Watercooling Duallies, Series vs. Parallel

[Colin]

Today I tried some wyes to split the flow before and after the waterblocks in my dually rig. My temps went up 3C. :( The system consists of a Danner Mag 7, Maze 1Cs and a Cooling Cube. My guess is cutting the flow in half to the waterblocks is the reason. The Maze 1Cs are known to be very sensitive to flow. To make things work I will have to dust off my Mag 12 or try a low restriction heater core, perhaps both. In the meantime, I'm going back to a series setup and get on with life.

Wyes have worked well in other folks systems. If you are going to try them, make sure the rest of your system is up to the task. I suspect they would work much better with a block that does not depend on high flow rates like the Maze 1C. The original Maze block may be a better option for this.

[The Overclocker]

you could try having everything except the water blocks in 1/2 inch and the blocks in 3/8. or have the rad going to a res with two small pumps each providing one block with water

[BillA]

nothing good about reducing the flow rate through a wb

not only is the quantity of the heat transport "media" being reduced,
so is the turbulence within the wb (and channels if present)

with 2 wps in parallel the flow resistance that the pump sees will be reduced and the system's total flow will increase
- but never enough to offset the reduced effectiveness of the wb's performance

not an opinion, have a ton of data

be cool

[Colin]

Thanks Bill, now I have a little data too. ;)

[Mictlan]

Some hydrolics might give you some good starting point for the tweaking of your dual system (I'm currently going thrue the design of one of this systems, more of this latter).....

A centrifugal pump, like the ones we are using (you can tell them appart as they suck the fluid (water) thrue the center of the pump and the outlet is tanget to the outside of the pump), have a curve of operation. This curve is between the flowarte and the head. The head of the pump is how much feets the fluid will be pumped straight up, this tells you how much force the flowrate has.....now when the fluid has travelled that amount of feet, it "spends" that power to defeat gravity. When you are saying that a pump will lift 5 ft 10 gpm of water, you are stating one point of this curve.

But in an hydrolic system (like the watercooling system: pump->waterblock->radiator->reservoir) fluid power is not just used to defeat gravity, is used to also friction. Every turn, every change in the diameter, every lift expends the fluid power delivered by the pump......even straight pipes causes some amount of friction that the fluid must overcome to flow.

All this can be seen with the pressure of the fluid. If you install a manometer at the outlet of the pump, and then another one at the end of circuit, you can messure this loss of power with the loss of pressure. When you have a fixed set-up, that produces a certain drop of pressure.....that can be checked at the pump curve and the flowrate that pump will deliver can be obtained.

When you add more equipment (like a division and a mixer) you increase the pressure drop of the system.....

Now onto some more quick quirks of fluid flow:

If you have two interconnected systems (in a Y fashion) and install a flow meter at the end of each one of the branches, you can see the branch that drops the less pressure will have more flow than the one with higher pressure drop (this is know as the less resistance of flow principle).....so in a dual watercooling system, both waterblocks have to have the same lenght of pipe and the same waterblock to have aprox. the same from to the waterblocks.....remember that gravity works also in the system, so if the two waterblocks are at different positions, you'll have to make some trimming in the lower of the two systems to equal the pressure drop.....

Taking all this into consideration is very importan if you plan to have a one pump-two waterblocks in paralell system.....It will be easier if you just use two pumps with two waterblocks in two independent circuits, that could dump all the water into a single reservoir.....One idea I'm playing with is the follwoing setup.....and how to squezze it into a full tower case. I might use only one pump for the two waterblocks but I'm in the design process, and right now I'm saving to buy 2 PIII 1.26-S chips.....

[BillA]

Colin

did you ever swap the hoses to see where the temp diff was ?
(I'm always curious)

Mictlan

slice the loaf as you wish, but of this I can assure you:

less flow through the wb = higher die temp

true the difference may become small (say the increment between 3 and 4 gpm - if one has such a pump);
but the difference between 0.5 and 1.0 gpm WILL be noticable

and all those pumps will be adding their heat contribution too, eh

be cool

[Mictlan]

The pumps are going to be inline pumps not submersible.....:) I have already thought that 2 or 3 pumps running inside the reservoir will be too much heat to handle.

AS for the problem that Collin has, if he has the money he can do the following: increase the diameter of the piping grom the pump to the waterblock, that might increase a little the flowrate to each of the waterblocks.....but surely you'll be needing to have the same kind of wb in each circuit.

[Aesik]

The water temperature increase after flowing though one block is only going to be in the range of .05 to .15 degrees C at best. This isn't going to significantly effect the ability of the second block to transfer heat to the water.

There is still only one radiator, so regardless of whether or not the blocks are in series or parallel, all the heat has to be dumped through the single radiator.

