Water block Temp Diff question

[UserName]

How much does your water temp change going thru your water block?

I mean if you measure the temp of your water RIGHT before it enters you block and again RIGHT after what is the difference?

Please refrain from posting if you think you know, or you know this guy who..., or you read about..., or some other i didn't measure it but... comment

Just facts from you to us as you measured it.

You can post links to articles that have someone else's data.

[Tiger]

My delta was 0.2C. It was measured in order for me to calculate the wattage of the compressor for a water chiller.

[The Overclocker]

a rough guess is around 4 degrees, my pc is 4 degress above ambient

[UserName]

the overclocker

Why is it a rough guess? You measured it didn't you? You wouldn't have posted if you didn't actually measure it, right?

[lennytiger]

Username who r u??

Why UserName?????

[Tiger]

Originally posted by the overclocker
a rough guess is around 4 degrees, my pc is 4 degress above ambient
That's not possible. If your water gained 4C for each pass you would never cool it. You are moving water through the block at approx 10 to 15L per minute. To transfer that kind of heat you would have be using an incredible conductor.

[BillA]

my lab notebook has 40+ pages of wb inlet and outlet temps, and more are added daily

you need to specify a heat load (NOT total, eh -just that to the die)
and a flow rate

all my data uses an inlet temp of 25.00^C (+ or - 0.01^C) from a Haake A82 lab chiller
temps measured with a matched pair of YSI 700 series thermisters
and a pair of calibrated Digitec 5810 thermometers having 0.01^C resolution

with flow rates from 3 to .3 gpm, and heat loads from 50 to 130W
the temp rise will range from 0.05^C to 1.75^C
depending on where in that range you are operating
and, perhaps, your wb

I'm starting on an article for JoeC on wbs, so eventually you will see a bunch of these graphs
you'll not find many with measurements, and far fewer with good measurements
a bit of such is also in my article on rads (http://www.overclockers.com/articles481/)

be cool

[ButcherUK]

Originally posted by Tiger

That's not possible. If your water gained 4C for each pass you would never cool it. You are moving water through the block at approx 10 to 15L per minute. To transfer that kind of heat you would have be using an incredible conductor.

Um where did you get those flow figures. Most people don;t run huge pumps you know. My 600l/h pump (the 10l/min figure you have) flows about 200l/h when connected, that's more like 3l/min. :P

[-EternitY-]

That's awesome dude, Lol... Im at 100F Right now........ When you are having NEGATIVE Temps. =(

[Tiger]

Originally posted by ButcherUK


Um where did you get those flow figures. Most people don;t run huge pumps you know. My 600l/h pump (the 10l/min figure you have) flows about 200l/h when connected, that's more like 3l/min. :P
Well that depends on your head doesn't it? Most people don't run 600l/hour more like 1200L/hr.
But lets look at the chart for the Maxijet 750;
@ 0m head its rated to pump 620L/hr
@ 0.5m (50cm) its rated to pump 600L/hr
Most people don't get near a head of 50cm.
So why is yours moving only a third of its rated amount?
Even if it were pumping 3L/min, what's your point? There is no way a waterblock is going to warm water 4C per pass.

[BillA]

Tiger:
Butcher's point was valid
perhaps you are lacking the data/experience to understand also what he did not explicitly state
so here is a bit of clarification

before you spout pump mfgrs "ratings", test your system
rig a water supply setup that will not vary the head that the pump sees
and elevate the whole system so that from an open-ended discharge tube you can fill a 5 gallon (20l ?) jug
now time your system and calculate your flow rate
- and, for drill, also back-calculate the system's "apparent" head loss (flow resistance)

it would be generous of you to share those numbers with us

to analise a bit your numerical ruminations:
for us on the old system; 600l/hr = 2.64gpm (rather high, eh ?), and 50cwH2O = 0.71psi (a bit low, no ?)
please spend a moment with this article (previously cited) (http://www.overclockers.com/articles481/)

select the radiator of your choice, and do tell us how you propose to put 2.6gpm through it, plus the waterblock, plus the tubing, plus any fittings
-- and still have only 0.71psi of head loss

you can't get there from here
Butcher's point was simply that the real-world "translation" of pump mfgr's ratings is to assume 40 to 60% of nominal
- and I don't see where Butcher commented at all on the 4^ fantasy

