Hoses, WB's, Pumps, and Performance???

[Dix Dogfight]

Alot of the dabate here has been around the issue of compairability of the different tests across the globe.
The problem with compairing person X results with Y's is the calibration of the measurement setup.
I've bee thinking about it for a while and this is what I've come up with as a part of a solution. Using a die-simulator and doing the measurements this way:

We want to measure how "good" the thermal interface between the cpu and WB is.
The problem with measurement of a WB is to know the exact amount of applied power and the effect of the pump/rest of system.
Therefore we must calibrate the setup.
First measure the exact amount of water in the system, se picture 1, we call this volume V.
Then run the pump without the cpu simulator on. For a half hour or so and measure the temps every 30-60 seconds.
The total time is depending on the amount of water and the C/W of the box/hoses. But the important part is that you measure untill the temp is stable.
This gives you a calibration curve/plot to correct our measurement with. The plot will/should look something like picture 2.
Now we can calculate how much power, P_pump, the pump adds to the water by using the slope, K_pump, of the curve in the beginning of the measurements and the measured amount of water V.

P_pump = K_pump * theta * rho * V (Watts)

(rho = 1kg/litre = desity of water)
(theta = 4180 J/(kg*C) = specific heatcapaciy of water)

A typical pump (eheim 1250) consumes 28W. With a setup with one litre this should give us a slope of K_pump=0.0067 C/s.
1/K_pump=150 s/C = it takes 150 seconds to heat one litre of water one degree celsius.

Now wait untill the water reaches ambient again and redo the temp measurement but with the cpu simulator on and measure the power it consumes.
P_cpu=U*I

Now we know the total applied power P_total=P_pump+P_cpu and the WB C/W of the system is easily calculated using the equilibrium temp at load.
If we take the slope in the beginning of the load measurement, K_load and subtract K_pump we get K_wb.
To calculate the C/W of the WB we need the temp at the cpu simulator an ambient temp:

(T_cpu - T_ambient)/((K_load - K_pump)*4180*rho*V)=C/W of the waterblock


If we want to compare different measurements from different places/persons we do:
EXAMPLE
Lets say person X measures a MAZE and a SPIRAL and person Y measures a MAZE and a SWIFTECH. Both persons has done the measurements the described way.

Person X:
MAZE_X=0.40C/W
SPIRAL_X=0.35C/W

Person Y:
MAZE_Y=0.55C/W
SWIFTECH=0.50C/W

Which is better SWIFTECH or SPIRAL?

SPIRAL_X/MAZE_X =0.40/0.35=1.14
SWIFTECH_Y/MAZE_Y=0.50/0.55=1.10

Since the SPIRAL har a relative difference that is bigger than the SWIFTECH the the SPIRAL is better.
(PLEACE NOTE THAT THESE NUMBERS ARE MADE UP AND IN NO WAY RELATE TO THESE WATERBLOCKS, i simply stole the names to make it easier to read)



The same setup can be used to measure different radiators. By simply exchanging the WB with a rad and heat up the water to, lets say, 60 degrees celsius.
Do the measurements like above, with/without fan on rad, to get the K values, the slopes in the beginning, and then calculate the C/W th same way.
(what about the fan some might think. Well if you use a monster fan that "saturates" the rad = any higer airflow won't make any difference,
then the rad is working under optimal/best conditions, then the airflow will not reflect on the measurement).



What do we need for this to work = conclusions :
To compare we need a WB/rad that both persons use as referense.
Since all comparisons are relative the difference between systems are irrelevant.
Since the calibration is made, the relative K_WB is indifferent from different setups.
We need good isolation around the reservoir to get a big delta temp.

Now I'm beginning to get tired and I've probably missed something somewhere but please look at the idea as a whole.

[Dix Dogfight]

forgot pic 2

[pHaestus]

Dix:

Yes I agree that you have to have a block or two as "standards" that you retest every time you break down/upgrade/change your system. I used a GLaciator2 and a Dynatron BH610 for this purpose when testing heatsinks a while back. It is the only way to have any confidence that your numbers are not being affected by some experimental error. Your proposed setup isn't so different than Bill's except instead of an insulated box he is using a lab chiller for his submersible pump and controls the set temperature so that it always results in an inlet water temperature that is constant. You'd really need more temp probes but the advantage of your design is that you could cheaply find a digital thermometer with type Ks that has two inputs (Ex Tech?). The problem is the time required (I suspect) to come to an equilibrium, and the fact that different blocks will have different restriction amounts, and so the pump will generate different amounts of heat (how to correct?). Flow rate data still needed, and also much more tedious for this since your pump will become more and more important in heat generation.

The thing that Bill and I have been kicking back and forth here is that most of the die simulators (Bill's excluded) seem to still have substantial secondary losses as well (or else their calculated W isn't much better than radiate's). So you are back to only internally valid numbers; no comparison between testers is good since their error in W created vs. W applied is all going to be different anyway.

No comparison between different testers is ever likely to be good; testers might as well just shoot for internally good numbers and then explain their methods. The TIM between die and block is a huge variable, as is (for most) mounting pressure. I can probably get a consistent paste application over time, but I wager that it will never be the same thickness settled upon by other testers.

