[conspire] Last Year's Supercomputer

[Rick Moen]

Quoting Mark S Bilk (mark at cosmicpenguin.com):

> The usual technique for cooling is to blow a lot of air through
> the case in order to get a small fraction of it flowing past 
> the few parts that get very hot.  This requires powerful, noisy
> fans.
> 
> My solution is to hang a small fan right up against each hot
> device.  I get them from Halte[k|d] -- 6cm x 6cm, 12V, 1watt,
> about $7, absolutely silent.

See, that's good, but I think one can do better.  With careful system
design and parts selection, it's possible to do mostly passive cooling,
where conduction via adjacent metal, convection of air through a
spacious case, and scrupulous spacing out of the problem components
keeps things from overheating.  I don't know about you, but I'd feel
really dumb replacing a $200 part (say, your pride-and-joy large hard
drive) just because a $7 fan failed.

Just before the dot-com crash, we had an example of forced-air cooling
run amok that I regard as an excellent cautionary tale.  My
then-employer was, I believe, the first firm to release a dual-Athlon
rack-mount 2U server system.  It drew ridiculous amounts of electical
power, almost all of which of course eventually got radiated out as
heat.  To prevent it from melting down, a hell of a lot of air got
driven through the chassis from front to back, and the case itself was
of careful (patented) design to maximise the benefit of all that
airflow.  We sold clusters of those damned things for scientific
computing, and the sorts of labs that bought them had to bring in
extra-heavy-duty power feeds for them -- not to mention air
conditioning.  And the din they put out was just unbelievable!

I wouldn't have taken one on a bet, and I'll bet that just the AC power
and air conditioning bills for a cluster of them would break the bank.
I wonder if they're still in use, anywhere?

So, anyway, the aforementioned K6 tower is my preferred sort of system.
It's relatively quiet, and probably _could_ get by on almost entirely
passive cooling, because of the huge case and large amounts of spacing
between the heat-producing components (which, on reflection, means
mostly the two hard drives).  I have a couple of case fans on the
belt-and-suspenders theory.  It would be _quieter_ without them, and could
probably survive without them and with a PC Power & Cooling Silencer
instead of the big-ass TurboCool.  But it's relatively quiet and runs 
very cool, while giving me the confidence that it would remain OK even
if a couple of the fans seized up.

> I have one blowing on the Riva TNT graphic card, one on the
> soon-to-be-retired swap drive, and one on the two main 3.5" drives
> that are stacked only 1/8" apart.  

See, that bit about 1/8" spacing sounds bad, to me.  I'd stretch out the
spacing a lot, if possible.  If the case didn't permit that, I'd shitcan
the case and get a bigger one.  In fact, that's exactly what I _did_ do,
to get the case the K6 now resides in.

I'm not in any way saying you haven't done a great job.  What matters is
results, and you say it runs cool.  That counts as success.  However,
you've achieved that at the cost of making the cooling reliant on a
bunch of $7 fans.  I get greater warm-fuzzies from removing that
reliance, is all I'm saying.

> I haven't had the side panel on the case for years....

Most cases are actually designed so that they do effective air flow only
with the case closed.  You may be impairing heat control, by operating
it that way.

-- 
Cheers,             "Don't use Outlook.  Outlook is really just a security
Rick Moen            hole with a small e-mail client attached to it."
rick at linuxmafia.com                        -- Brian Trosko in r.a.sf.w.r-j