I'm confused, Radiator Flow Rate, More not always better?

[Erasmus354]

I am having a hard time figuring out what methodology coolingmasters is actually using to get their numbers. In their review they mention two different ways of doing it ... and then never say which way they are using.

[Martinm210]

I am having a hard time figuring out what methodology coolingmasters is actually using to get their numbers. In their review they mention two different ways of doing it ... and then never say which way they are using.

I think they are using flow rate and delta across the radiator to come up with the heat dissipated. Then they probably use delta from the air in sensors to the average of the two water sensors.

I still struggle with the whole Q = M X CP X dT equation myself with units but I think I got it.

To simplify it further and convert it to flow rate:

Q in watts = 263.43 x (Flow rate in GPM) x (dT in C)

So you need 263 watts to raise the delta 1 degree at 1 GPM

Or if you want to solve for dT

dT in Celcius = (Q in watts)/(263.42 x (Flow rate in GPM)

So for example, I've been testing with a 590 watt heat load at 1.5 GPM, so my dT would be 1.49 degrees across the radiator.

I think I need a beer, this stuff hurts..

[HESmelaugh]

Thanks for this, Martin.
I'm going to have to change the calculations in my recent tests again, then. I'm quite curious if it will make a significant difference in my results.

[Erasmus354]

I think they are using flow rate and delta across the radiator to come up with the heat dissipated. Then they probably use delta from the air in sensors to the average of the two water sensors.

I still struggle with the whole Q = M X CP X dT equation myself with units but I think I got it.

To simplify it further and convert it to flow rate:

Q in watts = 263.43 x (Flow rate in GPM) x (dT in C)

So you need 263 watts to raise the delta 1 degree at 1 GPM

Or if you want to solve for dT

dT in Celcius = (Q in watts)/(263.42 x (Flow rate in GPM)

So for example, I've been testing with a 590 watt heat load at 1.5 GPM, so my dT would be 1.49 degrees across the radiator.

I think I need a beer, this stuff hurts..

Oh I understand the math and physics just fine, always have. I don't find it that complicated either. Protip: Stop using imperial measurements. The calculations are much simpler if you use liters per minute. What I don't understand is whether that is actually what they are doing. A google translation of their methodology page:

This temperature difference between water input and output is only a few degrees at most at very low speed and a big power to dissipate. It is less than 1 ° C typically (flow> 2 L / min). With this method, it requires appropriate instruments to measure just 0.01 ° C or better. Indeed, every hundredth of a degree equivalent to several watts after calculation, especially where high-speed input-output gap is reduced, which reduces the sensitivity of the measure:
English version

The advantage of this method is that we know exactly the power dissipated by each radiator. There is no loss side or powers unknown variables introduced in the measuring circuit (eg a pump.). Here, everything happens only between the entry and exit of the radiator.

The method is to measure the temperature rise of water compared to air in a single loop, where we introduce a constant heating power, is another method. It is valid if we take some precautions because it introduces factors more or less unknown, but it does not require tools for high resolution. Each method has its advantages and disadvantages vis-à-vis what is sought and what was available. Either one works at constant power deltaT to find the water-air, or are working on water-air deltaT to find the constant power.

Notice how they are pointing out the need for highly accurate sensors to measure the way you describe, and they go on to mention the other way of calculating it (which is using the known heating element). They never actually state, or at least I can't tell from the translation, which of the two methods they are actually using. If they are using the second method I can see how they might have made the same mistake you did.

[Martinm210]

Thanks for this, Martin.
I'm going to have to change the calculations in my recent tests again, then. I'm quite curious if it will make a significant difference in my results.

It depends on the heat load and flow rate. A smaller heat load and small difference in flow rate may not amount to much, but it made a huge difference in my test where I ran a 600 watt heat load over a range of flow rates from .5 to 3.4 GPM. My original test was showing a 22% improvement at .5GPM vs 3.4GPM, but once corrected the higher flow rate shows a 2% improvement over the low flow rate. My results were really messed up, but all better now. I think I'll switch out one of my water sensors over the inlet side though regardless to start recording the average water instead of water out.