It would be much simpler to just put the blocks in series. I see more harm in having multiple pumps for multiple loops. Even if they are in-line, they will still be dumping some amount of heat into the system. The only way I could ever see justifying multiple pumps would be if one pump simply can't supply enough head to get the desired flow rate. In this case it would most likely be better to upgrade the single pump.

[Colin]

Bill - I did not try swapping the hoses. While I can't say for sure (strange things do happen), I doubt there would be a change. My conversion from pelted Athlon rigs to Dual PIIIs has given me a tast of real hassle free speed. I think I will just park and enjoy using this box instead of tweaking for a while. :)

[UserName]

To clafify one small point in Mictlan's excellent post. Closed loop systems do not fight gravity, like open systems do. Head is not a factor in the traditional sense.

Although Aesik needs no credibility enhancement, i second what he said.

[Colin]

Originally posted by Aesik
The water temperature increase after flowing though one block is only going to be in the range of .05 to .15 degrees C at best.

This depends on the wattage of the CPU. I have not done before and after the block temps with this system. My current CPUs are putting out a little over 47 watts. From my experience with water temps before and after the block on Athlon systems, I would estimate the temperature of the water after the first block at about 1.5C to 3C above the inlet temp depending on the CPU load.

[UserName]

if you measure it before and after you will see it's nowhere near 1C

theres a simple way to get at the number.

figgure out your flow rate and see how much water flowes over your CPU in a second. then figgure how much 47 watts could heat that water.

no wear near 1C

[Colin]

It's not that simple. You need to factor in the suface area of the block and the turbulence. Given the fact tha CPU2 always runs 2C to 3C higher than CPU1 at full load, it's obvious the water is heating up more than 1C. Also the VP6 uses in-socket thermistors so there is compression in the CPU temperature measurements.

[UserName]

it might not be so obvious. i have seen many systems with 2CPU's that have different temps. i think there is a lot of varance in the thermister.
But you can controll for it by reversing the flow thru the blocks

but i am telling you that there is no way you could raise the water temp by 1C. it's just not possable with a little 47W heat source.

[UserName]

did some quick math

169000J = 47w*H
assuming 8 pounds per gallon
assuming 100GPH = 800lbs=298.6kg

Specific heat of water 4186

169000/298.6=566.64

566.64/4186 = 0.13C

So maybe your CPU's are putting out 470W. that would get you 1.5C

Aesik;
Have you done this calculation? Cause thoes numbers are so close...

[Les56]

Originally posted by UserName
did some quick math

169000J = 47w*H
assuming 8 pounds per gallon
assuming 100GPH = 800lbs=298.6kg

Specific heat of water 4186

169000/298.6=566.64

566.64/4186 = 0.13C

So maybe your CPU's are putting out 470W. that would get you 1.5C

Aesik;
Have you done this calculation? Cause thoes numbers are so close...

I first saw these sums calculated by Billa, in this thread:
http://hardforum.com/showthread.php?s=&threadid=276693

[UserName]

WOW

nearly a year ago.

We need a FAQ.

[BillA]

hey fellas

there's a whole set of actual numbers here (http://forums.overclockers.com.au/showthread.php?s=&threadid=26546&pagenumber=5)

other than your's truly, I know of no one "out there" taking coolant temp measurements to 0.01^C
- and that IS what's needed

be cool

[UserName]

awww bill i got to register to see those numbers and thats not gonna happen.

i just thru the numbers out to show that it's just not possiable to get a rise in temp that was being discussed, even more than .5C. the exact amount of the rise really wasn't the point. i guess i shoulda made that clear.

I didn't mean to say that's how much it's rising or anything like that. i was just saying that even if all the heat went into the water and you only had a 100GPH flow the max it could rise in one CPU was really small. Really really small.

So chaining CPU's is a non issue.

Heck, chaining everything in your system, from PS's to chipsets aint gonna really effect the water temps in the chain.

[BillA]

eh ? what ?
"i got to register to see those numbers"
I can fix that
quote:

load W . . flow . . die T . . bp T . . inlet T . outlet T . . C/W . . Btus*
. 45 . .2.0/7.57 . . 36.5 . . 27.1 . . 25.00 . . 25.06 . . 0.256 . . 108
. 45 . .1.5/5.68 . . 36.7 . . 27.3 . . 25.00 . . 25.09 . . 0.260 . . 121
. 45 . .1.0/3.79 . . 37.2 . . 27.7 . . 25.00 . . 25.15 . . 0.271 . . 135
. 45 . .0.5/1.89 . . 38.2 . . 28.7 . . 25.00 . . 25.31 . . 0.293 . . 139

. 75 . .2.0/7.57 . . 45.0 . . 29.4 . . 25.00 . . 25.12 . . 0.267 . . 216
. 75 . .1.5/5.68 . . 45.4 . . 29.8 . . 25.00 . . 25.16 . . 0.272 . . 216
. 75 . .1.0/3.79 . . 46.1 . . 30.4 . . 25.00 . . 25.26 . . 0.281 . . 234
. 75 . .0.5/1.89 . . 47.8 . . 32.0 . . 25.00 . . 25.54 . . 0.304 . . 243

