[Beowulf] CPUs are not hot enough
mathog at mendel.bio.caltech.edu
Fri Apr 1 18:04:15 EST 2005
I was musing over trends in computing, with vast
amounts of power going into machines, coming out as heat, and
then being dumped into the environment via equally massive A/C
systems. Or at the very least, a really big fan.
Wouldn't it be nice to reclaim some of that power?
For the purposes of thermodynamics CPUs are basically
just big resistors. Put a voltage across them, current
flows, heat is generated. Imagine using
the difference in T between the CPUs and the ambient
environment to make more electricity. Replace the A/C
with a heat driven engine that turns a generator,.
Here's where the irony comes in - the CPUs aren't hot enough.
room temp is around 300K and the CPUs are at most another
50K above that, giving a pathetic Carnot efficiency. Imagine if the
chips were designed to run really, really, really hot. For
instance, 600. At that point the heat engine could run at
about 50% efficiency and one could recover about half
the power that went in. Silicon doesn't actually melt
until around 1700K, so there's possibly even some leeway
above that. Have to use something else instead of metal
for the traces though, since aluminum will have melted
off at 933K. Silicon might not work at those temperatures
either but there's likely some other substance
that is a semiconductor at 1000K. The boot sequence should
be "interesting" if it becomes necessary to preheat the
CPU to a very high temperature before it can run.
Obviously this is not hardware you'd want on your desk
but with data centers churning along at 50 megawatts perhaps it's
worth thinking about creative ways to get the maximum amount
of crunch per watt. You've doubtless all seen suggestions that
exotic CPUs be run at superconducting temperatures. Probably
this is the first time you've heard anybody suggest they be run
at incredibly high temperatures instead.
And as a side benefit the next Venus lander could have
a computer that will run at ambient temperature.
mathog at caltech.edu
Manager, Sequence Analysis Facility, Biology Division, Caltech
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