[Beowulf] Innovative liquid cooling
j.sassmannshausen at ucl.ac.uk
Thu Feb 28 13:31:11 PST 2013
check out the Iceotope website of how they are doing it:
They are only using convection and the mother board is mounted upright as
opposed to be the more convential horizontal mounting. Also, they don't cool
the board to 20 °C but have it more or less at a higher temperature. Here the
used Nova has a low viscosity and moves well as far as I understand it. There
is still a chilled plate (one side of the aluminium container) which removes
The AC in that data-centres is a complete and utterly wrongly installed system
done my AC specialist. Sucking off hot air at the bottom is never a good idea
and blowing in cold air parallel to the racks with a high wind speed does not
help either. I got hotspots in the room and if one unit fails I can run into a
problem very easily. So yes, that number of 2 might be high but in my case it
is most likely more the correct number.
All the best from London!
P.S. I am not working for Iceotope!
On Donnerstag 28 Februar 2013 Mark Hahn wrote:
> > consumption from 41 kW (electricity, probably same amount again for
> > cooling)
> I really doubt it. there is something profoundly wrong if an
> HPC-type datacenter, with completely conventional servers, air/DX cooling
> runs at a PUE of more than about 1.3-1.4.
> immersion cooling sounds appealingly unconventional, but if you think
> about the heatflow, you've still got to move it around. you still need
> a heatsink on the CPU with fins, some way to move the fluid past these
> fins and get them to the secondary heat exchanger. as with airflow
> management in a conventional DC, surely one has to ensure that cool
> fluid gets to the CPU fins and heated fluid finds its way to the
> rejection exchanger. surely convection wouldn't be good enough without
> some serious re-engineering. or do these systems rely on boiling?
> so I wonder what the effective thermal resistance is for such a
> fluid system (assuming some fishtank-like pumping). risk-wise,
> I'm pretty sure I'd be more comfortable with a heatpipe inside
> the chassis mating to a coldplate/exchanger built into the rack
> running the water loop.
> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
> To change your subscription (digest mode or unsubscribe) visit
University College London
Department of Chemistry
email: j.sassmannshausen at ucl.ac.uk
Please avoid sending me Word or PowerPoint attachments.
More information about the Beowulf