Let's hope that we can clear this up, once and for all.

I found that most people that have a hard time understanding the whole concept, or have a difficulty getting into such a discussion, have the difficulty of not being able to differentiate the starting conditions from the running (aka balanced) conditions: sure your water temp is 20 deg, when you start your PC, but it will rise.


No. Twice heated, twice cooled (if you double the flow rate).

The quantity of water is irrelevant: it will reach a specific temperature (and temperature gradient) regardless: it would just take a bit more time with a larger quantity of it.

By having a lower flow rate, the water temperature will be spread over a larger range. That's your increased thermal gradient. The average temp remains the same.

Yes, there is a temp difference, within the rad, but the flow rate is such that it would be hard to measure. The average flow rate of a rig is ~50 gph, that's roughly 1 gpm (gallon per minute): think about it.

hey guys, we have been feeding a troll
I just wasted 15min reading the OCers thread Since87 (?) linked to previously
and this Greystar is gameing the world

but eliciting some explanations of great clarity, over, and over, and over, and over

I would suggest just following his posts with the caption:

Caution: lots of words, no understanding, too stupid to learn; Greystar is struttin' his stuff

I've been starting to wonder about that.

Man I'll be fuming if I've wasted that much of my precious revision time for no good reason.:mad:

8-ball

OK, I've got a question which is kind of related to the original title of the thread.

Suppose you have flow vs head loss charts for various lengths of tubing, a radiator and a waterblock, and maybe a gpu block as well. How would you combine this to produce a head loss vs flow curve for the whole loop from the pump outlet round to the pump inlet for comparison with the pump p-q curve.

Also, can you do the same thing with flow vs efficiency charts to try and predict a systems sweetspot.

8-ball

Absolutely.

You must have three things:
the pressure drop curve for the heatercore/rad
the pressure drop curve for the block
the PQ curve for the pump.

As long as you keep the flow speed less than 2 or 3 feet per second within the tubing, you can basically ignore it (unless you want to be accurate).

You then add the pressure drops for the block and core, overlay this total over the pump curve, and see where the lines cross: that's your resulting flow rate and total pressure drop.

It's easier said than done, but that's the principle. I usually run the numbers at different flow rate increments.

I don't think you have to worry about blowing off your fittings with that pump; I have one, and the 4' head spec seems pretty generous...:evilaugh:

It does have a built-in valve to tailor flow rate, but I doubt you'll run it at anything but wide-open.

@8-ball,

In priciple, you can "add" the curves as Ben has suggested. Practically speaking, it doesn't work to perfection. As you likely know, the impact on flow from a given object doesn't begin and end at the object's physical boundaries. Fortunately, the very nature of a centrifugal's curve means that it minimizes the errors in your estimate. What I mean is that if you guess your resistance higher than it really is, the flow will not go up much because increasing the flow will increase the delta-P. Likewise, if you estimate your resistance lower than it really is, the corresponding drop in flowrate drops the system resistance a little. It's a sort of attenuated error, I guess one could say. Does that make sense?

By adding, do you mean that I can develop a curve by considering each discrete flow rate in turn and adding the corresponding pressure drops for all of the components.

How about efficiencies? Or is that a bit more complicated.

8-ball

Yes, you can simply add the dP of each component, for a given flow rate.

As myv65 pointed out, this method is not terribly accurate, and a small margin of error on the pressure, won't amount to a big difference on the flow rate.

It gets a bit tricky if you have an odd configuration, where you've got an extra component running in parallel, like a chipset block.

Some handy, "good enough" techniques:

Most devices have a PQ curve of the form:

dP= Q^2 * Rf

where

dP is the pressure differential across the device
Q is the the flowrate through the device
Rf is what I call the flow resistance

Based on Bill's data, waterblocks generally match the equation very well, radiators are more aberrant but 'close enough' for meaningful quicky checks.

You can calculate Rf for a device by picking a point towards the right end of a device's PQ curve and solving the equation. A quicky shortcut for determining Rf, is to just take the pressure drop at 10 lpm and divide it by 100.

So, find Rf for the devices in the loop. Then add them up for a combined Rf, or:

dP-system = Q^2 * ( Rf-block + Rf-rad + Rf-tubing... )

Plug the equation into Excel and graph. Voila! System PQ curve.

As myv65 pointed out, the nature of centrifugal pump PQ curves tends to minimize the effect of errors, so although this is a somewhat sloppy technique, you still can get results accurate enough to do meaningful comparisons fairly quickly.

Correct me if I'm wrong, but that equation only works if the curve is actually a straight line, no?

From what I've seen, for a typical range of flows, the curve is actually "curved"!:D

Add all the head loss charts together. That is, for each flow rate, make the pressure the sum of the pressure on each head loss chart at that flow rate.

As you probably know, you can then take this overall head loss chart and plot against the pump curve. The interesection point is the head loss and flow rate of the entire system.

Not sure what you mean by "flow vs efficiency charts."

Alchemy

Ben... Since87's equation is that of a parabola.

I thought it would be somewhere between parabolic and cubic. Doesn't the "flow resistance" Rf increase as Q increases? If so, it wouldn't be parabolic, would it? I could be totally off base here, but I am curious enough to ask.

