[Beowulf] Interesting google server design

Lux, James P james.p.lux at jpl.nasa.gov
Thu Apr 2 10:37:10 PDT 2009

> -----Original Message-----
> From: beowulf-bounces at beowulf.org 
> [mailto:beowulf-bounces at beowulf.org] On Behalf Of Greg Lindahl
> Sent: Thursday, April 02, 2009 9:31 AM
> To: beowulf at beowulf.org
> Subject: Re: [Beowulf] Interesting google server design
> On Thu, Apr 02, 2009 at 10:11:07AM -0400, Prentice Bisbal wrote:
> > Or it could be because they make motherboards that convert 
> 12 VDC to 5 
> > VDC on the motherboard.
> All Itanium and some other x86 boxes take a single 48V input 
> to the mobo. I talked to a mobo designer once and he claimed 
> that there was no power savings to be had doing this. Beats me, mon.
> -- greg

There might not be power savings, but there could be cost savings, or convenience.

Conventional power supplies in PCs work like this (as do in just about every other consumer electronics widget that consumes more than a few watts.)

Incoming line voltage is rectified and used to create a DC bus at about 380VDC (if your line voltage is 240ish, then it's straight rectification to a capacitor input filter, if 120ish, a voltage doubler).

The DC bus voltage (which is unregulated, of course, so it can vary some 20% up and down) is fed to a switch which runs at some tens of kHz. The squarewave output of the switch is fed to a multiwinding transformer, with output windings all chosen to produce the right output voltages (in terms of ratios).  ONE of those output voltages (5V or 3.3V) is used to generate the control signal to the switch to adjust the duty cycle to provide regulation. The rest all just follow along. The low voltage AC coming out of the transformer is rectified (using fast Schottky diodes, usually, although synchronous FETs are also used for lower loss) and low pass filtered (because it's at tens of kHz, the filter components can be physically small).  There is some voltage drop through the filter, but most computer stuff doesn't require rock solid regulation (e.g. if the 12V to the disk drive rises to 12.5V, that's not a big deal).

If you need voltages for the CPU or logic that are lower (like 1.8V or 2.4V or whatever) , that would be done onboard by another regulator (because the voltage drop in a 1.8V 20A line would be intolerable) close to the load.

The only "BIG" components in this whole thing is the filter capacitors in the input filter, which are working at line frequency, and have to store enough energy so that at full load, the range of duty cycles available in the switcher can still hold regulation.  (for a variety of reasons, the duty cycle can't go from 0 to 100%..and still keep decent efficiency)

There will also often be some sort of powerfactor correction on the input, either a passive filter (big, bulky, expensive) or some sort of active preregulator that sort of turns the line into a constant current source before feeding the input energy storage.  (a rectifier feeding a capacitor has a very peaky current waveform, particularly at light loads)

If you're running off 48VDC, the design is much the same.. Instead of designing the primary for a 400V, you design for 48V, so there are fewer turns on the transformer primary, and the switch device has to handle more current. 

If you're doing onboard DC/DC conversion from a 48V bus, then you'll have an line voltage to 48V power supply (same schemes as outlined above), and then a raft of smaller 48V to whatever converters. (which are actually quite easy to build and have high performance, because the input voltage is more stable (you're not trying to filter out the power supply ripple coming from the line voltage).  The range of duty cycles is smaller, you can probably run a higher switching rate (so you need less stored energy in the output filters, making them smaller)

There are significant regulatory advantages to 48V.. It's below 50V, which means that it's "low voltage"

If I were king of the world and could dictate designs, here's what I would do:

A big bulk 3 phase rectifier to make the 400V bus, which would be fairly well regulated.  A 3 phase rectifier even with NO filter has a lot less ripple than the single phase plus a capacitor in most power supplies. A decent 12pulse design would have a ripple of a few percent.  This makes the downstream converter's job MUCH easier.

The batteries hang off this bus at that voltage. (this is standard stuff these days, with grid-tie inverters, for instance). 

You distribute the 400VDC (or 350, or whatever it works out to be convenient) to all the chassis.  Inside the chassis, you have a simplified power supply that resembles the back half of a conventional power supply.  You're distributing the high power at high voltage, so IR losses in the distribution are less. (schemes which do stuff like distribute at 12V, or even 48V, have to deal with more IR loss).  Because you're feeding from an essentially regulated DC bus, the "per PC" power supply can be designed with higher efficiency over a narrower input voltage range (2-3% should be easy to achieve, with fairly conventional wiring techniques... Normal house wiring is designed to have no more than 2% voltage drop at the outlet.)

There are some problems: Multihundred volt DC has some safety aspects.  If it arcs, it is tougher to extinguish than AC. You need different kinds of switches, if you need to open under load.  But this is all pretty conventional stuff in the solar panel and grid tie inverter business these days.

For all I know, this is what they're doing.


More information about the Beowulf mailing list