I'm thinking three simple points will do if the results are flat. Maybe one at .5GPM, 1.5GPM, and Max flow. This will help average out the three tests and give a good indication if there is any trends regarding flow rates.

Oh I understand the math and physics just fine, always have. I don't find it that complicated either. Protip: Stop using imperial measurements. The calculations are much simpler if you use liters per minute. What I don't understand is whether that is actually what they are doing. A google translation of their methodology page:



Notice how they are pointing out the need for highly accurate sensors to measure the way you describe, and they go on to mention the other way of calculating it (which is using the known heating element). They never actually state, or at least I can't tell from the translation, which of the two methods they are actually using. If they are using the second method I can see how they might have made the same mistake you did.

I see, yeah I'm not sure you can tell by the data presented, but I think they did it right.

My error was causing a very pronounced parabolic like curve starting high at low flows similar to what you see on a block C/W curve. Their data seems to be very different in that regard (more linear where it progresses downward) but there's not alot of number checking since they don't provide any of the data collection details...only the watts and deltas.

[skinnee]

I think I figured out the problem with my particular setup when trying to see flow rate effects:

A larger delta across the inlet and outlet of the radiator will net a lower "Water Out" temperature with lower flow rates. This is simply a function of the water's specific heat value and how lower flow rates does indeed mean a lower "Water Out" temperature, but if you were measuring the average loop temperature, THAT would be a different story.

What this means to me is that using "Water Out" temperature only in radiator testing while letting flow rate vary is BAD!!!

More appropriately, we should be measuring "Water Out" and "Water in" and averaging those two temperatures. This way the Delta across the radiator which is VERY dependent on flow rate is canceled out in testing.

That or you always need to test at exactly the same flow rate. That's not necessarily taking any pressure drop differences or flow rate performance differences into account, but it's at least canceling out the In vs Out delta dependency on flow rate.

Bottom line, measuring radiator performance must either fix the flow rate, or preferably measure and average the temperature of the inlet and out.


This same issue goes for CPU testing. Using the water out only is again going to create some skewing of results if flow rates vary and you compare water to core. If ambients were held exactly the same, the blocks that are the most restrictive will produce water temperatures from the radiator that are slightly lower, this is not so much due to lower pump heat dump, it's mostly because of the delta across the radiator.

Anyhow, I think I finally understand and learned some important test procedure things:

1) You should not test radiators with "Water Out" only and allow flow rate to vary. This will make the more restrictive radiators appear to perform better, when it's really just a specific heat and flow rate variable that's making the water out temperatures lower.
2) Radiator testing or comparing of water temperature should compare the water loop temperature average "Water In + Water Out/2". This will cancel out this delta T problem that's extremely sensitive to flow rate.
3) Radiator testing using "Water Out" should ONLY be used if you fix the flow rate constant for all tests. This still does not test the real world where the flow rate is dependent on the loop, and affected by the radiator pressure drop, but it's still at least eliminating the radiator delta problem.
4) CPU block testing - Again this "Water Out" comparison is not good with varied flow rates. Ideally you need to fix ambient and compare core to ambient only. If you fix the ambient temperature and compare core temps to water out temps, you'll artificially be making the lower restriction blocks look better because by nature of the higher flow rate the radiator water out temperature is higher. You will indeed get lower water out temperatures with the more restrictive blocks simply because of the lower flow rate across the radiator and larger delta across the radiator. To me this just means using "Water/Core" as a means to compare blocks is not an apples to apples comparison. Apples to apples is a direct comparison to ambient, and that will mean two slightly different water temperatures running across the block.

Anyhow, my testing setup is not working for flow rate testing. I either need to fix flow rates at 1.5 GPM, or if I want to see flow rate effects...I need to measure both inlet AND outlet water temps.

Regardless, I think the Coolingmasters test was good, so there is still some interesting thing happening...just not to the extent that my first test was showing.

I thought something wasn't right.....

I also realize that my early CPU block testing and comparison to water out temperature was WRONG! The rad testing is still ok though as I think fixing the flow rate at 1.5 GPM saves that one.

First, sorry for quoting the whole post, but I think this needs to be carried over on page roll. Also, glad I am fixing flow rate on all the radiator tests I do...whew, bullet dodged there.