105 . .2.0/7.57 . . 53.3 . . 31.6 . . 25.00 . . 25.17 . . 0.269 . . 305
105 . .1.5/5.68 . . 53.9 . . 32.1 . . 25.00 . . 25.24 . . 0.275 . . 323
105 . .1.0/3.79 . . xxxx . . 33.0 . . 25.00 . . 25.37 . . xxxxx . . 332
105 . .0.5/1.89 . . 56.9 . . 35.2 . . 25.00 . . 25.75 . . 0.304 . . 337
* sorry for the archaic units, equation given here on pg 1

a note re the Watts, these are from a very controlled heat die - and quite accurate with NO secondary path losses
what these temps suggest, indeed prove, is that the utilization of a CPU on a mobo as a "heat source" is just as pointless/inaccurate as using the on-board thermister for "temperature measurements"
-- furthermore that these reported "100W" CPU numbers are pure smoke

I have pages of this kind of data, and pages of analysis - it's quite consistent
(this is in-process as I am still formulating the assembly and test procedures to ensure the comparability of the data)
the next article is to be on wb C/Ws, and influencing factors
the second on wbs themselves

comments solicited
unquote

be cool

[UserName]

what is BP temp?

[BillA]

bp = base plate of wb, drilled for a TC per the AMD spec
(#53 drill hole, 2mm on center "up" from face, to center of die area)

be cool

(I've never heard of a forum that one had to register in to READ)

[UserName]

me neither.

and we have completly hijacked this thread. hooray for us!!!

AMD has a WB standard?

[Colin]

Thanks guys, as I remove my foot from my mouth. ;) I am glad I'm happy with my system as it is. Putting my home made inline temp sensors in would be a waste, of time, especially now that I understand they must not be very accurate. Perhaps I should try putting both of them in a bowl of water to see if they read the same.

[BillA]

"completly hijacked this thread"
not so sure
Colin and I have been chasing die temps for several days on different forums

AMD's cooling doc #23794 (http://www.amd.com/us-en/Processors/DevelopWithAMD/0,,30_2252_739_2983,00.html) says how to instrument an hsf bp

be cool

[Les56]

Originally posted by BillA

"

load W . . flow . . die T . . bp T . . inlet T . outlet T . . C/W . . Btus*
. 45 . .2.0/7.57 . . 36.5 . . 27.1 . . 25.00 . . 25.06 . . 0.256 . . 108



a note re the Watts, these are from a very controlled heat die - and quite accurate with NO secondary path losses
what these temps suggest, indeed prove, is that the utilization of a CPU on a mobo as a "heat source" is just as pointless/inaccurate as using the on-board thermister for "temperature measurements"
-- furthermore that these reported "100W" CPU numbers are pure smoke

be cool

Billa,
Are you saying that the heat absorbed by the water is less than that produced by the "heat die"? As shown in e.g. line one of table:
108 x 0.2931= 31.7 = 70% of the heat imput of 45watt.

Or have completely misinterpreted the data and misunderstood this point.

[ken257]

I had 2 Abit BP6's in the years past. One had dual 366's and the other dual 500's. On both boards one cpu always ran a few degrees hotter then the other, didn't matter if it was air cooling or water. The in socket thermistors must have a +/- a few percent tollerance.

I ran water on both boards in series and never had a problem. The 500's went to 600 and the 366's topped 600 with 80w pelts. I never had a problem with water temp when using the pelts with in series blocks. Just need to have a good strong pump with lots of flow and a non restrictive rad. The problem with Y's is that even though there will be less resistance and more "overall flow" there is only 1/2 the amount of water going through each block. The high volume is more important then the slight temp difference that may occur with the blocks in series.

Bottom line if it works and is overclocked to a satisfactory level you are good to go. Don't sweat a little temp dif between the cpu's the error just may be in the thermistors.

Here is a quick way that I measure water temps. I see a lot of faq's and people making inline temp sensors which I don't see the point of. I just take a com-u-nurse and tape the probe to a metel fitting. You can tape a little insulation over it to if you like. The metel stays at the same temp as the water that flows through it in the worst case you may just get a little lag in the reported temp in periods of rising water temp.

