PE1950 - EnergySmart - any disadvantages?

Kuba Ober kuba at mareimbrium.org
Sun Oct 21 02:36:56 CDT 2007


On Tuesday 16 October 2007, michalwd1979 wrote:
> Thanks for the info Faris.
>  From my test it turns out that the efficiency of PSU is rather poor. 1 PSU
> of dell2850 takes about 40-50W idle. In my configuration no loaded server
> uses 180W (3 SCSI drives 2x 3GHz Xeon, 4G RAM). Fully loaded it needs
> 240-250W. In fact I could save about 40W just by taking out one PSU. Some
> other problem is cooling system.  6 high speed fans can take 70-80W alone
> when running on full speed, much less when running on reduced speed. I am
> using water cooling and only one PSU so I saved about 100W (0.4A).  Now we
> are strongly thinking about making custom DC-DC PSU that will run directly
> form 24V battery, thus we will not need UPS, that alone can take some power
> just to covert energy 2 times. The problem is the connector to main board,
> I need to get documentation somewhere.

The power supply likely not only supplies the voltages, but also communicates 
with the motherboard somehow (about temperatures at least).

You may try to look at the pins with an oscilloscope.

As for DC-DC converters, you may end up having to do custom PCB design and 
whatnot: I presume that those supplies are pretty "stiff" on their 3.3V 
outputs, and with the power they transfer you will have to know what you're 
doing.

It's one thing to make a 1W DC-DC based on an application note. A PC supply 
like you want to make will take at least a well equipped lab and someone with 
some experience (or otherwise brilliant and really understanding physics). 
You'd want it to be CE-compliant as far as emissions go, at least. Not a 
small undertaking. As far as the server goes, with all the holes you've 
punched through the case, it probably already is not in compliance for 
emissions. Adding your own power supply will, at first try, likely drown out 
most of the long, medium and shortwave radio in the neigboring apartments. 
Not a nice thing to do. Power-efficient DC-DC converters run at lower 
frequencies (say 50kHz) and they radiate like hell by default ;)

You may want to look into the power supply and find out whether it has a PFC 
(power factor corrector). If it has, and it's a separate block, then the 
DC-DC is designed to run off 300V and isn't that easy to interface to. If 
there's no PFC, you will be able to power it with 100V DC going directly to 
the far side of the rectifier -- those supplies take AC, rectify, and then do 
DC-DC.

If there's a PFC, *and* you'd be brave, you can always rewind the primary on 
the high-frequency transformer in the DC-DC converter with 1/3 of the turns 
of a slightly thicker wire, so that the DC-DC would run on 100V DC instead of 
300V. A PFC is a step-up converter, and its output will be in 280-350V range, 
depending on how it was designed. Getting rid of PFC and doing a DC supply 
will save you 10-15% on efficiency (assuming your DC source is 100% 
efficient). Measure it all first, of course.

You may just as well use a 9 x 12V batteries connected in series. They can be 
on-line-charged using regular, line-frequency transformer-based power 
supplies, as those will by default have a primary-to-secondary insulation 
that can take a 100V offset. Not pretty, but simple, and way simpler than 
making your own supply from scratch.

By the way, the power supply should take ~120V DC without any modifications.  
Oh, the joys of switching power supplies :)

Cheers, Kuba



More information about the Linux-PowerEdge mailing list