Quick Liquid Physics Question

[masshuu]

While I'm designing a water cooled server system cause I'm bored,
but i stumbled upon an interesting question.

IE:
http://forums.x10hosting.com/attachm...1&d=1252221514

The question is, will the pump have to work to pump water back up to the top, or will the force of the water travailing down counter act that force?

and the same with a Reservoir on the ground, will the pump need to work to pump the water from ground level, or will the flow of water from the pump level to ground level counter act the force of pumping it up?

[zen-r]

Gravity working on the downward flowing water will assist in pushing the water back up the other side, but a pump will still be needed to compensate for the slight losses caused by friction. Something, such as a pump, would still be needed anyway in order to get the flow started in the first place.

A cleverly re-designed system may remove the need for a pump altogether, by taking advantage of the fact that hot water naturally rises above cold water.

I hope this helps!

[Sharky]

This is true, however, depending how hot the components get, the flow of water/coolant may not be sufficient by using gravity alone. Gravity fed systems work for getting things hot, eg, older central heating systems. You'd have to have the cooling fins fairly high. Better off having a couple of pumps and a fan on the water cooling block. Even better, put the cooling block outside (in the shade if it's always hot) and use a limited amount of antifreeze if it's that cold (be aware that antifreeze has different thermal properties to water, but that's only relevant really if you're hitting car/bike engine temperatures... I can't remember where I read that now. )

[masshuu]

ok, so yeah. There was a voice in the back of my head saying that it wouldn't work.
Note though: im not actually making this, though i wish i could.


But i think if i would build this, i would use pumps. The design i made only has 15kw per rack O.o, so im sure you want a pump to move that water.

and if your interested, the specks i hammered out for it go like this:
Each server is water cooled, a single 1 inch tube which splits off 1/2 inch to each server

Each server is 1U in height, has 2 Quad core Xeon Processors and 24 gigs of ram each, costing 7139.26 each.

1 42 U rack filled with 40 servers will cost $285,570.40

Together in a cluster, they will have a combined:
960 Gigs of ram
80 Quad Core 3.2 Ghz

and a theoretical 5.120 Teraflops

now all i need is $300,000 and a location with a 20kw hookup.
I wonder if these skills will be usefull in the future.

If i could locate a 1U case that can hold a 12x13 mobo(i didn't look very hard), i could double the specs of each machine, and have in each rack:

1.875 TB of ram
640 3.2 Ghz Cores,
and a theoretical 10.24 Teraflops of power

So back to the orignal setup.

if i had 6 racks, using only 1 node as a master node, all the others as slave nodes, here are the specs of it:
5.6 TB of ram
1912 3.2 Ghz Cores
and a theoretical 24.4736 Teraflops of power

And double it if i can find that case for the larger mobo:
11.2 TB of ram,
3824 3.2 Ghz Cores
theoretical 48.9472 Teraflops of power

Power use of these six racks will only be around 100 Kilowatts of power.

Normally i think datacenters use 3-4U cases, so if you expanded those 6 racks into 3U cases, you would need 20 racks, or 30 racks for 4U cases