I do not like this approach as I wish to DESIGN a system, but my views are not agreed with by most WCers

be cool

[The Overclocker]

UserName you are right, i haven't checked and didn't read the question.

last time i checked i used basic equipment so the difference was 1 degree (does not go lower) os it must be 0.5 degrees or greater (all in celcus)

[lennytiger]

24c load is a good temp!!!!!!

[Tiger]

Comment for BillA;
Yes everything you say is quite correct. In my world I get my own figures based on my particular parameters and I work out for myself what I need to do.
In my system the only bottle neck is the w/b otherwise it is totally open except I am pumping high viscosity fluid. As you said earlier each case has its own individual charactaristics and no two systems will ever be the same for very obvious reasons.
The original question posed was; what is your delta - please don't supply annecdotal evidence or guesses.

[lennytiger]

his delta?? is it a fan???


lol

[ButcherUK]

My figures are real (namely I mesaured my own setup off the outlet pipe), not something I cooked up with a manufacturers data sheet and a pail of moonshine. The waterblock is a significant reducer in most setups. Mine knocks 23% off my flow (i.e. I get 23% more without it) and it's not particularly restrictive. It's not uncommon for people with much more powerful pumps to have flow rates below 300l/h. I don't the head chart for my pump (eheim 1048) here so I can't tell you what my flow should be according to your system but i'll bet it's a lot more than 200l/h.

As for your "totally open" system, I take it you're running huge bore tubes and no res / rad then because even a length of 1/2" tube is a restriction on a system.

[Tiger]

I have achieved what I want to achieve through doing the necessary research. I think my oc and temps are pretty good. Each project I do is carried out using good scientific principles based on the equipment I am using. So yes, I have measured flow rates, temp deltas and I have manipulated flow rates to achieve the desired effect.
This topic could be expanded into a very interesting area of discussion. There is a great deal of scope for developing a system whereby the specific thermal charactaristics of water blocks could be determined and evaluated mainly because water is the standard by which specific heat is measured.
It would tremendously helpful if you dessisted from comments like "pail of moonshine" and massaging your own ego and concentrated on topics at hand.

[lennytiger]

I just wouldn't want to risk, putting water and electricity so close togeather, even though 12v is the highest current in the comp what if it leaks???? all over my mobo and cpu?????? I just wouldn't risk it!!

[BillA]

Tiger

I'll take a look at a series down to 0^C and see if there is an apparent change in wb performance
(Btus transferred)
I kind of doubt it

be cool

[Tiger]

BillA - Have some ideas about using a w/b as calorimeter to determine the efficiency of them. I have not quite got the plan together in my mind but there must be a way of doing this by knowing the energy output of the proc which is being used as a heat source. Got any ideas on this approach?

[BillA]

Tiger

er, that's what I'm doing
if you want to see how, look at this thread for an outline

http://www.tekforums.co.uk/posts.php?threadId=5140

your CPU is not an appropriate heat source, look here for that

http://hardforum.com/showthread.php?threadid=266016

then you're going to have the problem of taking a number of measurements to VERY HIGH accuracy

etc, etc

be cool

[Tiger]

