Dual AMD systems in rackmount cases

Robert G. Brown rgb at phy.duke.edu
Mon Jun 24 16:05:21 EDT 2002


On Mon, 24 Jun 2002, Mark Hahn wrote:

> > KW/meter^3 seems unwise -- any sort of bobble in your cooling system
> > will rapidly cause your stack to reach furnace temperatures, and even
> 
> ah, well, I think you have to start with assuming *something* stays up.
> I'm assuming that I can shut down my space heaters if (when) our chilled
> water goes out (again)...

Assume makes an ass outta u and me (as my wife is fond of saying:-).

We learned a lot about the dual's (lack of) thermal tolerance when our
new room had such a bobble.  At least nothing broke permanently, as far
as we can tell...

We have a thermal kill switch for the room, but it is set pretty high as
powering down everything at once is not our most desirable solution to
cooling failure.

Once we got enough, cold enough, air flowing over our nodes we were able
to knock the case exhaust temperatures (and the sensors-reported CPU
temps) down quite a bit.  The air is downright cold going in, but if you
sit behind a rack to work you can avoid freezing although it isn't
really exactly warm back there.  The warm air in the exhaust vents
themselves is quickly diluted by the overall airflow pattern -- our
measurements of 75F are a few inches out and reflect the overall exhaust
air mixture, not the case temperature just inside the rear fan.  We also
upgraded our fans so that we MOVE air through the cases, although I'm
still not happy with this.  We still (as I said) get some inexplicable
(possibly thermal) crashes on the 2460 nodes, although the 2466's seem
stable so far.

   rgb

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



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



More information about the Beowulf mailing list