[Dix Dogfight]

pHaestus:
Nice to see that somebody acually took the time to read it :-)

I agree on the timeconsuming aspect of it all, but if you choose a small amount of water. Then it will go faster.

However I think you missed one of the points with this setup, or i explained it poorly.
So here is another try:
1* Since you calibrate each WB to get the certain K_pump for that particular WB/restriction then the differences that occur in the pump gets cancelled out. No flow measurement needed!.
If you decide to change the flow with a restriction the you must recalibrate.

2* As you see i can extrapolate the power the pump is adding to the system. By measuring the deltaT, the time and the amount of fluid. This calculation is also possible on the second measurement and gives the total power added to the water by the simulator and the pump. Subracting the total power with the pumps gives you the exact power that the WB recives and transports to the water. If you at the same time meassure U and I to the simulator (SIM) you can caculate the losses from the SIM to the secondary heatpaths.

3* If we would isolate the WB and SIM from the sourrundings with a box the the WB would collect heat/power from the hot air inside the box which it normally wouldn't and it would perform better than in a real world application.
However you are correct that there is no way to compensate for the differences in secondary heatpaths with this setup and a real CPU.



More comments/ideas/thoughts anyone.

[pHaestus]

What I am saying is that the waterblock itself is a significant restriction, and will cause the pump to add a different amount of heat to the water for every single block tested. You can't account for this by testing the pump in the absence of the block; you can only hope to account for this by adjusting flow and monitoring it with a fair degree of accuracy. Bill might have some estimates on how much more heat a pump adds to water when throttled (actually I wager he does). The accurate measure of power into the system (from die and from pump) becomes a serious problem here IMO.

[Dix Dogfight]

Read it again and you will se that the WB is there during the calibration and therefore any difference in the pump between two different WB is cancelled out.

The only difference between the actual measurement and calibration is SIM off or on. No changing any physical parameters.
Thats why its called CALIBRATION.

[pHaestus]

Ah ok. Gotcha. I still think you would have to do multiple flow measurements or else you would get skewed (pump dependent) results in terms of relative performance. I have another question though.

If you have a constant heat input (a CPU die real or simulated transferring heat to the water through a block) and a completely sealed and insulated reservoir (that contains water and a pump), aren't you going to just constantly add heat to the system until parasitic losses take over? What makes you think this would be a situation that you could easily find an equilibrium value? I would think for equilibrium measurements you would have to have a heat exchanger for the input of heat to come to equilibrium with or else wait REALLY long times (and end up with extremely high water temperatures)?

[Dix Dogfight]

The point with the islolation is to delay the parasitic heatloss to the surroundings.
The numbers we are interested in lies in the begining of the measurement = gradient of the slope. And not in the end where we reach equilbrium. The isolation makes th initial gradient appear for a longer time and therefore contributes to a better result.

If we, lets say, add a rad to the system it will reach equilibrium much faster and at a lower temp which gives us less data to correctly approximate the slope in the beginning of the measurement. A chiller is the total killer for this type of measurement because it regulates the water temp actively and that way destoys the gradient we are interested in.

[BillA]

Dix Dogfight

don't think so, your method is mass dependant and a wb could be designed just for the test
look at some of Jeremy's comments on OCAU re just this point
AND you are summing the effects of a number of variables; are you quite prepared for the math involved with multivariate analysis ?
- then you would have to validate your test by testing just that parameter under observation

why not test just that variable to begin with ?

and I REALLY do disagree with looking at a transient response, not the way its done

but give it a whirl, try it

a question: can a circuit analysis be done with no reference to current ?
nor can you evaluate a heat exchanger without reference to flow

[Dix Dogfight]

unregistered:

Well the mass or volyme of the water needs to be measured to extrapolate the C/W but that is the total mass of all the water in the system so i can't se how you can produce a WB that works better with more water in the system.

The problem all you WB testers/reviewers are facing is a combination of poor equipment and difficult testingparameters.

What do we need do calculate the C/W which is the common way to compare the performance of different WB.
We need the Power/heat that is injected into the water through the WB and the difference in temp.

Injected power is hard to measure because you don't know the secondary cooling (mobo and air around socket).
Temp is hard to mesure because we can't afford to by calibrated instruments etc.

So we use poor tempmeasurment devices and guestimates the powerinput.

Now tell me. Where does the measurement of the flow make any of those measurements better?

The setup I am proposing here uses the same crappy temp measuremend device but by first calibrates the setup and then make a relative measurement. Any temp offset/unlinearity in the probe will be cancelled out. As a bonus it is possible to extrapolate the exact amount of incected power into the water. And that also tells you how big your secondary heat/power-losses are if you at the same time measures U and I to the SIM.

"Transient response, not the way to go."
Why not when it works?

Some one once said:
If it's stupid but works it ain't stupid

[BillA]

different situation here
don't use a CPU to test with
have good equipment
its calibrated

I'm not into guessing, I measure

"The setup I am proposing here uses the same crappy temp measuremend device"
not worthwhile, gigo
can't make a silk purse from a sow's ear

if you already know its crap, dump it for something that will be effective

but you need to address the methodology

[Dix Dogfight]

The methodology is EXACTLY what iv'e adressed with this proposal.