As for CPU blocks: When you're looking to find a constant ambient but cannot keep the ambient dead on a specific temperature (my test lab has a 2ºC natural variance even with a dedicated HVAC zone in my house) you have two options to calcute the average core temps to a set ambient. (<- holy run on sentence batman)

Air to Average 4 Core dT + Set Ambient
and
Water to Core dT + Water to Air dT + Set Ambient.

Since we can't rely on the Water to Air and Water to Core numbers unless we test all blocks at set flow rates, you have to use Air to Average 4 core. Or I could be misinterpriting your #4. Without using an environmental chamber, true control of ambient is near impossible. Other possibility...I am completely missing something obvious.

I guess the only way to be sure is use both calcs to check your data calculations and also use two sensors for Rad Water Out.

edit: 1 more thing...varied flow rate testing at 3 seperate flow rates is a good idea. would you still calc down to an average C/W though for overall performance?

[Jah]

Good to see you figured it out.

My first assumption was increasing heat dump from the pump.
Assuming you have a varying flow rate through the pump, not 100% sure how you vary the flow. Increased flow --> more power required by the pump, and this should give you a larger heat dump. From my experience with pumps they draw much less current with no/low flow then on max flow, as you your self noticed when you created an overly free flowing system and tripped the over current protection.

On a side note:
Nice to see you back up and testing, even if its on a smaller scale, you have teched me allot about water cooling, keep up the good work.

[Martinm210]

First, sorry for quoting the whole post, but I think this needs to be carried over on page roll. Also, glad I am fixing flow rate on all the radiator tests I do...whew, bullet dodged there.

As for CPU blocks: When you're looking to find a constant ambient but cannot keep the ambient dead on a specific temperature (my test lab has a 2ºC natural variance even with a dedicated HVAC zone in my house) you have two options to calcute the average core temps to a set ambient. (<- holy run on sentence batman)

Air to Average 4 Core dT + Set Ambient
and
Water to Core dT + Water to Air dT + Set Ambient.

Since we can't rely on the Water to Air and Water to Core numbers unless we test all blocks at set flow rates, you have to use Air to Average 4 core. Or I could be misinterpriting your #4. Without using an environmental chamber, true control of ambient is near impossible. Other possibility...I am completely missing something obvious.

I guess the only way to be sure is use both calcs to check your data calculations and also use two sensors for Rad Water Out.

Yeah, I think it's still fine to do some correction on the data for varied ambient, but you might want to figure out the error per degree since processor heat generation varies with heat. This can be done by simply logging one long run over a large range of temperatures. I would just avoid using the water out number by itself for comparisons if flow rates vary. OR if you want to compare to water, it should be the average water temperature between inlet AND outlet.

I still kind of like thinking in terms of water out of the radiator temperature since that's whats running through our block if setup that way, but it's just not a good way to compare stuff since flow rate messes with it.

I think the key here is that any delta measurements from water outlet temperature are affected by flow rate. That problem is eliminated by referencing either the average of inlet/out, by fixing flow rate, or by referencing ambient instead.

For CPU testing I think referencing the ambient makes the most sense, this is also something more users understand than water deltas, although it does introduce more errors such as dust building up on a radiator and stuff like that you have to watch for also.

[MpG]

Heh, very interesting. A simple source of error, but definitely one that's capable of skewing results without anyone realizing it. I wonder if this issue is entirely responsible for the funny curves on the radiator performance charts?

[Bond Number]

