Power requirement for Athlon-MP diskless nodes

[Robert G. Brown]

On Tue, 26 Jun 2001, Thomas R Boehme wrote:

> Hi,
>
> We are currently planning to buy a new Athlon-MP cluster
> I know AMD recommends a 460W power supply, but I am wondering if a less
> powerful power supply (e.g. 300W) would be enough for a diskless node. Any
> experiences or educated guesses?

The chips themselves run very hot -- as much as 72-80 Watts apiece.  So
you start off at (say) 150 Watts load when the system is running "just"
the CPUs.  Now you add in the motherboard, the memory, the fans, the
power supply waste, and the NIC(s) plus a surplus to facilitate the
switching and peak loads, and you're likely up there in the 200-250W
range.

At a guess 300W would therefore "probably" suffice to run a diskless
dual node, although AMD might be unhappy (enough to void the CPU
warranties?  enough to not "certify" correct operation?).  Still, their
460W probably assumes that the duals will be turned into loaded servers,
not stripped compute nodes.

I doubt that you'd actually break anything by trying -- one gets into
trouble with too much voltage/too much current but the most probable
consequence of too little is that the system won't work at all or won't
work correctly (without actually breaking).  If the trial system has
lm-sensors you can probably monitor the CPU core voltages and see if
they drop off too much (out of spec) under various loads.  Otherwise,
your real problem will be to "certify" correct operation -- if your
power supply is ever truly marginal things can fail without failing
catastrophically and your system will be not-often-fatally flaky as a
consequence.

Of course, trying is at your own risk.  But I think that the risk is
minimal -- I've inadvertently "tried" this a number of times in the past
with systems shipped with too small a power supply, with systems with a
failing power supply.  In almost all cases that did't involve a short
circuit or power surge, just putting in a new/adequate power supply
fixed everything.

    rgb

>
> Bye, Thommy
>
>
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
>

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu