Testing waterblocks vs pumping power

[nikhsub1]

Originally Posted by pHaestus

I don't see any of their canadian distributors stocking the thermochills. I guess I'll need to order direct from DD.

You asked about North America? AFAIK, DD has an exclusive at least in the US on the Thermochill's. www.mountainmods.com has them too I believe (via DD). Ben Rising over at Mountain Mods is a very good guy.

[pHaestus]

I was looking at the DD resellers in canada for someone who might carry the thermochills. Figured I'd save on customs fees and shipping costs that way.

[pHaestus]

Joe just PMed me from IRC that the madshrimps people (jmke and liquid3d) have been banned from the forums and their IPs are banned from even viewing this site. I apologize for the interruption in this discussion earlier. Hopefully it won't happen again.

[Incoherent]

(from Nexxos thread regarding PQ)

Originally Posted by Cathar

To do it properly you really need to measure it.

For a "ballpark" though. Classic flow mechanics tells us that the level of flow resistance is proportional to the flow rate squared (P = Q ^ 2). This doesn't take into account turbulence and boundary layer conditions though.

For a fairly decent ballpark equation to extrapolate from OC.com's results I've personally found that P = Q ^ 1.85 offers a pretty decent approximation across the typical range of flow rates seen in water-cooling. P = pressure. Q = flow. Not as ideal as having real data, but when given a single pressure/flow point like OC.com I personally found it to give a fairly close ballpark curve.

Re the P=Q^1.85 fit, I think there is a better way.

I assume you are familiar with the idea of a "k-factor", I stumbled across it recently when exploring the flow measurement problem.
Your PQ curves are not behaving, shouldn't one be able to generate a constant which, even allowing for transitions between flow regimes and boundary conditions, is relatively flat across flowrate?. I am getting this with measured data, (see attachment) it is constant enough that I would propose that every block has a "K-factor", a constant encompassing restrictivity described by the equation: K=Q/sqrt(dP) or P=(Q/K)^2, essentially the flow rate squared relationship. It's probably your curve fit that I am seeing, it is not generating a constant. If it is real, that's very interesting, there's probably a way to extract a Reynolds number curve from it.

Sorry if I am being disruptive, I have not been following this closely but I thought the K-factor thing might be relevant. It for sure is another way to generate a PQ curve from a single data point...

[BillA]

Originally Posted by Incoherent

(from Nexxos thread regarding PQ)

Re the P=Q^1.85 fit, I think there is a better way.

I assume you are familiar with the idea of a "k-factor", I stumbled across it recently when exploring the flow measurement problem.
Your PQ curves are not behaving, shouldn't one be able to generate a constant which, even allowing for transitions between flow regimes and boundary conditions, is relatively flat across flowrate?. I am getting this with measured data, (see attachment) it is constant enough that I would propose that every block has a "K-factor", a constant encompassing restrictivity described by the equation: K=Q/sqrt(dP) or P=(Q/K)^2, essentially the flow rate squared relationship. It's probably your curve fit that I am seeing, it is not generating a constant. If it is real, that's very interesting, there's probably a way to extract a Reynolds number curve from it.

Sorry if I am being disruptive, I have not been following this closely but I thought the K-factor thing might be relevant. It for sure is another way to generate a PQ curve from a single data point...

yes, you are correct
I am in the process of converting all Swiftech flow resistance curves to a 'K-factor'

pH or Cathar
why not write an article on flow resistance characterization ? (yes, I'm lazy)

[pHaestus]

Bill I have a question about dP measurement. It seems to me that the easiest way to deal with all the different possibilities of tubing and fittings and such is to:

1) zero the xmitter
2) Run a length of the tubing to be used over the flow range of interest (blank)
3) Cut the tubing in half, mount block/rad, and repeat (sample)
4) sample - blank = dP of only block/rad (this takes care of the dP of crosses and fittings and tubing and all leaving only wb).

Is that unnecessarily complicated? I was concerned that the friction factor of clearflex or silicone would be difft enough from copper pipe to make doing a calibration with just copper pipes of difft ID to be a mistake.

[Cathar]

Hmmm, maybe some of you more engineering types could help me out to understand as I may have the following wrong (being self-educated and all on this stuff):

P=(Q/K)^2

seems to me to be simplification of Bernoulli's equation. However if we consider viscous head, being the energy that gets lost as heat (and absorbed into the liquid itself), then for turbulent flow the friction co-efficient of the viscous head term will fall away as the Reynold's number increases (which is proportional to the flow rate).

So what we then have is an effect where P=(Q/K)^2 isn't really followed cleanly. For the more restrictive blocks I intuitively would have thought that they would deviate even further from the classic P=(Q/K)^2 equation.

Now, quite true, Bernoulli's viscous equation consists of two primary terms, being the constant head loss due to inviscid flow (which is proportional to the flow rate squared), and the head loss due to viscous flow. The relative effect of viscous flow term diminishes as the flow rate increases, meaning that we end up with a curve that doesn't fit P=(Q/K)^2 any more.

