Does loop order matter? (pressure)

[MikeTimbers]

there is only one place where water velocity is important but it is a crucially important place and that is in the block. The great sea-change in water-cooling was the discovery by Cathar that high-speed turbulent water impacting directly over the cpu broke down the water's innate surface tension to remove the hot water-layer that builds up in the older-style maze-type blocks. This is why all modern cpu blocks use inlets directly over the die's centre.

This is why it is generally a good idea to have the cpu block as soon after the pump as possible as this is where the greatest pressure is which helps overcome the extremely restrictive nature of impingement blocks. The more restriction within the overall system, the less pressure the pump is able to create and the overall speed of the loop is slowed down.

This is simple common sense. Everything in the loop adds restriction:

the tubing itself
the blocks
the radiator
every slight curve in the tubing
every barb

The more of those you have, the greater the impact on the ability of the pump to push the water and the slower the water must go.

Now the impact of all those little curves may be minimal but please don't say they're beneficial!

 

[JonJ678]

Reason for asking is that some waterblocks are pressure dependent, so putting my ek supreme immediately after a pump would be better than just minimising tubing length.

This is why it is generally a good idea to have the cpu block as soon after the pump as possible as this is where the greatest pressure is which helps overcome the extremely restrictive nature of impingement blocks.

Happy days! I am thrilled to have you agreeing with me, even if it has taken 40 posts for anyone to do so

I'm pretty confident in this loop order being an improvement. I don't want to hazard a guess at how much of one. But its a million small things which get you to the result you want, so no 90 degree barbs for me either.


Anyone want to hazard a guess on pump->pump-> system vs pump->block->pump->system?

The former is the only way I've ever seen serial pumps done. But the water coming out of a ddc is not going to be beautifully laminar, nor will it be predictably turbulent (at least not by me). So the idea to feeding it through a block before the second pump is to break up the flow, isolating the pumps from one another.

If the block in question is the gpu one, then between this and the cpu looks like the ideal place to put a second ddc. Mount it by using compression fittings and tubes, and if needs be running a wire harness around it which I tie to the top of the case if it sags when warm. Shame it can't be run upside down, perhaps between a mosfet and the processor would work better.

 

[Jokester]

Nooooooooooooo, an impingement block will have the same pressure drop across it regardless of where it is in an otherwise identical loop. Sticking the block on the pump outlet will not increase the flowrate in the loop or increase the available pressure from the pump.

 

[fornowagain]

I don't know where people get these ideas, bends increase the speed of the water. Not that speed matters much anyway, i've seen a few people running naturally convective loops.

Tut tut, now I didn't say velocity did I? I said each bend has a pressure drop, and its proportional to flow rate. Its usually given an equivalent pipe length. The total system has a cumulative pressure loss wrt flow, which gives a curve. The pump also has a curve starting from static head (no flow). You get the system flow rate where the two curves meet. Now local velocity can up or down depending on the cross sectional area. Therefore pressure changes along with the pressure drops from losses.

Say you're flowing 1l/s going through a rad

And its easily shown, in heat exchangers (rads & blocks) the fluid flow is continuously heated.

The mean heat transfer can be expressed as

Power = mass flow rate x Cp x temperature difference

or W = dQ/dt = M Cp dT

where
Q is heat energy entering the coolant, once equilibrium is reached its the heat leaving the radiator in watts (W).
M is mass flow rate (density x velocity x cross sectional area)
Cp specific heat capacity (water is 4186 J/(Kg°C))
T2-T1, delta T, difference in temperatures from inlet to outlet for any block or rad etc.

Or basically the heat transfer is directly proportional to the mass flow rate. So if you increase the flow rate you increase the heat transfer. Now if you assume the cpu is dumping a constant heat load Q (dQ/dt is Watts or joules/second), then increasing the mass flow rate decreases the water deltaT.

---------------
Example.

