[Beowulf] bring back 2012?

Stu Midgley sdm900 at gmail.com
Mon Aug 22 17:00:01 PDT 2016


of course, that measurement was done under full load in air and in fluid.
The inbetween case  of the direct contact solutions would obviously use
less... but I've got no idea how much.

On Tue, Aug 23, 2016 at 7:40 AM, Stu Midgley <sdm900 at gmail.com> wrote:

> I measured the power draw of our 2RU 8 phi nodes with and without fans...
> the fans draw about 20% power.
>
> On a 2200kW node that's considerable.
>
> On Tue, Aug 23, 2016 at 5:36 AM, Mark Hahn <hahn at mcmaster.ca> wrote:
>
>> Extra, the direct contact cooling stuff must still have the fans in the
>>> nodes to keep the dimms/disks/power supplies cool... thus use
>>> considerably
>>> more power than nodes without fans.
>>>
>>
>> why do you say "considerably"?  1u muffin fans are annoying and not that
>> efficient, but even 1U could still use blowers.  and it's not as if
>> you need a lot of airflow to keep 5-20W of support hardware cool.
>> (that would be for non-extreme numbers of disks/dimms/nics - though it
>> occurs to me that motherboards also have some power regulation in them,
>> and if 2 sockets are drawing 250w, then 90% efficient regulation would
>> still mean 28W extra heat to dissipate.)
>>
>> iirc a typical 1U fan dissipates up to ~3W, but that's only when running
>> full on (screaming 14krpm...).  not to mention the wide availability of
>> simple 4-system-in-2u chassis (more efficient fans).
>>
>
>
>
> --
> Dr Stuart Midgley
> sdm900 at sdm900.com
>



-- 
Dr Stuart Midgley
sdm900 at sdm900.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.beowulf.org/pipermail/beowulf/attachments/20160823/96ccc6e7/attachment-0001.html>


More information about the Beowulf mailing list