Originally posted by BillA
Tiger

er, that's what I'm doing
if you want to see how, look at this thread for an outline

http://www.tekforums.co.uk/posts.php?threadId=5140

your CPU is not an appropriate heat source, look here for that

http://hardforum.com/showthread.php?threadid=266016

then you're going to have the problem of taking a number of measurements to VERY HIGH accuracy

etc, etc

be cool
The 1st thread came up fine but 2nd is not accessable.
Comments on the first thread - having looked at the angles and thought about it I don't think a live proc is clever heat source. I'll tell you why from my own personal experience. I have about 6 different apps that are supposed to stress the cpu to 100% but there are 2 apps that cause more stress than any of the others and will raise the cpu temps 2C over the others. 1st is the first 2% of seti and the second in my virus scan when it scans compressed files. In terms of deviation this could be as high as 10%. Having said that is there an alternative? A few months ago I saw someone that manufactured an artificial one that was made from a copper block and had a core machined into the top face and had an electrical heater unit inserted into it.
The positives of this approach would be 1st'ly a standard calibrated heat load and secondly it is not a core so it is possible to manufacture it with the studs necessary for block mounting. and therefore get standard block pressure. As to the problems with TIM. If the standard is coolant at 25C and AS2 then that is the standard. The precision of a constant could be taken to nth degree but it is an accepted fact that no to cpu core faces will be identical and no two w/b faces will be the same and it would therefore be logical for me to say that there is x% of error or deviation inherent in such an approach.

[BillA]

links ok
(their server is soft)

be cool

[Tiger]

That's the article I was referring to regarding a heat source simulator.

[UserName]

Excellent;
Gentlemen, I laud you. All this good thinking and all I can say is yes, yes, and yes.
I will try to add a little. I don’t think you can quantify and control every aspect if this complicated system. At least most of us can’t. Our Heat sources wattage will change as the voltage changes, and it does every day or every hour. And one time you will get a better seating of the block than others, and one batch of heat goop might be better than another, and etc. etc. The whole system will be very hard to quantify, and impossible to verify.
This brings me to verification. If you do lots of work and I can’t reproduce it, is it of any value? Until others can reproduce our findings, I don’t think they have value.
So I asked a question that I hoped would generate a consensus that water temps should be about this much on exit. I think we could say mostly between 0 and 2°C. Perhaps we could standardize the Degree units too. After that we will take over the world.
I also suggest that, all other things being equal, a lower delta is better. What do you say?

[Tiger]

If it is accepted that a heat simulator is the best way to go then the next step is calibration of the heat sink. To my way of thinking the mass of the heat sink should be determined and internal volume determined by filling with water @ 4C and measuring the water volume. The water block should then be heated and the time to heat the contents 10C be determined a number of times with static water content. The theoretical specfic heat of water can be be determined and compared to actual. The specific heat of copper is known and should be accepted as a constant or you could run the same sequence on the block empty to determine the SH of the block as well.
The block could then be tested with water flows of increasing magnitude to graph heat transfer chractaristics at these various flows to determine the optimum flow rate.
Is my thinking correct or am I way off the mark?

[UserName]

I think we have jumped ahead.
How will a good heat sink perform? More importantly why do we say that?

If I have 2 water blocks, one very good and one very bad, how can I tell the difference?

I guess the good one will keep the CPU colder, all other things being equal.

What are the other things that could change?
1) Incoming temp.
2) The ambient temp is constant or it is insulated.
3) The pressure is constant
4) The flow is constant.
3 and 4 are very difficult because every design is different.

I am not done, but is this making any sense?

[Tiger]

The object is determine how effective the wb is at transferring the heat to the water. There have to be designs that optimise factors such as surface area, minimise laminar flow charactaristics and optimise the flow rate.

[UserName]

Oh, so that’s why you heated the block.

Then we could just control the input water temp and flow rate and observe the water pressure and block temp changes. That would give us a graph so we could see how good the block is for any particular application.

Agreed?

[Tiger]

There are standard factors. The base for specific heat is water. The standard is that it takes 1Joule to heat 1g of water 1C at 20C and 1Joules=4.1868 calories. The idea would be to see how close the heat sink comes to this standard. In power terms 1000W(/hour) or 1KW=3 600 00J. I hope you see where I am going with this.

[UserName]

But if the water is not flowing then given enough time the water will reach the temp. I guess the idea is how fast?