A computer WC rig with a flow of 2GPM & 7/16" bore. The water moves at 1.3m/s. A 250W heat load. Water at 22C has a density of 997 kg/m3 giving a mass flow rate of 0.126 kg/s, specific heat capacity for water is 4186 J/(Kg°C)

250 = 4186 x 0.126 x dT

Therefore dT = 0.473°C

If you doubled the flow to 4GPM? Fluid velocity increases to 2.6m/s, mdot (mass flow rate) increases to 0.251 kg/s.

Therefore dT = 0.237°C

So for the 250w going into the cpu block, say with a coolant inlet temp of 20C, the outlet is 20.47C. If you seriously increased flow by doubling it you only get a marginal improvement of 20.23C.

All it means with the kind of flow rates we use, once the system has reached equilibrium heat in = heat out, then increasing flow even doubling it does sod all. It also means the temperature anywhere round the loop varies very little, and all more flow does is reduce the slight deltaT's towards the equilibrium temp.

---------------

 

[JonJ678]

And the idea is down again. Damn the analogy with electricty.

Your first reply says this though

Water pressure decreases round the loop with the most pressure at the pump outlet and least pressure at the pump inlet

which certainly suggests that for a pressure sensitive pump it should be near the outlet, no?

Sorry fornowagain, I wrote this before I saw your post. I agree completely, but question the range over which your equations are valid. Would they still be applicable at a flow rate of 0.2gpm for example?

Quite curious about what you do for a living, this isn't the first time you've produced mathematics on request for me. Thank you

edit: 0.2gph really would have been slow...

 

[fornowagain]

Sorry fornowagain, I wrote this before I saw your post. I agree completely, but question the range over which your equations are valid. Would they still be applicable at a flow rate of 0.2gph for example?

Quite curious about what you do for a living, this isn't the first time you've produced mathematics on request for me. Thank you

That is a really really low flow rate, you'd have to experiment with laminar flows. But the heat transfer delta holds yes, it would be mental. 284C. I'll assume you meant GPM. That would give 4.7C across the block for 250W.

Me? I'm a Mechanical Engineer. Many years ago I designed services like cooling towers etc.

which certainly suggests that for a pressure sensitive pump it should be near the outlet, no?

The pump is blind, it just sees the system pressure loss. The block will see the same flow rate and give the same pressure drop anywhere in the loop. It can but perform exactly the same for a given flow rate. The greater the flow, the greater the pressure drop and the better performance up to a point, as it starts to level off.

 

[Jokester]

which certainly suggests that for a pressure sensitive block it should be near the outlet, no?

No, when they say pressure sensitive they mean the pressure drop across the block. If you only have that block in the loop, then it has the full pressure drop and the maximum water velocity through it giving it the most turbulent flow you can get.

If you have other components in the loop and just for convience they are the same amount of restriction as your CPU block then it will only have half the pressure drop across it meaning that the velocity will be a lot less, meaning less turbulent flow and even laminar flow below a certain value, I'm sure fornowagain will show you the delight of Reynolds numbers and all that good stuff if you ask nicely.

It doesn't matter if that is the first pressure drop or the second pressure drop - the velocities will be the same for non-compressable fluids (ie water) meaning it has the same cooling performance.

If you really want the best position for your block place it immediately after the radiator as that is where the water is coolest, but even then the temperature delta round a loop is fairly small, typically of the order of 1C.

Edit: In fact fornowagain's calcs show delta t round around a typical loop.

 

[JonJ678]

I feel slow on the uptake today, thank you for sticking with me. I'm studying to become a mechanical engineer, which perhaps explains the interest in the theory behind this. I did not do well in thermofluids.

Non-compressible was when the penny dropped I think. At least that spares me the difficulty of obscure pump locations, though I still like the idea of hanging them above the motherboard.

So. The conclusion is that we return to the model of 'loop order doesn't matter, reservoir before pump is helpful for filling and try to minimise tubing length.' I'm happy with this.

However one issue remains, that of the ek supreme. I think I'm convinced that the rest of the loop wont care one jot if there's one ddc or two running it, but the cpu block might do. Is there any way to judge this except for getting another pump and trying it myself?

