only a jumper and a change to the input power connector. Code compliance could be a problem!

Why do you think they originated the study? Cooling is now about one third of the total power usage. Power cost is a large part of doing business. Every watt saved goes directly to profit!

It sounds like the idea behind the 340V DC distribution was that a simple modification that could be done to any PC power supply could eliminate the AC components of the power supply. Since that would consist of just a rectifier/voltage doubler, I think that it would hardly save any heat generation at all. A few centralized 340V DC power supplies, if well designed (that is, not just a simple rectifier), would present a much nicer load to the power company for sites where that matters, as well as eliminating the overloaded neutrals that lots of switching power supplies on a three phase system often have.

Actually, the overloaded neutral problem could be solved if computers designed to use 100-260V supplies (that is, most of them) were connected phase to phase getting 208 volts rather than phase to neutral 120V. No neutral current at all, and the current spikes on the hots would be two smaller spikes rather than one large one.

On Mon, 18 Dec 2006 16:49:48 GMT, VWWall Gave us:

The new trend is entire high rise buildings that have many self sufficiencies and such in them.

On Mon, 18 Dec 2006 16:49:48 GMT, VWWall Gave us:

I have an idea. Use 6 inch masts that have a central "flue" to the roof. Use a 48 volt rail supplied from a battery room next door. Each mast has ultra clean, noise free linear supplies for each computer voltage requisite. Those linears mount inside the masts and vent their heat and EMI noise up the flue. The computer banks get fed, and the MOBO local voltage regulation circuits still do their thing. All are happy. The computer room environmental stabilization costs are cut by two thirds.

All heat is then strictly computational, and the hard drive spindle motors.

My idea. Of course this forces all computer server rooms to within two floors of a rooftop. Actually not if one forced air through the masts.

V I O L A ! :-]

On Mon, 18 Dec 2006 00:22:04 -0500 snipped-for-privacy@aol.com wrote: | On 18 Dec 2006 02:24:53 GMT, snipped-for-privacy@ipal.net wrote: | |>The idea I have in mind is one single standardized voltage for all parts |>inside a computer case, especially the mainboard. Some machines would |>have a PSU to convert the mains AC voltage to this standard DC voltage. |>Other machines would have a direct DC connection for integrating into an |>array of machines served by a common PSU. Both could have the same |>mainboard. | | | Phil, they call those laptops. Virtially every mobile PC I have seen, | from the old IBM convertible to the latest shiny P4 uses a single | ended DC supply. Any further slicing and dicing of the voltages is | internal.

But that's more like a PSU that takes DC instead of AC. My quest is for the optimal direct-to-mainboard single voltage that can be provided by PSU, wall wart, or battery, either inside a single computer, or to multiple boards in a rack, cabinet of racks or maybe even a few cabinets. Figure the same boards would be designed for universal use in all these places.

| It sounds like the idea behind the 340V DC distribution was that a simple | modification that could be done to any PC power supply could eliminate the | AC components of the power supply. Since that would consist of just a | rectifier/voltage doubler, I think that it would hardly save any heat | generation at all. A few centralized 340V DC power supplies, if well | designed (that is, not just a simple rectifier), would present a much | nicer load to the power company for sites where that matters, as well as | eliminating the overloaded neutrals that lots of switching power supplies | on a three phase system often have.

So do you have any problem with a wall wart putting out 340VDC to a connector that plugs into a tiny computer and is directly fed to the mainboard? I suspect someone would object to such a voltage for reasons of safety.

| Actually, the overloaded neutral problem could be solved if computers | designed to use 100-260V supplies (that is, most of them) were connected | phase to phase getting 208 volts rather than phase to neutral 120V. | No neutral current at all, and the current spikes on the hots would be | two smaller spikes rather than one large one.

That just pushes the problem further upstream if the 208 comes from a D-Y transformer. A large data center using single voltage boards (like Google apparently plans to do) could power a few hundred boards in a few cabinets from a single large PSU fed directly from 480VAC in three phases.

| You'll find that the power lost varies with the input voltage. A few | watts from each server PS can add up. They would be more efficient | running directly from a regulated 400V DC input. | | Having to accept that input range, they picked the most efficient | switching voltage. I'm not surprised that it came out somewhat above | the common mains derived, unregulated, ~340V.

Is this a voltage you would want going directly to a computer mainboard from conversion sources such as a wall wart with detachable cord, or a battery?

| Why do you think they originated the study? Cooling is now about one | third of the total power usage. Power cost is a large part of doing | business. Every watt saved goes directly to profit!

My question is, is it possible or practical to make all boards accept a single incoming DC voltage regardless of whether that voltage comes from a central PSU for 100's of boards in a data center, or a wall wart for a small home computer? And what would that standardized voltage be?

First, such a voltage isn't an input to the motherboard in what I described. It would be an input to a standard (minus the powerline AC-DC conversion) power supply.

The advantage would be the centralized power supplies would present a much nicer load to the power company, if designed properly. There would also be a minimal (two diode voltage drops times input current) heat reduction since the centralized power supply shouldn't need computer room AC. There is no advantage to having 340V wall warts supplying power to home PCs, this system would only be useful for large herds of computers.

That idea isn't intended to help with harmonic generation as far as what the power company sees. It neither helps it nor makes it worse. It just decreases neutral current on bus bars and the transformer neutral connection, to zero if everything used this.