I threw out P = Q ^ 1.85 as a really quick and dirty approximation of this effect, perhaps only valid for within 0.5 to 2.0x the data point given at somewhere like OC.com.

I measured various data points from 2.0 - 5.0LPM for the Storm/G4, and it didn't seem to me to quite follow a simple P=(Q/K)^2 curve.

That could also just be due to measurement error too.

Or perhaps one of you kind lads could tell me if viscous flow effects don't even apply here?

[BillA]

nothing is too simple once one gets into it

- the dP units are temp sensitive, if not temp controlled you'll need a zero reading at the start and end; use this to 'correct' the dP reading
(I would not suggest re-setting the zero, this is part of the unit's cal)

yes, your procedure is correct
I lay out the crosses flat on a table and connect them with a short piece of tubing, run a dP curve, then cut that tubing in half to connect wbs; then the single connection loss can be deducted from the wb reading at each flow rate

but you need to do this for each tubing size you will be testing with

[pHaestus]

What about the fact that the tubing has a curve in it when the wb is connected, but is straight through w/o the block?

[lolito_fr]

I threw out P = Q ^ 1.85 as a really quick and dirty approximation of this effect

Isn't that the same law used by the Hazen William equation?

[BillA]

Originally Posted by pHaestus

What about the fact that the tubing has a curve in it when the wb is connected, but is straight through w/o the block?

nooo, the cross must/should (?) be close coupled to the wb/rad and that connection tube kept straight
- a bend will have an effect

[Uncle`BuZZ]

Originally Posted by pHaestus

In MY opinion, the solution is to produce C/W vs flow data and dP vs flow data for waterblocks and a few common radiator/fan combos. Then to use that data along with the other components in the loop and the pump P-Q curves to predict a final C/W for the system. That is what groth was going for here:

http://pages.sbcglobal.net/water.gro...20process1.htm

The interface would have to be simplified a lot and it's awaiting my dP vs flow data...

It's waiting on my dP vs flow numbers to be generated; that's next up after this Alphacool nexxos XP review is done

I think we tried to do same thing in different way...


you can found my solution (version alpha) here (in french). I stopped development cause I miss data to put into my db...

you can define your loop with some tubing, radiator and waterblock, in series or what you want, you just need to drag elements where you want on the loop and see results...

You need the JVM from SUN and not the JVM from microsoft to use it. You could see some difference from your results because I probably haven't the same data of you (very few data in fact) but they should be minim.

In my dream I hoped to add some C/W data to do a prediction of performences, but I miss drop pressure data, so I don't imagine having enough C/W data... :shrug:

[pHaestus]

I am happy to open source all data. Let me get it in a reasonable format. I suspect dP vs Q numbers will be more useful to others and I'm in the process of fiddling w/ that now

[pHaestus]

Danger Den most graciously sent me a Thermochill 120.2 radiator (thanks guys!)
I will buy some Papst fans (anyone have a source for the ones used in these simulated loops?) and then try to reproduce some of the dT vs pump results using our modeled cooling loop.

[BillA]

good luck
correlating that systems stuff will be a chore, you'll have no 'free' time this xmas

[pHaestus]

I am a dummy. Danger Den had the 4312L's in stock; should have gotten them to drop a few in the box. I ordered them just now and should have them next week.

I'm in research scientist mode still: NO correlation is still worth knowing right? LOL

[pHaestus]

OK I am setting up the thermochill-based test loop tonight when I get home. I'll post an update when I get it running. The New AMD64 system I built for general testing will allow me to keep using my TBredB box for interesting stuff like this I hope.

[MaxxxRacer]

dont kill the winnie man...

[plywood99]

Originally Posted by pHaestus

OK I am setting up the thermochill-based test loop tonight when I get home. I'll post an update when I get it running. The New AMD64 system I built for general testing will allow me to keep using my TBredB box for interesting stuff like this I hope.

Hope this hasn't died down. Looking forward to the results...

[YEP]

Originally Posted by pHaestus

You guys are too damn clever. Now I suppose all that you need is pressure drop numbers for all the wbs I've tested right?

Not to be rude ..... sure "buddy"

This thread is really really nice. Good work you guys are doing.

When putting a new system together it's reall nice to have an idea how much pressure drops across a waterblock.

Again invaluable work you guys are doing for the community

[bobo5195]

Nice work, the computer model saves me answering many questions.

I havent been able to find this information anywhere else:
What are the assumptions in the model though.
Is the friction from the pipe based on tubulent or lamina flow and is the small diameter and possible boundary layer problems consider (ie has boundary layer thickness been considered as a check).
When comparing different widths of pipes is the head loss coefficent from varying diameter considered. For example thermochills are 3/8ths but piping might be 1/2 (this might be taken into account by bills testing) and the inlet and outlet to blocks or pump.