Are you speculating that if the water heats faster standing still then it will work better when it's moving?

Surely one may work better with slow water and another with fast?

[Tiger]

That's the question. Go to ProCooling.com where they have just reviewed a batch of Intel based water blocks and have found that flow rates did not appear significant and there were liitle differences in the performances. But the question is was the method of test flawed for the reasons above?

[BillA]

I’m going to run at this only wrt wbs, not hsfs

let's first attempt to eliminate (or control) as many variables as possible

1) looking at the heat source:
- use a Sola transformer (to "eliminate" line variations)
- use regulated DC power supplies to power a resistance heater
- simultaneously monitor voltage and amperage (millivolts and milliamps, above 2A is difficult)
- adjust voltage to apply a predetermined and constant power
- eliminate secondary heat path losses (through the heat die's design, construction, and ultimately insulation)
- monitor the heat die temperature using an embedded fine wire TC
--> calibrate the TC after installation (NOT so easy to do, still thinking on this, TCs used to measure hundreds of degrees are NOT designed to measure tenths, the tolerance on even special wire, even cut from adjacent pieces from the same roll, can exceed a degree - how to quantify, and over what temp range ?)

2) looking at the die face:
- standardized dimensions, and tolerances, and flatness, and finish
(do you all know what is needed to do this ?)

3) looking at the goop and it's application:
see the thread previously cited
-->I know of no way to control the applied thickness

4) looking at the coupling of the wb and the heat die:
unless one wishes to test wb mounting hardware, to establish wb performance wbs must be connected with the heat die in a controlled manner
AMD specs a 24 pound maximum applied load, and springs seem to be the easiest way to achieve this
- measure the spring height at the desired load, and replicate that height when installed

5) --> calibrate the heat die itself by heating water in a "standard" container ?
- this is a basic part of "round robin" testing and is needed only if the comparability of data from different sources is desired
- note that the water must be kept in motion (in order to read it's "overall" temperature), so how much agitation, how done, etc
- this standardized container just went to thousands of $$ to design and build

moving right along here . . .

6) looking at radiation and convection losses from the wb exterior:
I ignore them as the wb body is at the coolant temperature which is only slightly above ambient, and such losses will be small compared to the effect of the coolant
(additionally such losses occur as well in the actual use of wbs and will be the same order of magnitude)

7) looking at the coolant flow rate:
- the use of a precision sensor is essential, and it must be calibrated (to 2 significant digits presently)

8) looking at the differential pressure (loss) across the wb:
- use a digital pressure gauge (reading to hundredths of a psi)

9) looking at the coolant temperature control:
as the temperature increment to be observed is small, this is an ABSOLUTLY key parameter to be controlled, and will define the repeatability of all work
- the coolant temperature variation MUST be held to less than the smallest increment to be observed, with my present setup that is 0.01^C [a Haake A82, controlled with a precision DC microvolt reference standard - normally used for instrument calibration]


10) looking at the coolant temperature measurement:
the coolant's temperature measurement is another aspect of it's temperature control, hence the same requirement
- temperature resolution to 0.01^C is required (or the results will be trash)
I do know this for a proven fact, I have spent over a year in a continuous series of sequential equipment upgrading to achieve the necessary sensitivity to be able to distinguish differences on a repeatable basis
--> conventional thermocouples of any type will not serve; only RTDs or specially designed (dual "linear") thermisters [I use YSI 700 Series with Digitec 5810 thermometers]

in summation I would observe that the easiest "cross calibration" is probably done by testing the same wb
--> but if the test equipment capabilities are not the same, neither will be the results

look at the thermodynamic "testing" of watercooling components presently on the 'net
it's not difficult to quantify the lack of technical education and experience, inappropriate equipment selection, and inadequate testing procedures

be cool

Edited for UserName

[Tiger]

Everything looks really good.
Two areas of concern that you have raised here;
1. Tolerances of the die face - are there any specs?
2. calibration of the die. You mention "and is needed only if the comparability of data from different sources is desired" can you clarify this?

