[Beowulf] Dunking for Density: New Projects Pursue 3M’s Take on Immersion Cooling
Ellis H. Wilson III
ellis at cse.psu.edu
Fri Nov 15 19:30:08 PST 2013
On 11/15/13 03:57, Eugen Leitl wrote:
> (go visit the site for pretty pikchers)
Cool stuff, as immersion cooling always is. However, this got me
thinking in the complete opposite direction...
Obviously there are cases where you need absurd density for latency
critical applications or, I suppose, if the cost of the land in which
you needed the computing to be local too was very costly. Or, as they
show, maybe you just want the efficiency of cooling to be near to perfect.
However, (here begins a very possibly insane set of ideas) what if
instead of reaching for low density and really efficient cooling, you
went the other way and spread things out and tried not to actively chill
air at all? Sure, your network latency will shoot up, but for many
applications (data centers in particular) this may not matter at all.
As a broken example, in the fall, my windows are open here in PA, I have
no "chillers" in my house, and my server/desktop/microcluster runs just
fine. This is clearly a straw man example, as I have far too few
machines for this to be reasonable. But my core question here stands:
Are there places in the world so arid and stable in temperature that you
could effectively run a data center or compute farm outside (or "almost
outside") and just let some big fans move the heat away from the cluster
(maybe with just a pavilion-style roof on it)? Maybe even someplace
that has reasonably stable temperatures that you could just slightly
condition the air for humidity before pushing it through? I know
condensation is a problem if you were someplace too cold or too humid,
but I'm not sure what the relative sensitivity for that is.
Just throwing this out there -- the people who actually know things
about material science on this list should feel free to laugh and
explain to me why this could never work. Just a beer-infused
More information about the Beowulf