Thanks, that's what I thought.

By efficiency, I mean C/W.

I guess what I'm asking is can you produce a C/W vs flow chart for each stage in the transfer of heat from the cpu to the air, then sum these to determine an overall C/W vs flow for the entire loop, including all of the waterblocks.

Factor in the variation in heat load for different flow rates and try and predict the sweet spot for the system.

Or have I just ridiculously oversimplified what people are trying to do in the thermal simulator forum.

8-ball

It's just some messy nomenclature on Since87's part. The dp is really the flow resistance for the component, as a function of the flow. The term he's calling the flow resistance, Rf is really a constant, needed to shape the parabola.

Hot damn, you're right!

Let's put it to the test, with some data I've extracted from Bill's roundup:

Using the "Surplus"'s curve, I extracted the following info:
dP @ 2.0 gpm: 1.6 psi
dP @ 1.5 gpm: 1.0 psi
dP @ 1.0 gpm: 0.3 psi
(a rough extraction, but let's take a look!)

Ignoring the units,
using the first set of numbers, Rf = 0.4
using the second set of numbers, Rf = 0.44
using the third set of numbers, Rf = 0.3

Hum...:shrug:

as has been pointed out, the individual resistances can be simply added
but this sum will be 'correct' only at that specific flow rate as each component's rate of variation is different

and consider the potential accuracy, 15% would be unusually good
this is due to the quite small 'individual' resistances being measured; the instruments' accuracy and the substantial effect of rounding errors

Null-A Ben
if you are going to calculate; use the Chevette hc data, it is much more accurate

what a pleasant thread without having to push a rope uphill

:confused:

You got it right. Last I heard, that's exactly what's supposed to be going on in the simulator. Except I don't think the variation in heat load will be considered, since it's hard to say with inline pumps how much heat is going into the water. And if you just consider the energy input to the water by pressure increase you get . . .

Hm. 1.5 gpm for an overall pressure drop of 5 psi (~10ft H2O) gives you an energy increase in the water of 3 W. So, maybe 5% of the CPU load.

If you have flowrate constant, flow resistances add up just like resistances in an electrical circuit.

[Edit - removed non sequitir. Thanks, Since87.]

And to nitpick on what someone else said, centrifugal pumps *are* analagous to voltage sources. But they are not analagous to *ideal* voltage sources. They induce a potential in the fluid (pressure), but that potential is a function of flow rate. I've only studied basic circuits so I don't know if there are any voltage sources that act this way. But it's still a decent analogy, in my opinion.

Alchemy

I'm tempted to have a look over at the ideas in the simulator forum, but I really don't have the time, what with finals this summer. Maybe afterwards.

8-ball

Actually, a better place to nitpick would be my comment that, "Positive displacement pumps are not current sources." If you ignore the pulsation in the output flowrate, a positive displacement pump is fairly analogous to a practical electrical current source.

However, no electronics manufacturer is going to sell a voltage regulator, who's output voltage sags as much with output current as the dP of a centrifugal pump does with flowrate. Voltage regulators will typically maintain an output within 5% of nominal (or substantially better) over their entire operating range.

I'd still have to say that a practical voltage source is an awfully poor analogy to a centrifugal pump.

Edit: I suppose a battery with its inherent series resistance behaves somewhat like a centrifugal pump. No competent electrical engineer would consider a battery a 'voltage source' though.

Here is a link to a post I made at OCAU for working out the target flow rate of your system given a few things that you can figure out for yourself.

As Bill says, getting to more than a 10% accuracy is hard, especially when hand-plotting/reading information from crappy PQ graphs/plots, but for the most part, 10% slop is "good enough" unless you're down below the 2lpm mark.

Thread is here:

On pressure drops and flow rates

Translation: Nice thread when the "chaff" is gone and the discussion remains focused and based on reality.

Yes. I'm an electrical engineer. My formal training in fluids consists of about a week during freshsman Physics in college. Virtually everything I know about fluid flow comes from the forums so my nomenclature is definitely sloppy.

Per Bill's suggestion I did similar for some of his most recent data. I chose to do it for my HC since it was the most restrictive and allowed me to get the most accurate dP numbers off the graph.

I printed the graph, and used digital calipers to get as accurate data as possible off the graph. My results:

dP = 0.71 @ 12 lpm, Rf = 0.00493
dP = 0.32 @ 8 lpm, Rf = 0.00500
dP = 0.092 @ 4 lpm, Rf = 0.00575

Pretty good match at high flowrates. Less good for low flowrates.

The following graph illustrates the effect of modeling my HC with two different waterblocks, and using both 0.005 and 0.00575 for Rf.

http://uffish-thought.net/wc-gifs/hcrf.gif

Fairly small errors IMO. (Especially when you consider the C/W change associated with the difference in flowrate.)

But, like I said, the techniques I suggested were for quickly doing "good enough" analysis. This is not the way to get the most accurate results.

If someone understands the limitations and can make use of my suggestions, great.

I didn't really expect my suggestions to enable you to do more useful analysis than you already do Ben.