[BillA]

ken257
with all of the discussion re thermister measurement "difficulties", it's quite hard for me to understand why one would quibble over a few degrees

ahh, Les
someone who reads the numbers and thinks about them, thanks

as the old saw goes, the numbers speak for themselves
and here my lack of formel training in thermo leaves me unable to state exactly why

the Btu equation is the same as that used in the rad article, and is widely used in industry to calculate the work done by boilers, hot water heaters, etc
- it is obviously a "net output" calc as the input values are not considered, and a calc such as yours will indicate a gross efficiency
- a heat balence calc would account for "all" of the heat

my description of the heat die's having "no secondary (heat) path losses" is somewhat relativistic as I'm comparing it to a CPU in a package, mounted in a socket, on a motherboard, etc
- the die's copper slug (in which the TC is embedded in a step on a side of the raised die face) is encased in a phenolic enclosure with air gaps, etc; which is then covered on 3 sides and the bottom with 1 1/2in. of styrofoam, and a rigid urethane foam shim is used between the phenolic top and the wb bp (there is still a slight air gap between them)

a considerable portion of the "missing" heat presumably is represented in the temp rise
- but a puzzlement to me is just why the "efficiency" is inversly related to the flow rate

perhaps Aesik will shed some light on this for us
or even my French friend Antoine

be cool

[Aesik]

Before diving in too deep, let me toss out the governing equation for calculating the fluid temperature rise:

qconv=mdot*Cp(Tout-Tin)

qconv=heat transfer by convection
mdot=mass flow rate
Cp=specific heat of the working fluid
Tout=temp of fluid coming out of the channel
Tin=temp of fluid coming in to the channel

Rearranging a bit, if we just want the temperature rise:

Delta T=Tout-Tin=qconv/(mdot*Cp)

It must be remembered that there is one HUGE assumption that has to be made to use this equation, and that is that ALL of the heat from the die is being transferred into the fluid. Of course this is not true in any case as heat is being lost to parasitic sources all over the place. Conduction into the socket/motherboard/etc, convection from the block itself to air in the case, radiation heat loss, etc. etc. We are also assuming that the die is a consistent source of power output and of course this isn't quite true either. One last assumption made is that the system is in equilibrium and the block temperature and Tin are no longer fluctuating.

What Bill has done with his test setup is attempt to control the exact and consistent power applied and to eliminate any heat loss to other sources than the water. He has done a great job with is setup (IMO) but there are still going to be some parasitic heat losses, thus the descrepency between exact wattage in and exact heat out. The next big step to improving a test system like his would be to suspend the heat source and the water block in a vacuum chamber.

My whole thesis was testing highly conductive composite thermal links for which I had to test the pure conductivity of. Thus I can attest to what a major pain in the ass it is to attempt to eliminate all sources of parasitic heat loss. I was using a kevlar suspension system (very little conduction into the system from outside sources) to hold the link in a dewer filled with liquid nitrogen on one end, and pulling an extreme vacuum. The thermal link was also surrounded with multiple layers of MLI (Multi-Layer Insulation) that is used to shield satellites from radiation heat transfer. This system came close to isolating the thermal link from parasitic heat losses, but it was still not perfect.

As far as to why the efficiency is inversely proportional to the flow rate in Bill's test, I have a few speculations but would have to see more data before I'd dare come to much of a conclusion.

[Aesik]

Just as a quick side note, I'm tossing together a quick and dirty spreadsheet that will do some VERY simple calculations when making LOTS of assumptions. But it will be a decent way for people to get an idea of how changing certain variables (flow rate, channel size, etc.) could effect their system.

I have several more changes to do to it, but I'll post it up soon if anyone wants to play with it.

[BillA]

I have the power input pretty well controlled and quantified
(SOLA transformer, DC regulated power supplies, mA and mV simultaneously monitored, Wattage coutinously calculated)
- and all measurements made ONLY after equilibrium established (amb controlled, etc. etc.)

I personally am not so interested/concerned with the T calcs, I can measure those of interest to me
however

I AM MOST INTERESTED in the Btu business

what data do you need ?
have flow rates form 0.5 to 5 gpm, power from 40 to 130W applied
as well as data at 4 different wb/die applied load (force) levels

given the calculated results, I'm not so sure just what it is that's being calculated ???

given the small deltaT, the wb losses will not vary enough with the changed flow rate to affect the Btus significantly

what DOES change with the flow rate is the setpoint of the bath temp to hold the inlet temp ABSOLUTLY constant (25.00^C)
- as the flow rate decreases, the temp setpoint must also be decreased; and by an ever increasing amount at the lower rates

to me this says that the bath is having to remove MORE heat from the coolant as the flow rate decreases

obviously I'm missing something here

be cool

[LegumaN]

Thx Bill for invited me to this thread ( I'm the frog's legs eater ).
Bill could you translate the btus in watts please ?