Hi Martin,
I totally agree with your idea of taking into account flow rate. I mentioned this in one of HES’s posts.
http://www.xtremesystems.org/forums/...=220874&page=2 post #38
In HES’s test,he was plotting delta Ts by a constant wattage, so in that case you want to non-dimensionalize your delta T with some characteristic temperature (which takes into account flowrate). This doesn’t take into account all the flow rate dependence because the water side heat transfer coefficient can change with flow rate (boundary layer), but it will account for the m*Cp term. If you were to plot this dimensionless delta T for the same radiator at different flow rates any change you would see would solely be due to changes in the waterside heat transfer coefficient. You could also plot this for various radiators and/or radiator configurations as a function of flow rate and see which radiator scales the best with flow rate.
As for the average temperature difference, I see what you are trying to do. It is common practice to use a mean temperature when trying to determine the overall heat transfer coefficient of heat exchangers (usually a log mean temperature difference). If you were to plot your average temperature differences versus flow rate at some constant wattage one would see your average temperature difference decrease with increase flow rate (Ideally). You could do this for multiple heat exchangers and see which exchanger scales best with flow rate (which I think is the point of this exercise). But I guess I don’t see what is to gain with this method relative to a characteristic temperature. Looking at your method the only difference I see is measuring an extra temp.
As for the cool masters, their strange plots might have to do with error in their measurements. At low fan speeds the air is clearly saturated and no matter how much you increase the flow rate you will not reach a lower the water temp. But then why does the total wattage decrease? They have a calibration probe with an error of .0001C but they measure with probes of .02C. Because they are taking the difference they could have a max error of .04C. For every liter per minute there could be difference of 2.8W. So over the range of 10 L/min one could see a difference of 28W. When fan speed is increased the air is no longer saturated and we start to see increase dissipation vs. flow rate like we would expect. If there was no error in the probes this trend at higher fan speeds might be greater.

[twwen2]

You are closer to the truth than NaeKuh thinks! I wasted my breath on this one a long time ago, blocks may like as much flow as you can feed them but radiators certainly do not unless you ramp up the amount of air that you are horsing through the core.

Closed loops mate, i thought we settled this back yonder!

Nice to see the mind is still sharp Martin Good work on picking up on that inconsistency.

[hellcamino]

Closed loops mate, i thought we settled this back yonder!

Nice to see the mind is still sharp Martin Good work on picking up on that inconsistency.

I'm guessing you didn't actually start at the beginning of this thread and read it?

[shazza]

Great stuff here.

Would it be fair to say that the majority of us running just one pump in a loop are not going to experience any major issues, since we are working within a relatively narrow range for flow rate?

[hellcamino]

Great stuff here.

Would it be fair to say that the majority of us running just one pump in a loop are not going to experience any major issues, since we are working within a relatively narrow range for flow rate?

If your pc doesn't have overheating problems already then you won't ever experience any major issues due to this test:P

[shazza]

If your pc doesn't have overheating problems already then you won't ever experience any major issues due to this test:P

Thanks, hellcamino. Not too worried about my system, as I can run pretty consistent temps with varying fan speeds (CPU only on PA120.3). Really just looking to get confirmation that this is not a real issue most people have to deal with. I consistently see new folks believing their temp problems are caused by water flowing too quickly through the radiator.

[Erasmus354]

I dunno, it still doesn't seem right to me. From a pure thermodynamics and fluid mechanics standpoint it doesn't make sense. More flow means faster fluid velocity through the tubing. Faster fluid velocity directly correlates to a higher Reynolds number. This means less instance of laminar flow and therefore better heat transfer from the water to the tubing.

Now admittedly there are a lot of variables to consider here. Slower air flow means you need a higher fin/tube temperature to dissipate the heat load. The higher the temperature the less "eager" the metal is to absorb more heat from the water. And on and on, it is like a set of domino's one thing effecting another which effects another etc... Meh, all this thinking makes me remember a really good read about thermodynamics in respect to processor cooling.

BASIC PRINCIPLES INVOLVED IN HEAT TRANSFER AND COOLING RELATED TO PROCESSORS

This was referenced by Bill Adams in something I saw back in like 2004. Pretty interesting read for people who want to understand a bit more about the properties of how heat energy is moved through a system.

[SNiiPE_DoGG]

I initially think when reading these results that this is what is happening (alluded to by Erasmus)

Higher flow brings more heat to the radiator. This we know; but with the slow fans the heat is dissipated from the radiator less quickly. Thus the delta T between the Rad tube surface and the water is smaller and less heat is transferred away from the water.

[relttem]

mdot*Cp*(T2-T1) = h*A*(T2r-T1r)

[rosco]

There are principally 2 reasons to describe the small downturn at very low airflow and high water flow rate. Bond Number had some clues.