[BillA]

"Everything looks really good", well after a year's "schooling", it should
(I spent 3 months just learning what TCs could NOT do !)

"Tolerances of the die face - are there any specs?"
sure, the AMD (or Intel) specs for the top of their CPUs
- I have not seen such for flatness and finish, anyone got a link ?
(and I know I will not be buying the equipment to measure such)

"the comparability of data from different sources"
so long as I compare my results with my results, no correlation to any absolute "standard" is necessary
(assuming good procedures, equipment, calibration, etc)

BUT

if I wish to compare my results with JoeC's results, then a common reference point must be established
(in addition to the comparability of equipment, etc, etc)

this is the present problem with C/Ws
look at this thread
http://www.tekforums.co.uk/posts.php?threadId=3670

be cool

[BillA]

double post
??

[UserName]

Bill;
Could you go back to your post, "I’m going to run at this only wrt wbs, not hsfs ", and add 1) 2) 3) and A) B) etc. as you feel appropriate. It is an excellent approach, but rather difficult to discuss bit by bit.

I think your concern hits the nail. If you set everything up and get awesome performance out of the block, but no one can reproduce your findings, it is of no value as a guide to the general public and to designers of blocks.
.

[BillA]

UserName

"if ... no one can reproduce your findings, it is of no value as a guide to the general public and to designers"

well, not really
we are already at that point with radiators, but the fact that no one else has an adequate test bench setup does not deny the utility of the information I presented

the pending article on wb is oriented towards a testing method (and equipment) description, and will use the "best" 4 wbs as as (incidental) examples

all this grew out of my disgust with the super inept "reveiws" that purport to describe thermal performance

and I quite doubt that others will go to the trouble of setting up an effective measurement system
- it's too much easier to just swap units, and then pronounce "A is better than B"
now THAT is the valueless information

be cool

[Tiger]

Yes, well I have been quietly watching this debate re C/W with some interest because it is only relevant when ambient is of significance. If you happen to be me or any person using any form of phase change then it becomes irrelevant.
I wrote an article sometime back looking at the cooling charactaristics of air cooled HS's. Basically using the idle app WPCRSET to switch off the processor and examine th cooling curve which was very interesting. I later swiitched to the heating phase which was far more interesting in that different HSF demonstrated very different charactaristics. I also wrote an article about this but Joe decided not to publish. The gist of the article was that for very light and very heat sinks there was standard rise to about 33C which I think is the point that heat saturation occurs at ambients of around 20C but thereafter there was radical and predictable difference. The lighter HSF climbed to their operating temp very quickly while the heavier ones took twice as long. The point is that I beleive that any good heatsink should have one attribute and that is that they should heat slowly and have a low peak. So if the graph is examined the light ones have a spike while the heavy ones have a hump. I suggested that this fact gave a good indication of HSF performance. While the terminal performance of the HSF's is about par the stability of the proc is affected by rapid changes in temperature. This is why water is so good - it has great thermal capacity.
Coming back to the topic - Core specs; I think given the extreme tolerances that the wafers are manufactured under I doubt whether there is anyone out there who have the kind of measuring equipment capable of measuring irregularities. The only kind of device that i know of would be some kind of electron microscope.
standards- well someone has to be a developer of standards. Don't you do standards testing Bill? I seem to remember a reference to this somewhere. Its a case of well here are the parameters I have set and like you say - prove me wrong.

