[Beowulf] More about those underwater data centers

Prentice Bisbal pbisbal at pppl.gov
Thu Nov 8 07:46:08 PST 2018


One comment - my dissertation below is specifically about non-ebullient 
immersion cooling. As Jim Lux pointed out in a later e-mail, in 
ebullient cooling, some kind of surface feature to promote nucleation 
could be beneficial. Ebbulient cooling is a whole different beast from 
normal (non-ebullient) immersive cooling, since in that case you have 
changes of state and gas bubbles flowing through a liquid.

However, in all of the live and video demonstrations I've seen of Novec, 
the processors were completely bare, bubbles were forming at a pretty 
rapid rate, so again I think creating some sort of heat sink for this 
would add cost with no significant benefit.

Prentice Bisbal
Lead Software Engineer
Princeton Plasma Physics Laboratory
http://www.pppl.gov

On 11/08/2018 10:40 AM, Prentice Bisbal wrote:
>
> Heat fins are used to increase the surface area used for heat 
> transfer, since the rate of energy transfer by conduction is directly 
> proportional the surface area. Heat fins are needed when air is 
> involved because air has such a low thermal conductivity.
>
> Thermal conductivity of liquids are much high, so heat fins aren't as 
> necessary. For example, I've read that water can transfer heat orders 
> of magnitude better than air, so using water to remove hear from a 
> processor would need orders of magnitude less surface area for the 
> same energy transfer rate.
>
> Also, liquids have higher viscosities than gases, so we have to worry 
> about 'boundary layers'. A boundary layer is area where the edge 
> flowing fluid is in contact with a solid. The friction between the 
> liquid and the solid slows down the fluid near the solid. This affects 
> both gases and liquids, but since liquids have higher viscosities, the 
> effect is more noticeable.
>
> Think about a car's radiator - the air side has all the fins on it, 
> and the liquid side has smooth pipe walls.
>
> https://en.wikipedia.org/wiki/Boundary_layer
>
> Convection heat transfer is an equally important mode of heat transfer 
> in fluids, and in the boundary layer, where the liquids aren't moving 
> as fast, heat transfer isn't as good, so you need to keep your 
> boundary layer from becoming too thick.
>
> Since fluids have much higher thermal conductivities, and boundary 
> layer effects are more of a concern, I actually think a smooth heat 
> transfer surface would be better in these immersion cooling cases. I'm 
> sure smaller,  more spaced out fins would probably help heat transfer 
> without  creating too much of a boundary layer, but making those heat 
> sinks adds cost for increased performance in a situation where it 
> probably isn't needed.
>
> Now direct-contact cooling systems like Asetek products  do have fins 
> on the liquid side, if I remember correctly, but that in those 
> systems, there are pumps to provide forced convection. In immersion 
> cooling, you are relying on natural convection, so there isn't as much 
> driving force to overcome viscosity/boundary layer effects to force 
> the liquid through the heat fins.
>
> That's my thoughts, anyway.
>
> Prentice
> On 11/07/2018 04:12 AM, John Hearns via Beowulf wrote:
>> Thinking about liquid cooling , and the ebuillient cooling, the main 
>> sources of heat on our current architecture servers are the CPU 
>> package and the voltage regulators. Then the DIMMs.
>> Concentrating on the CPU die package, it is engineered with a flat 
>> metal surface which is intended to have a thermal paste to transfer 
>> heat across to a flat metal heatsink.
>> Those heatsinks are finned to have air blown across them to transport 
>> the heat away.
>>
>> In liquid immersion should we be looking at having a spiky surface on 
>> the CPU die packages and the voltage regulators?
>> Maybe we should spray the entire board with a 'flocking'' compound 
>> and give it a matt finish!
>> I am being semi-serious. I guess a lot of CFD simulation  done 
>> regarding air cooling with fins.
>> How much work has gone into pointy surfaces on the die package, which 
>> would increase contact area of course and also act as nucleation 
>> points for bubbles?
>>
>> One interesting experiment to do - assuming the flat areas of the CPU 
>> in an immersive system do not have (non thermal paste) heatsinks 
>> bolted on:
>> take two systems and roughen up the die package surfacewith sandpaper 
>> on one. Compare temperatures.
>>
>> ps. I can't resist adding this. Sorry Stu . 
>> https://www.youtube.com/watch?v=kHnifVTSFEo
>> I guess Kenneth Williams is a typical vendor Site Engineer.
>> pps. the actress in the redress had her career ruined by this film - 
>> she ver got a serious role again after perfectly being typecast.
>>
>>
>>
>>
>>
>>
>>
>> On Tue, 6 Nov 2018 at 22:57, Prentice Bisbal via Beowulf 
>> <beowulf at beowulf.org <mailto:beowulf at beowulf.org>> wrote:
>>
>>     On 11/06/2018 02:03 PM, Lux, Jim (337K) wrote:
>>>
>>>     True enough.
>>>
>>>     Ebullient cooling does have some challenges – you can form vapor