First, the whole measurement chain has a responsability. The bigger the water flow rate is, the bigger the error about heat dissipated will be. Even if the RTD probes for water are very sensitive (0.01 °C) and well calibrated against a quartz thermometer, the deltaT "water_out-water_in" become so small (~0.1-0.3 °C) at high flow rate that the smallest error on this deltaT leads to a bigger error when calculations are made. A 0.01 °C error or rounding seems to be very insignifiant, but at 10 L/min, it's not ! To compensate with the method used (deltaT water-air imposed), better probes and calibration are needed to achieve 0.001 °C resolution and precision. I let you imagine the price of such material and this requires specific environnement... Other parameters have also a very small inherent error : air is taken by multiple probes to 0.1 °C, water temp, etc. For instance, the chiller is very good but water inlet variation is about 0.03 °C, so we have to take a mean value between the min and max recorded by probes and that introduces an unavoidable error, very small but real. The same applies to outlet. All these tiny errors and roundings from measurement devices have less influence on the results at low flow rate especially (deltaT are greater, so relative error tend to be very small), but it's not possible to get the same confidence range for all possibilities, especially at high flow rate and low airflow. I won't show you the big excel sheet with all factors, averages and corrections parameters taken in account, but when we're dealing with global error bars, we can see how the deviation increases with flowrate. See below.



For a clarity purpose, these errors bars weren't put on charts. As usual, curves are "mean" curves by a statistic way (tests are done several times), their uncertainty is not shown (there's always one !). Anyway, the very high flowrate zone is the less important one because it's quite rare (let's say >500 L/h) and not really useful because gains are undetectable on a typical machine.


The second point is about some sort of saturation because of the air weakness. Air passing through the radiator will be warmer at the outlet, no mystery. From a thermophysical point, this air can't be warmer than the hottest point on the radiator. We can approximate that maximum saying it's the water temperature on inlet, let's say 30 °C. So, at best, air at the outlet will be at 30 °C too. Problem with very low airflow is that this air (20 °C at inlet) passing through the radiator will be warmed very quickly and it will hit a thermal wall without the possibility to absorb more energy (impossible to go >30 °C). Here, more water flow rate with an imposed deltaT water-air of 10 °C is directly equal to give more thermal energy to the rad. So, we inject more and more energy, but the small air volume provided by fans can't absorb such amount of energy (due to its limited heat capacity), the radiator efficiency is then +/- leveled, even if the convection efficiency in flat tubes is a bit improved. With the flow rate increase, more and more area of the radiator become "useless" in a manner of speaking, the fins tend to be at the same temperature than air surrounding them (all at ~30 °C), the thermal transfer tends to be very weak on these parts.

Someone talk about heat generated by friction into the rad due to its pressure drop, it's a nice try because it's real, but the power generated between the I/O is very negligible here (~0.9 W @ 10 LPM for the CR-22T). Heat dumps from fans are also to be neglected, especially at very low airflow.

We can suppose that the dissipation curve at very low airflow should be a horizontal line in the bottom right of the graph, like a asymptotic platter we can't cross, but it's a bit difficult to be sure because of multiple factors involved. Need a different way of measure and a cross-comparison to investigate this particular zone more deeply. In a typical loop, this problem can't be seen directly. At low airflow, water temperature will go higher to compensate the inefficiency.

At the end, the 2 classical ways of measuring radiator efficiency have to lead to same global conclusions with a proper analysis and modus operandi. Don't be too obsessed with ultimate accuracy, because all setups are flawed in a way or another, even with top notch equipment.

[Martinm210]

Thanks guys, awesome discussion, I've learned alot.

This is what I was seeing with my error in using water out only. It simply doesn't work right because it's affected by both flow rate and heat load. This also explains why I had a hard time with multiple heat loads as well. Using water out also tend to give a higher heat dissipated number than is probably correct.



I'm switching my testing over to measuring both inlet and outlet water temperatures, but I still plan to measure heat using deltaT because my probes aren't that accurate. Anyhow with this first test, I wasn't able to measure any downward trends once I figured out my error, only a very very small upward gain, but pretty small at only 2% difference between .5GPM and 3.4GPM.