Strange things on the pointed lines, why btus are constant ?
That's not coherent with the other measurments !

load W . . flow . . die T . . bp T . . inlet T . outlet T . . C/W . . Btus*
. 45 . .2.0/7.57 . . 36.5 . . 27.1 . . 25.00 . . 25.06 . . 0.256 . . 108
. 45 . .1.5/5.68 . . 36.7 . . 27.3 . . 25.00 . . 25.09 . . 0.260 . . 121
. 45 . .1.0/3.79 . . 37.2 . . 27.7 . . 25.00 . . 25.15 . . 0.271 . . 135
. 45 . .0.5/1.89 . . 38.2 . . 28.7 . . 25.00 . . 25.31 . . 0.293 . . 139

. 75 . .2.0/7.57 . . 45.0 . . 29.4 . . 25.00 . . 25.12 . . 0.267 . . 216 <---
. 75 . .1.5/5.68 . . 45.4 . . 29.8 . . 25.00 . . 25.16 . . 0.272 . . 216 <---
. 75 . .1.0/3.79 . . 46.1 . . 30.4 . . 25.00 . . 25.26 . . 0.281 . . 234
. 75 . .0.5/1.89 . . 47.8 . . 32.0 . . 25.00 . . 25.54 . . 0.304 . . 243

105 . .2.0/7.57 . . 53.3 . . 31.6 . . 25.00 . . 25.17 . . 0.269 . . 305
105 . .1.5/5.68 . . 53.9 . . 32.1 . . 25.00 . . 25.24 . . 0.275 . . 323
105 . .1.0/3.79 . . xxxx . . 33.0 . . 25.00 . . 25.37 . . xxxxx . . 332
105 . .0.5/1.89 . . 56.9 . . 35.2 . . 25.00 . . 25.75 . . 0.304 . . 337

[BillA]

so much for thinking things were "under control"
thanks Les for making me think about all this a bit harder

recalibrated everything and found that my coolant outlet temps were reading 0.02^C low
so the corrected table is below and expanded per Les's comment:

load W . . flow . . . die T . . bp T . . inlet T . outlet T . C/W . . Btus*. . Watts . %eff
45.00 . .2.0/7.57 . . 36.5 . . 27.1 . . 25.00 . . 25.08 . . 0.256 . . 144 . . . 42.2 . . 94
45.01 . .1.5/5.68 . . 36.7 . . 27.3 . . 25.00 . . 25.11 . . 0.260 . . 148 . . . 43.4 . . 96
45.01 . .1.0/3.79 . . 37.2 . . 27.7 . . 25.00 . . 25.17 . . 0.271 . . 153 . . . 44.8 . . 99
45.01 . .0.5/1.89 . . 38.2 . . 28.7 . . 25.00 . . 25.33 . . 0.293 . . 148 . . . 43.4 . . 94

75.02 . .2.0/7.57 . . 45.0 . . 29.4 . . 25.00 . . 25.14 . . 0.267 . . 234 . . . 68.6 . . 91
75.02 . .1.5/5.68 . . 45.4 . . 29.8 . . 25.00 . . 25.18 . . 0.272 . . 243 . . . 71.2 . . 95
75.02 . .1.0/3.79 . . 46.1 . . 30.4 . . 25.00 . . 25.28 . . 0.281 . . 251 . . . 73.6 . . 98
75.02 . .0.5/1.89 . . 47.8 . . 32.0 . . 25.00 . . 25.56 . . 0.304 . . 251 . . . 73.6 . . 98

105.04 .2.0/7.57 . . 53.3 . . 31.6 . . 25.00 . . 25.19 . . 0.269 . . 341 . . . 99.9 . . 95
105.04 .1.5/5.68 . . 53.9 . . 32.1 . . 25.00 . . 25.26 . . 0.275 . . 350 . . . 102.6 . . 98
105.04 .1.0/3.79 . . xxxx . . 33.0 . . 25.00 . . 25.39 . . xxxxx . . 350 . . . 102.6 . . 98
105.04 .0.5/1.89 . . 56.9 . . 35.2 . . 25.00 . . 25.77 . . 0.304 . . 346 . . . 101.4 . . 97

in looking at this, I suspect that the measure of a wb may be described as it's "efficiency"

need to massage the #s a bit more, and look at a wider range of values

LegumaN
added Watts, some of the values are the same as the many assorted rounding errors make them look that way
- but note that the input power is carefully controlled for each group

be cool

[Mictlan]

Originally posted by UserName
To clafify one small point in Mictlan's excellent post. Closed loop systems do not fight gravity, like open systems do. Head is not a factor in the traditional sense.

Just one quick clarilfication to the clarification made by USerName (always wanted to say that :P).....gravity always works against/with you in an hydralic system.....

Let's assume the following setup: you installed a reservoir in the top of you case with a submersible pump at the bottom of it. Then your radiator located in the side panel behind the mobo tray. And oviously you'll have your waterblock over your CPU (around the mid point of your case). The radiator will be an "U" shaped one, so that the inlet is in one side at the top and the outlet in the same side at the other side. The flow will be the reservoir -> pump -> waterblock -> radiator -> reservoir.

