[ExI] homebrew cold freon bath super computer

Eugen Leitl eugen at leitl.org
Fri Mar 9 19:43:26 UTC 2012

On Fri, Mar 09, 2012 at 09:56:52AM -0800, spike wrote:
> We have a tool at work which is a cold Freon bath. submerge the whole mess

Where do you work now, if it's not a secret?

> in cold freon and BOOM, we have a great poor-man's highly parallel super
> computer, ja? spike

Fluorinert & Co is *really* expensive. People have been using submersion
in mineral oil, though. As long as you don't use hard drives (SSDs are fine)
mineral oil works well enough, albeit messy.
> Another thought: in the last several years, I have heard a lot of moaning
> about how processor clock speeds suddenly stopped increasing, settling

Purely sequential people are screwed, yes. Notice that multithreaded people
will be equally screwed at sufficiently high core count (no memory coherency
for you), but they yet blissfully unaware.

> around 4 GHz.  But it looks to me like we have enjoyed tremendous progress,
> not in calculations per unit time, but rather in calculations per unit
> energy.  That these little phone processors can do so much is really

Sure, Koomey's Law:

> exciting, because there are so many useful tasks that do not require
> trillions of calculations, but perhaps only a few billion.  Chess is a good
> example: it is playing at the level of the best humans using less than a
> watt.  If we can do all that for just that little bit of energy, we can
> stack them closely and carry away the waste heat using a simple device.
> This reduces the latency between nodes and opens up a whole nuther
> collection of capabilities and a whole nuther field of cool challenges.

Watercooled racks and since recently whole turnkey supercomputers are

Now 3D-stacking with TSV with forced (microliquidics) liquid (metal as
in Ga or K/Na eutectic?) cooling has not made mainstream yet. 

I anticipate liquid cooling in enthusiast APU/memory stack setups very soon
now (couple more years) and then in the mainstream by 2020 latest.
> Processor hipsters, have we had major advances in calculations per unit
> energy, or watt?

Sure, websearch Koomey's Law.

More information about the extropy-chat mailing list