[Erasmus354]

Great post rosco, thanks for the input. However, riddle me this:

So we are assuming that the air reaches water temp by the time it leaves the radiator. On any curve on the chart you have a constant airflow and a variable flow rate. You should as you said reach a point where the curve plateau's and it shouldn't go back down. At this point you have reached the limit of what the air can absorb. Since the airflow is constant, air in is constant (20C for arguments sake), air out is constant (30C - water temp), you should have constant temperature dissipation. You have the same flow of air through the same area with the same change in temperature. Yet obviously your measurements show that the water temperature is not decreasing by the same amount. It is decreasing less, so how is the air increasing by the same number of watts but the water is decreasing by less and less as the flow rate increases? You are gaining energy somewhere, it just doesn't make sense.

[nikhsub1]

Good to see you Rosco! You need to post more... Threads like these remind me of the old days (that is a good thing)

[relttem]

So we are assuming that the air reaches water temp by the time it leaves the radiator. On any curve on the chart you have a constant airflow and a variable flow rate. You should as you said reach a point where the curve plateau's and it shouldn't go back down. At this point you have reached the limit of what the air can absorb. Since the airflow is constant, air in is constant (20C for arguments sake), air out is constant (30C - water temp), you should have constant temperature dissipation. You have the same flow of air through the same area with the same change in temperature. Yet obviously your measurements show that the water temperature is not decreasing by the same amount. It is decreasing less, so how is the air increasing by the same number of watts but the water is decreasing by less and less as the flow rate increases? You are gaining energy somewhere, it just doesn't make sense.

This can be answered by that balance I posted:

h*A*(T2-T1) this is the air side heat transfer. IF, the air flow rate is the same, then h is constant, A and (T2-T1) are both constant as you mentioned.

This has to be balanced by

mdot*Cp*(T2-T1), which is the water side heat transfer.

mdot is your mass flow rate. Cp is going to be pretty constant at these small temperature changes. So, as mdot slows (T2-T1) (larger temperature difference) has to go up to balance the equation..or the other way around...as mdot gets higher your (T2-T1) (smaller temp difference) gets smaller..

[Erasmus354]

This can be answered by that balance I posted:

h*A*(T2-T1) this is the air side heat transfer. IF, the air flow rate is the same, then h is constant, A and (T2-T1) are both constant as you mentioned.

This has to be balanced by

mdot*Cp*(T2-T1), which is the water side heat transfer.

mdot is your mass flow rate. Cp is going to be pretty constant at these small temperature changes. So, as mdot slows (T2-T1) (larger temperature difference) has to go up to balance the equation..or the other way around...as mdot gets higher your (T2-T1) (smaller temp difference) gets smaller..

Which is exactly what I am getting at. The situation as described by rosco has the low speed fans saturated from a certain flowrate onwards. This means that the airflow half of the equation is constant. The problem here is that the chart shows the water half of the equation going down .... how?

[skinnee]

Great post rosco, thanks for the input. However, riddle me this:

So we are assuming that the air reaches water temp by the time it leaves the radiator. On any curve on the chart you have a constant airflow and a variable flow rate. You should as you said reach a point where the curve plateau's and it shouldn't go back down. At this point you have reached the limit of what the air can absorb. Since the airflow is constant, air in is constant (20C for arguments sake), air out is constant (30C - water temp), you should have constant temperature dissipation. You have the same flow of air through the same area with the same change in temperature. Yet obviously your measurements show that the water temperature is not decreasing by the same amount. It is decreasing less, so how is the air increasing by the same number of watts but the water is decreasing by less and less as the flow rate increases? You are gaining energy somewhere, it just doesn't make sense.

Air temp would be equal to water temp if you have full (100%) water to air heat transfer, meaning the radiators are 100% efficient. I am running into weird numbers on my second round of radiator reviews because I am seeing 100%+ effeciency of water to air heat transfer, and I am controlling flow.

So I am going back through and using the water in sensor in my water average numbers for calculated C/W. The water temps do rise (~.2-.4C, thus far), but this does change the calculated C/W. I do not have enough data sets converted yet to show a C/W chart for any of the 6 triples tested. I will share the differences as soon as I can get at least one radiator using the new formulas.