[Beowulf] Innovative liquid cooling

Lux, Jim (337C) james.p.lux at jpl.nasa.gov
Thu Feb 28 15:23:06 PST 2013


> 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?

Actually, since the thermal conductivity of the liquid is so much better than air, you probably don't need finned heat sinks. Heat transfer rates is dependent on 3 factors: temperature difference, area the heat is flowing across, than the thermal conductivity of the conductor. The main reason for heatsinks is to increase the surface area to increase heat transfer.

Since a liquid is about 1000 times better than air at conducting heat, you would only need about 1/1000 the surface area to have the same heat transfer rate as a gas, essentially eliminating the need for heatsinks.

>> yes.. and you can package more densely.  For a lot of parts, most of the heat is carried out through the pins to the board and then radiated or conducted from there. So improving the heat transfer from the board itself helps.  It also helps reduce temperature differentials across the board, which should help reliability.  There's always a CTE mismatch between board and parts and every thermal cycle with some delta loads and unloads the connection. 

Another angle is that when working with a liquid, the closely-spaced fins of a heat sink would retard heat transfer in a liquid. Why? Because the viscosity of a liquid is so much higher than a gas. The liquid can't flow through those narrow gaps between the fins, causing stagnant liquid which would collect the heat, but not carry it away through convection.

For a real-world example of this, think of a car's radiator, or hot-water baseboard heating. In both of the cases, the liquid flows through a smooth-walled pipe with metal fins on the outside. The fins are on the air side, not the liquid side where the increased surface area is needed. Putting fins on the inside would hind the flow of the liquid and create a thicker-boundary layer.

For the reasons stated above, I think air-cooling requires more engineering to do right than liquid-cooling.

>> No question there, as far as the "heat transfer" part goes.  But "system engineering" is more complex with distributed liquid cooling, particularly if you want to do swaps of components without draining the system.  You can push air where you want it to go with sheet metal, cardboard, and duct tape.


Boiling at these temperatures would require a relatively volatile liquid, which would probably flammable (explosive), or bad for you in other ways (carcinogenic, etc.)  Change of state is very efficient for heat transfer, though. That's why many still consider steam heat to be the best form of heating.

>>  Ebullient cooling using Fluorinert has been around a long time. There's the "immerse the board and let the bubbles rise to the surface" scheme, and the "spray liquid on the surface and let it evaporate" scheme, as well as a variety of "boilers" for localized high power components  (high power vacuum tubes are a good example of the latter)


More information about the Beowulf mailing list