hahn at physics.mcmaster.ca
Fri Jun 30 12:20:53 EDT 2006
> details of your case's design. Some cases rely on very particular
> airflow patterns -- suck cool from front, pull over memory, disk, CPU,
> and either out directly or through power supply and out. If you open
that's pretty much the standard (informal AT standard, but formal
ATX/BTX/etc). it's pretty close to universal to have the PS and
possible aux exhaust fan pretty close, near the top of a minitower.
intakes are another matter - through the front is desirable, since it
introduces space from the outflow, and usually means airflow over drives.
(although drives are usually <10W nowadays, you'd still like flow...)
but PCI slots are often open/perforated, and sometimes case vendors
will introduce airflow innovations, which may or may not work well
(blow-hole on the side, presumably as a forced intake, PS on the bottom,
passive vents on the sides, etc.)
IMO, the best plan is to simply have substantial air exhausted from
one general area, and no intakes near that. as such, I don't think the
case on a minitower makes much difference - removing the sides/top/front
shouldn't matter if there are exhaust fan(s) in the back and no open space
between the nodes.
> (especially towers) that have a front fan push fan as well as rear pull
> fans are a bit more likely to keep enough cool air mixed into any
> recirculated air to avoid problems.
in particular, I don't really like push/pull systems. more moving parts,
and it's not clear that they really work together. PS + 1 fan with
appropriately placed intakes is really a better idea. for 1U, centrifugal
fans really are nicer than a bunch of bitty axial fans...
heh, this thought just struck me: make your raised floor deeper, and
mount your nodes facing _down_ into the floor. no tiles, just the nether
ends of your nodes. convection and chiller blowers to do all the work
for you, and it would look quite striking ;)
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