When you are pumping the water down your system, you need less power as gravity works with you, but when you need to pump it upwards gravity works again you. In other words, in a system like this you'll be pumping downward from the pump to the waterblock, then upward to the radiator, down and up thrue the radiator and finally up to the reservoir. In this setups if you do the math, you'll see that you need no extra power due to the gravity as you travel almost the same distance downwards than upwards (1-1=0), but you do need the power to start up :).....And this apply to the reservoir setup or with an inline pump as both follows the same basic layout. THe only difference is the real pressure at a given point.

I haven't seen this kind of calculations shown nowhere, so give time (and if no one beats me first as with the heatsink's material and general heat transfer articles) I might work something out.

[Les56]

I have been looking at the data here and comparing with that posted at Tekheads:
http://www.tekforums.co.uk/posts.php?threadId=6913

http://www.jr001b4751.pwp.blueyonder.co.uk/Billa1a.jpg

Most intrigued by :
1) The high temp gradient at the "Die/ TIM/ WB " interfaces. -- a 10mmx10mm rather thick(0.1mm) layer of Arctic Silver( Thermal cond 8 W/m*c ) gives a Thermal Resistance of 0.125 c per w.(according to my 0ne horse calculator Kryotherm). It appears that WB can be unimportant,
2) Is it legitimate to include your old Tekhead data ?
Or are the conditions completely different eg 25c inlet temp.If legitimate is it the WB or a different TIM

[BillA]

Les

1) I'm thinking you're right
IF the wb is sufficient, it is "transparent"
and it's the TIM that is the limiting factor
- but this premise CANNOT be true, witness my modded 462 bp die temp reductions

2) no to combining the data, each of these setups was for a particular purpose
and I'm changing equip so fast that I am often unable to "re-create" the data

I do of course still have that 462 bp,
and will re-run it as one of the bp thickness series

the TIM is always the same,
ASII a bit thick - after 24hrs under a 24lb load at 40^C (or longer)

and a clarification of my recirculating lab chiller "set point" description earlier:

I was not thinking about how the chiller actually operates
- the compressor is always on, and the heater element cycles on-off "against" the compressor to hold a set temp

so when I reduce the set point (as I reduce the flow rate) to hold a fixed wb inlet temp, I am also reducing the total heat input into the system

as it "should" be

be cool

[LegumaN]

Bill, in your values there are no coherent variation of %eff vs flow rate, it seems to me that these differences are only due to measurements errors, round off errors, secondary heat path ( natural convection and radiation of the wb and the die simulator, conduction from die simulator to workbench, conduction through temp probes) even relatively small, added together they may fill the gap.

Could you calculate relative error of the computed Watts ?

What about ambiant air temp ?

[BillA]

LegumaN, Les, and Aesik

sorry for the delay, but I've been calibrating, re-calibrating, and re-re-calibrating
- a tedious exercise but worthwhile for the cleaner data
I've e-mailed each of you a complete dataset (notebook pg scan)
(and will do so to others interested, PM me)

LegumaN
it would seem we are in agreement
re the Wattage calcs:
milliamps read on a Data Precision 1350, millivolts on a HP 3490A
both past the calibration due date but as the same range on each is used for all measurements, the relative basis is unchanging
I take a single reading of each for each data point, so range and variance is not applicable; if there is a fluctuation I will observe it, and take the mid-point
(the HP will average it for me)

an eg.
volts = 37.296 VDC
amps = 1.072 A
calculated Watts = 39.98 W (this is the value that is used in subsequent calcs)
- Note: depending on the purpose, I may indicate this as 40W, or even 40.0W

amb is 22.3^C, + or - 0.3^C

I understand that the "% efficiency" (as calculated) is NOT that at all, more like a very crude heat "balance"

BUT MY QUESTION still remains:
why are the (work done) Btu calcs constant when I KNOW that the total heat being removed from the wb decreases with a decreasing coolant flow rate ???????

(BTW: the flow rate is measured with a Sparling magnetic flow meter, to 3 decimal places; calibrated to 0.25% accuracy !)

Les
for each power level, the die minus bp temp is about constant; and divided by the W will give the TIM C/W (0.193 for this run)
- MOST useful to eliminate TIM induced variations

another item apparent is that the wb C/W is unaffected by the applied power; it varies with the flow rate (as expected)
- I am most pleased to observe this as it means that I can do all testing at one power level
(probably 70W, ACTUAL - as contrasted with "CPU output")

as always, comments solicited

be cool

[rogerdugans]

I have been following this thread with great interest, since I am preparing to undergo my first water cooling experience!

I have learned quite a bit from all you folks who are posting in here.....but wow! You guys are so far over my head its not funny:D (Well, yes it is, actually!)

I really have no input to add on the calcs, just wanted to thank you all and let you know that I am looking forward to the results.

