Liquid cooling?

Jim Lux James.P.Lux at jpl.nasa.gov
Wed Apr 24 15:47:07 EDT 2002


Jim makes a number of good points, some of which would also apply to 
conduction cooled equipment (as used in space applications and some 
avionics, where you can't depend on the air density). In my efforts to 
design a "field usable Beowulf" which is entirely sealed (mud/dust proof, 
etc.), I had come to similar conclusions about the viability of liquid 
cooling (mass gets to be a bit of a problem, though).

However, a basic philosophical issue arises...  The "beauty" of a Beowulf 
is that it uses "commodity" computers, and so, can capitalize on enormous 
efficiencies of scale for the huge consumer market.  As you start to go 
towards less consumer configurations, you're straying, to a certain extent 
from the "pile of cheap PC's" paradigm and back towards the "behemoth in 
the machine room" model.  Certainly, the 500+ node computers that folks are 
putting together with 1U cases, dozens of machines in dozens of racks, with 
dedicated cooling, etc., is straying pretty far from the original Beowulf 
concept.

It still is cluster computing, and maybe what makes it Beowulf'ish is not 
the hardware, per se, but the fact that you are using "off the shelf" 
cheap/free software (Linux, e.g.), and "off the shelf" interconnects??

By the way, I wouldn't fool with DI water in a totally immersed system.. 
too corrosive.  If you don't want to pop for the Fluorinerts, then various 
silicone and mineral oils would work well, are non-toxic, and inexpensive. 
These have been used for decades for immersed cooling of all sorts of stuff 
(transformers, for instance). You'd want to assess compatibility with 
adhesives and existing coatings.  And, you better hope that it really does 
improve reliability... it's going to be a mess to service.  You might want 
to do some hard core burn in first and get past the infant mortalities, 
before you "literally take the plunge".

And, while plunging the Mobo in oil wouldn't bother it, I wonder if the 
same is true of things like the hard disk drive?  They're probably vented, 
and/or have moving parts that are outside the sealed area.

At 02:14 PM 4/24/2002 -0400, Jim Fraser wrote:

>While I think overclocking is kinda silly in extreme cases (like hot-rods)
>and nearly pointless for most real serious computing applications, I think
>the water-cooling has real merits and could be considered for dense
>clusters.
>1) Water cools orders-of-magnitudes better then air
>2) It is far quieter
>3) CPU temps hardly vary as compared to air (even under load) (better
>stability)
>4) It does not have to be that much more expensive then high-end air cooling
>(there is a real price to cool dual cpu's in a 1U steel box.)
>5) leaks are almost unheard-of, and are not as catastrophic as they sound
>(distilled water is generally not a problem, but a leak is a possible mode
>of failure)
>6) Dense water cooled systems could be easily be engineered to remove bulk
>heat far better then the rows of tiny little cheap jap fans whirring at 7000
>rpm...talk about failure rates!?! Fans are most prone to failure that result
>in hardware breakdowns.  Don't discard water cooling.
>
>jim
>
>
>---
>On Wed, 2002-04-24 at 15:10, Worsham, Michael A. wrote:
> > Has anyone attempting to create a beowulf cluster using extreme methods of
> > cooling, such as the liquid cooling?
> >
> > Example sites: http://www.koolance.com/, http://www.senfu.com.tw/, &
> > http://www.overclockershideout.com/
> >
>
>Well, I think Robert Brown has FINALLY been beaten here.
>You're not going to install Freon tanks, complete with plastic
>fish are you Bob?
>I just have this bizarre vision of Bob in an aqualung visiting
>a Freon-flooded machine room...
>

Jim Lux
Spacecraft Telecommunications Equipment Section
Jet Propulsion Laboratory
4800 Oak Grove Road, Mail Stop 161-213
Pasadena CA 91109

818/354-2075, fax 818/393-6875

_______________________________________________
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