>>>     films, which are good insulators, but if you get the system
>>>     working right, nothing beats phase changes for a heat transfer.
>>>
>>     If I recall what I learned in my Transport Phenomena classes in
>>     engineering school, you need a reasonably high temperature
>>     difference to get a stable film like that. For that to happen,
>>     radiant heat transfer needs to be the dominant heat transfer
>>     mechanism, in the range of operation we are talking about, the
>>     temperature difference isn't that great, and conduction is still
>>     the dominant form of heat transfer.
>>
>>     Here's an example of what 3M Novec ebullient cooling looks like.
>>     It doesn't look like it's anywhere near the film boiling regime:
>>
>>     https://www.youtube.com/watch?v=CIbnl3Pj15w
>>
>>     --
>>     Prentice
>>
>>>     *From:*Beowulf [mailto:beowulf-bounces at beowulf.org] *On Behalf
>>>     Of *Prentice Bisbal via Beowulf
>>>     *Sent:* Tuesday, November 06, 2018 8:17 AM
>>>     *To:* beowulf at beowulf.org <mailto:beowulf at beowulf.org>
>>>     *Subject:* Re: [Beowulf] More about those underwater data centers
>>>
>>>         . And serviceability is challenging. You need to pull the
>>>         "wet" boards out, or you need to connect and disconnect
>>>         fluid connectors, etc.  If you're in an environment where
>>>         you can manage that (or are forced into it by necessity),
>>>         then you can do it.
>>>
>>>     I think everyone on this list already knows I'm no fan of
>>>     mineral oil immersion (It just seems to messy to me. Sorry,
>>>     Stu), but immersion cooling with other liquids, such as 3M Novec
>>>     engineered fluid addresses a lot of your concerns. It as a low
>>>     boiling point, not much above room temperature, and it was
>>>     originally meant to be an electronic parts cleaner (according to
>>>     a 3M rep at the 3M booth at SC a few years ago, so if you pull a
>>>     component out of it, it dries very quickly and should be
>>>     immaculately clean.
>>>
>>>     The low boiling point is an excellent feature for heat transfer,
>>>     too, since it boils from the heat of the processor (ebullient
>>>     cooling). This change of state absorbs a lot of energy, making
>>>     it very effective at transferring heat away from the processor.
>>>     The vapor can then rise and condense on a heat exchanger with a
>>>     chilled water heat exchanger, where it again transfers a lot of
>>>     heat through a change of state.
>>>
>>>     Prentice
>>>
>>>     On 11/05/2018 06:30 PM, Stu Midgley wrote:
>>>
>>>         I refute both these claims.
>>>
>>>         You DO want to run your boards immersed in coolant.  It
>>>         works wonderfully well, is easy to live with, servicing is
>>>         easy... and saves you almost 1/2 your power bill.
>>>
>>>         People are scared of immersion cooling, but it isn't that
>>>         difficult to live with.  Some things are harder but other
>>>         things are way easier.  In total, it balances out.
>>>
>>>         Also, given the greater reliability of components you get,
>>>         you do less servicing.
>>>
>>>         If you haven't lived with it, you really have no idea what
>>>         you are missing.
>>>
>>>         Serviceability is NOT challenging.
>>>
>>>             You really do NOT want to run boards immersed in coolant
>>>             - yeah, there's folks doing it at HPC scale
>>>
>>>             Whatever the coolant, it leaks, it oozes, it gets places
>>>             you don't want it to go. And serviceability is
>>>             challenging. You need to pull the "wet" boards out, or
>>>             you need to connect and disconnect fluid connectors,
>>>             etc.  If you're in an environment where you can manage
>>>             that (or are forced into it by necessity), then you can
>>>             do it.
>>>
>>>         -- 
>>>
>>>         Dr Stuart Midgley
>>>         sdm900 at gmail.com <mailto:sdm900 at gmail.com>
>>>
>>>
>>>
>>>
>>>         _______________________________________________
>>>
>>>         Beowulf mailing list,Beowulf at beowulf.org <mailto:Beowulf at beowulf.org>  sponsored by Penguin Computing
>>>
>>>         To change your subscription (digest mode or unsubscribe) visithttp://www.beowulf.org/mailman/listinfo/beowulf
>>>
>>
>>     _______________________________________________
>>     Beowulf mailing list, Beowulf at beowulf.org
>>     <mailto:Beowulf at beowulf.org> sponsored by Penguin Computing
>>     To change your subscription (digest mode or unsubscribe) visit
>>     http://www.beowulf.org/mailman/listinfo/beowulf
>>
>>
>>
>> _______________________________________________
>> Beowulf mailing list,Beowulf at beowulf.org  sponsored by Penguin Computing
>> To change your subscription (digest mode or unsubscribe) visithttp://www.beowulf.org/mailman/listinfo/beowulf
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.beowulf.org/pipermail/beowulf/attachments/20181108/8480d490/attachment-0001.html>


More information about the Beowulf mailing list