[Aesik]

Originally posted by BillA

BUT MY QUESTION still remains:
why are the (work done) Btu calcs constant when I KNOW that the total heat being removed from the wb decreases with a decreasing coolant flow rate ???????
be cool

Perhaps I'm a little confused here, but are you thinking that because the flow rate is decreased, less energy is being removed from the block? Once at equilibrium, no matter what the situation, the exact amount of energy is being removed from the block no matter the flow rate. What happens is that with a lower flow rate, the temperature of the block must be raised in order to have a high enough temperature differential to transfer the heat to the water.

If I'm mis-reading what you're asking, please correct me. Otherwise I can go into more specifics and equations of how to describe the physical situation.

[BillA]

"are you thinking that because the flow rate is decreased, less energy is being removed from the block?"

yes
why else would the temp rise ?

backtracking a bit,
equilibrium is established for each regime, with temps whatever;
when the flow rate is changed a new equilibrium will be (eventually) established, with new temps
no problem

BUT (as I commented earlier)
"what DOES change with the flow rate is the setpoint of the bath temp to hold the inlet temp ABSOLUTLY constant (25.00^C)
- as the flow rate decreases, the temp setpoint must also be decreased; and by an ever increasing amount at the lower rates"

to which I later added:
"I was not thinking about how the chiller actually operates
- the compressor is always on, and the heater element cycles on-off "against" the compressor to hold a set temp

so when I reduce the set point (as I reduce the flow rate) to hold a fixed wb inlet temp, I am also reducing the total heat input into the system"

and so from this tortured train of thought I conclude that
less heat is being removed at lower flow rates

I believe that part of my difficulty is that I have no means at all to measure the (positive/negative) heat contributions of the chiller and it's 2 pumps and heater
nor do I see such as being necessary given that for EACH trial I wish the wb inlet temp to be the same

I'm just having difficulty getting my "head" around the calculation method
(or indeed as you suggest, my understanding of it)

do feel free to explain

be cool

[Aesik]

Bill, just so that we are on the same page and so that it doesn't take a thousand posts going back and forth, check you PM and give me a call or send me your phone # and I'll ring you up.

[UserName]

Hey, posterity and all that.

Goin off line don't help the rest of us.

But i'm sure you will come back when you all get it sorted out, right?

[LegumaN]

Same as Aesik, a bit confused or mis-understanding your question ?

The only thing I see that could match your question is that when flow rate is lower, overall wb temp is higher, then for a constant ambiant temp, heat transfer through natural convection is higher then water get less heat.

About wattage relative error calculation, sorry I was not clear enough, I was talking about the one calculated with the relation
Q = Ww C delta T
I would like to know delta Q / Q ?

There is something that disturbs me in your set up, the bp temp measurements : the TC you put in the bp will introduce a pertubation of the heat path through the bp ( constriction ). Moreover it is where there are big gradients.
The influence of the Tc is different with different wb type, I think about swiftech wb ( impinging jet ) where heat transfer is concentrated above the cpu. For maze wb type, the effect should be less important.

The TC wouldn't have a constant influence if you compare different wb type, and for each type different bp dimensions.

I'm not an experimental guy, but the fact of introducing a perturabtion in what you want to test is not natural for me ( I think about Heisenberg principle ).

Is this remark justified ? waiting for reactions...


About your power level selection, I think you should take the higher one in order to get the fewer measurements's relative error.

[Les56]

Ta for data Bill. Should keep me amused for a while.

My understanding of Bills "brain searching" is that he wishes to explain the "higher cooling" settings on the cooling unit to maintain a 25c inlet temp at lower flow rates,This being the case even when the heat per unit time absobed by the water block (Temp diff ,flowrate sums) is, near as damn, the same as at higher flow rates.. At least I cannot understand this unless as Bill sugests it is inherent in the cooler unit's "modus operandi"
Les

[UserName]

great point LegumaN. disturbing the heat flow in a spot that is critical is a bad thing.

That begs the question how would you measure the block temp?
any sort of measurment would cause issues wouldn't it

[BillA]

UserName
posterity ?? more like posterior
e-mail me if you want the dataset

Aesik and I had a nice talk, and I thought that my question was resolved
but upon reflection the fog settled back in

Les's re-statement is correct

I seem to have difficulty understanding why, with a self controlled/regulating chiller, I should have to adjust down the set point as I decrease the flow (to maintain a constant inlet temp).

the chiller uses an RTD and variable pulse width/rate to control the temp with variable loads
- the chiller simply corrects, up or down, for any change in the bath temp

so I return to my difficulty:
with no change in the heat "load" being applied to/through the die; after decreasing the flow rate, WHY do I also then have to reduse the set point to maintain the SAME coolant (inlet) temp ???

LegumaN
"when flow rate is lower, overall wb temp is higher"
this is true, but not enough to make a difference as the inlet temp is constant (and therefore the body of the wb, a MCW462) and only the wb bp center area is doing much changing

"delta Q / Q ?"
if I understand you, in the data I sent you:
subtract (3) from (14), and then divide by (3)
??