The voltage the motherboard uses depends on the what the logic on the board needs. I believe the largest current consumption in a processor chip is the repeated charging and discharging of the capacitance of the internals (in particular the gates) of CMOS circuitry. To minimize heat dissipation, you'd want to minimize voltage. This is why, I believe, processors went from using +5V to +3.3V quite some time ago, and supplying a processor with +340V would be absurd.

On Tue, 19 Dec 2006 14:59:33 +0000 (UTC) Michael Moroney wrote: | snipped-for-privacy@ipal.net writes: | |>On Mon, 18 Dec 2006 19:00:14 +0000 (UTC) Michael Moroney wrote: | |>| It sounds like the idea behind the 340V DC distribution was that a simple |>| modification that could be done to any PC power supply could eliminate the |>| AC components of the power supply. Since that would consist of just a |>| rectifier/voltage doubler, I think that it would hardly save any heat |>| generation at all. A few centralized 340V DC power supplies, if well |>| designed (that is, not just a simple rectifier), would present a much |>| nicer load to the power company for sites where that matters, as well as |>| eliminating the overloaded neutrals that lots of switching power supplies |>| on a three phase system often have. | |>So do you have any problem with a wall wart putting out 340VDC to a |>connector that plugs into a tiny computer and is directly fed to the |>mainboard? I suspect someone would object to such a voltage for |>reasons of safety. | | First, such a voltage isn't an input to the motherboard in what I described. | It would be an input to a standard (minus the powerline AC-DC conversion) | power supply.

The thing I'm looking for _is_ the motherboard input single voltage that would be the best to choose to be standard based on whatever factors do matter (including safety of home users, energy efficiency, and code compliance for data centers that might supply the voltage as part of the infrastructure).

| The advantage would be the centralized power supplies would present a much | nicer load to the power company, if designed properly. There would also | be a minimal (two diode voltage drops times input current) heat reduction | since the centralized power supply shouldn't need computer room AC. | There is no advantage to having 340V wall warts supplying power to home | PCs, this system would only be useful for large herds of computers.

Basically you're saying (as a separate issue) it would be good to have a centralized DC power that doesn't have to be converted. Since the internal rails of a PSU are around 340VDC, by using just that voltage for the big infrastructure, you can eliminate half of the PSU.

But I think the objective is to get the motherboard input down to ONE voltage. I suspect that voltage would between 12VDC and 48VDC and perhaps optimally be 18VDC, though 12VDC would certainly have many conveniences. But anything less than 48VDC would be problematic for a data center wanting to centralize the DC power telco-style (due to such high amperage and massive conductors or bus bars). So how hard would it be to make a simple PSU-like unit that steps 340VDC down to

12VDC or 18VDC or such? Could that be done in 1/2 or less space of a comparably power AC mains to multi-voltage DC PSU of today?

|>| Actually, the overloaded neutral problem could be solved if computers |>| designed to use 100-260V supplies (that is, most of them) were connected |>| phase to phase getting 208 volts rather than phase to neutral 120V. |>| No neutral current at all, and the current spikes on the hots would be |>| two smaller spikes rather than one large one. | |>That just pushes the problem further upstream if the 208 comes from a |>D-Y transformer. A large data center using single voltage boards (like |>Google apparently plans to do) could power a few hundred boards in a |>few cabinets from a single large PSU fed directly from 480VAC in three |>phases. | | That idea isn't intended to help with harmonic generation as far as what | the power company sees. It neither helps it nor makes it worse. It just | decreases neutral current on bus bars and the transformer neutral | connection, to zero if everything used this.

If the building is supplied with 480Y/277 and that's stepped down to

208Y/120 with a delta-wye transformer, harmonic loads on the 208 L-L will be passed back to the MV to 480Y/277 transformer as L-N currents. But if you're going to have a central DC system supplying 340VDC, you could perhaps run that from 480 L-L directly? Or maybe do the 480 to 208Y/120 as a wye-wye?

On Tue, 19 Dec 2006 15:06:47 +0000 (UTC) Michael Moroney wrote: | snipped-for-privacy@ipal.net writes: | |>On Mon, 18 Dec 2006 16:49:48 GMT VWWall wrote: | |>| You'll find that the power lost varies with the input voltage. A few |>| watts from each server PS can add up. They would be more efficient |>| running directly from a regulated 400V DC input. |>| |>| Having to accept that input range, they picked the most efficient |>| switching voltage. I'm not surprised that it came out somewhat above |>| the common mains derived, unregulated, ~340V. | |>Is this a voltage you would want going directly to a computer mainboard |>from conversion sources such as a wall wart with detachable cord, or a |>battery? | | The voltage the motherboard uses depends on the what the logic on the | board needs. I believe the largest current consumption in a processor chip | is the repeated charging and discharging of the capacitance of the internals | (in particular the gates) of CMOS circuitry. To minimize heat dissipation, | you'd want to minimize voltage. This is why, I believe, processors went | from using +5V to +3.3V quite some time ago, and supplying a processor | with +340V would be absurd.

Yes, I think 340VDC direct to the board is absurd. OTOH, I suppose I could use that for my ham transmitter :-)

And those voltages are even lower today, and nothing particularly standard. It's argued that it is best to let the motherboard host the appropriate voltage regulators to get the voltage it needs and derive that all from a single input DC, such as 12VDC (or maybe 18VDC if more is needed to get a well regulated 12VDC within the board).