Can I even say that two ddcs in series running everything would give the same cpu temperature as one for the cpu, one for everything else with a common reservoir (in which perfect mixing conveniently occurs)? The EK will be designed to expect a certain head pressure, I don't know what this is but it may well be over what one ddc can offer it if it's all in one loop

 

[fornowagain]

Funny enough I run two DDC is series through a Supreme, see the radbox link in my sig. I run two, well tbh because I like overkill and I had two anyway. I run them just above stalling on a controller and like the idea of some redundancy if one should fail. The difference I see in actual use from one to two DDC is about 1C according to my water temp sensor.

Plug all your data into this spreadsheet > http://martin.skinneelabs.com/img/MartinsWaterCoolingFlowRateEstimatorv3_1.zip

Borrowed this off XS. Follow the curve for the EK. A single DDC should get you enough for 1 GPM (227 l/h), two DDC maybe 1.5 GPM (340 l/h), earns you a degree or two.

 

[shadowscotland]

Good thread - nice to see some maths backing up statments for a change - and also a big fan of martin's flow graphs and testing.

imho - There are four key statments with water loops
Res before pump (or better still no res)
Ideally keep flowrate above 1GPM (use spreadsheet don't guess)
Minimise loop length (loop order doesn't matter)
Buy quality buy once (inc. tube)

A typical computer WC rig with a flow of 2GPM & 7/16" bore.

A single DDC should get you enough for 1 GPM (227 l/h), two DDC maybe 1.5 GPM (340 l/h), earns you a degree or two.

@fornowagain - having just read this I should point out that 2GPM is NOT typical.
If number was just for the maths, then fine - but 0.8 to 1.6GPM is much more likely for multi block loops.
A DDC ultra but XSPC res top with a PA120.3 could acheive 3.23GPM with a XSPC edge or 1.51GPM with an EK supreme.
and that's just the variation with a single block!
I know you know this - but other readers may not - If you what to KNOW use martin's flowrate spreadsheet (links above) and then just try and keep flow in the 'good' bracket or above.

@JonJ - don't use a EK supreme in a multi block loop. Just not worth it. Get a heatkiller LC or XSPC edge (if you can find one) both are high flow and top performers.
Or have two loops - but even then both heatkiller and a few others out perform the supreme (but somewhat pointless 'upgrading' for 1-2c I'm sure you'd agree).

 

[JonJ678]

I'm very much in favour of mathematics as a means of defending your point, its a language which is difficult to misinterpret. Especial thanks for the graph above.

The graph above for the EK suggests that the difference between 60lph and 240lph is a few degrees. Which I consider significant. However, a seperate loop is unsuitable as it wastes cooling capacity of the second radiator (an 8800t and chipset blocks is far less than a 240 can cope with). So I want both 240 radiators cooling the i7.

For your four points, the first I have no issue with at all
I also think it has to be above a minimum, why have you set this value at 1gpm?
This one is rather included in point 2
And the forth is good, though I might amend to buy second hand quality parts (for some, I'm not sure about second hand pumps)


I think the choice for me is between
Using two ddcs in series, everything on one loop, no reservoir
Using one loop with pump- ek supreme - 240 radiator - reservoir and the other drawing from the same reservoir and running everything else.

The former probably offers less flow rate to the ek supreme, but the latter introduces all form of inefficiency with regard to the combined radiator. So it's a difficult decision.

One more thing to try is taking the steel plate out of the ek. This will cut down resistance enormously but kills the impingement mechanism, could be better or worse. I suspect better with low flow rate.

 

[fornowagain]

@fornowagain - having just read this I should point out that 2GPM is NOT typical.

It was just a round number, I don't use '1' for examples its annoying to work backwards. But not to worry I took the 'typical' out just for you

 

[MikeTimbers]

I really wouldn't bother with two DDC's in a single loop. I once built a radbox with two Eheim 1250s in series and the water noise was ridiculous! And it was impossible to bleed as the turbulence was so huge. God knows what two DDC's would do. If you feel the need to use two pumps, you should do two loops.