WELL, WELL, WELL
hard to believe, but I think I just stumbled upon the answer to my riddle

my oil-cooled aluminum-cased 500gph pump is in the chiller bath,
to reduce the flow rate I'm closing a gate valve at the end of the line prior to dumping into the bath;
- I'm increasing the backpressure on the pump, causing it to have to work harder, which generates more heat
--- which then in turn decreases the heat needed from the heater
(so I have to lower the set point)

god damn
something so simple !!!

well, now I also know how to measure the heat load added by a pump
(oh no, not another effing article)

many thanks for the helpful feedback fellows, I owe you all a beer !

be cool

[BillA]

re the TC hole in the wb bp:

in an absolute sense, I'm in agreement that it would have some effect
and yes, the effect would be slightly different for different wb bp designs

but how much ?
it is a VERY simple test, and for sure I will run it

but in thinking about it, I suspect that the difference may well be to small (for me) to measure

we're looking a thermal gradients which are less than 50^C
and the thermal conductivity of the material is high

it would have more effect if it ran along an edge ("shadowing" a side),
but going to the center rather less

in any case, I'll run a trial

LegumaN, your critical organ is healthy

there is a "problem" with higher power levels:
most will assume, quite erroneously, that my heat die's 100W is the same as their CPU's 100W
(the obvious difference being that the heat die has very limited secondary losses, whereas such are rather large with a CPU mounted on a mobo, also using software to try to drive the CPU)

I'll run at different power levels 'till it seems more clear

be cool

[LegumaN]

Bill,


I'm surprised, don't you ever compute errors of what you measure ?


Relative errors are computed as follow :


Q = Ww * C * ( Tout - Tin )


then take Neper log of it


ln Q = ln Ww + ln C + ln ( Tout- Tin )


then differentiate ( assuming C is constant )


dQ / Q = dWw / Ww + d( Tout - Tin ) / ( Tout- Tin )


= dWw / Ww + dTout / ( Tout - Tin ) - dTin / ( Tout - Tin )


then to the first order


deltaQ / Q = deltaWw / Ww + deltaTout / ( Tout - Tin ) - deltaTin / ( Tout - Tin )


then take maximum absolute value limit


|deltaQ / Q| =< |deltaWw / Ww| + |deltaTout / ( Tout - Tin )| + |deltaTin / ( Tout - Tin )|





I need to know the errors of your measuring devices : deltaWw, deltaTdie and deltaTin.








Yep natural convection is negligible, that's why I was confused, I did not get the question was about the chiller part.





I will wait for your test of TC hole influence. Something linked to this issue : I found a thesis about TC influence on measurements ( constriction ), at first glance it seems that it could have a huge impact for some situations.

[BillA]

LegumaN
"I'm surprised"; do not be, the liability of self-education is that so much can be missed

a question: is the "error" being calculated related to a single measurement, a group of measurements, or the measuring device ?

the definitions are (that I am aware of, eh):

Absolute Error is the absolute value of the difference between the true value (or expected value) and your measured or calculated value.

Absolute Error = |True Value – Found Value|

Relative Error is the absolute error expressed as a fraction of the true value.

Relative Error = |True Value – Found Value| / True Value

Percent Error is the relative error expressed as a percentage of the true value.

Percent Error = Relative Error x 100% = x 100%

but as I never know the "true" value, what then ?
what I suggested was the Relative Error, but based on the deviation from the input

when you speak of delta"x", that to me suggests an Error Propagation calculation to produce an Uncertainity Estimate

I must admit that this is moving quite out of my area of knowledge
I will study some on it
but even now am inclined to ask you:

if I provide you the data, can/will you run the calcs ?

re the TC hole:
yes, data is needed, am even now doing testing

be cool

[LegumaN]

When I speak of delta X, this include the given ( by manufacturer ) error of the measurement device plus the reading error ( variation around recorded value and precision of "eye reading" ).

FYI, in the general case, when you want to obtain relative error of some quantity that could not take the form of a product, you can't use logarythm :
an exemple :
V = A * B / C + D
just differentiate
dV = dA * B / C + A * dB / C - A * B * dC / C² + dD
then to the first order
deltaV = deltaA * B / C + A * deltaB / C - A * B * deltaC / C² + deltaD
then divide by measured value V ... and take the maximum absolute value
|deltaV / V| =< ( |deltaA * B / C| + |A * deltaB / C| + |A * B * deltaC / C²| + |deltaD| ) / |V|

I hope someone will correct me if I'm wrong.


I can do the calcs, just send me needed datas.

[BillA]

UserName
I must now agree with your earlier assessment, indeed this thread has been hijacked

now on to statistical analysis

LegumaN
many thanks for the generous offer, I will provide data, data, and more data
I'll do this via e-mail as all this stuff is trash to most readers
(20 pg threads between 2 posters don't make much sense)

be cool