[BillA]

Tiger

the heat die face is a potential problem, particularly as with use and abuse and the time comes to "re-face" it

my die is 10x10mm
how the hell can that be lapped without rounding the edges ?
- not by hand (at least not by my hand)
(the whole side of the heater block was lapped before the die face was cut)

and the CPU die specs; for sure to be ignored
wafer polishing is far beyond what we can inspect for

yea, I used to be a member of ASTM, principally involved with stress regression testing of fiberglass pipe and adhesive joints

be cool

[Tiger]

Originally posted by BillA
Tiger

the heat die face is a potential problem, particularly as with use and abuse and the time comes to "re-face" it

my die is 10x10mm
how the hell can that be lapped without rounding the edges ?
- not by hand (at least not by my hand)
(the whole side of the heater block was lapped before the die face was cut)


Is there any particular reason why copper has to be used? Surely stainless steel would be an ideal material for use. Provided the unit is producing the specified output does it really matter what it is made of?

[ButcherUK]

Originally posted by Tiger

Is there any particular reason why copper has to be used? Surely stainless steel would be an ideal material for use. Provided the unit is producing the specified output does it really matter what it is made of?

I may be off the mark here, but if Bill's using a larer than die sized heater, a material like copper with high conductivity would be better than say stainless steel as otherwise the conduction of heat from the heater to the die surface may be affected. Maybe cusil is an option (assuming Bill has a large wad of cash), it;s not much worse than Cu thermally but is apparently quite a bit harder so would wear less.

[BillA]

interesting tangent, had not considered other matls

but:
yes, the heater is larger than the die (were it the same I'd just use it)

while the die being simulated is silicone, it's thermal properties are not so good - apart from the fabrication impossibility

with any heat transfer there will be a thermal gradient, and with copper (ignoring silver - unless Butcher's paying) such is minimized

there is also the problem of the temp that the embedded TC is reading, again the lowest gradient is desired

decreasing the die's thermal conductivity will also increase the magnitude of all secondany path "leaks" - and also lower the response rate - and therefor have a higher temp than a matl with higher thermal conductivity

I think that's straight, other things I've missed ?

be cool

BTW: those Cusil plates may become rather cheap, to the price of silver anyway; I have a couple of them but can't think of an easy way to test (no TEC setup)
any ideas ?

[Aesik]

Some very interesting discussion going on here. When it comes down to it, for testing and comparision purposes, all that really matters is getting a consistent setup that will load and measure the test specimens. It doesn't have to perfectly simulate a cpu, in fact I woudln't want to do that because of the way that cpus vary their loads over time.

Initially I'd think copper would be the best solution because it's 'reaction time' would be better and the temperature drop because of contact resistance at the block/thermocouple junction would be reduced. However, there is the problem of damage to the copper and how to repair it if that were to occur. I know I certainly couldn't lap a 10mm square slug. If any damage where too sever, there is also the issue of removing so much material that the thermal path is shortened enough to make a difference in the measurements being taken.

Interesting. I'll have to mull this over some more before drawing too many conclusions. What we need is a material with the conductivity of copper, and the durability of steel ;) If enough care is taken and the copper block is treated like your first born, it will probably stay in good enough shape for a good long time anyhow. But inevitably, something will eventually happen to it that could compromise the test results.

[Tiger]

If u used the the neoprene pads as per cpu then potential damage should be minimsied and you should get pretty good mileage out of the simulator. Thinking about it why cann't u use a neoprene shim?

[BillA]

I do use a rigid urethane foam shim for heat insulation, but it provides no support
- not really TOO concerned, just recognize the die's "pivotal" position in the system

and the fact that the residual ASII that's on it will be so "forever",
not a problem as it's pretty much the std goop

be cool

[lennytiger]

ASII is useful although remember there are fake tubes about which will do allmost nothing for you cpu temp!! :)


also what "fluid" to u guys use just water, or something else??

[UserName]

wrt the CPU die. Has any one thought to ask AMD if they have this thing specked out? They might have heat dies that look and act just like standard CPU's. A shim might decrease the performance, as the area exposed to heat is reduced. If a shim is the standard thats fine, but a copper block will be at a dissadvantage as it could absorbe the secondary heat better and, perhaps, reduce the die temp.

We would want to as close as possiable duplicate the CPU as it itself is a whole system. if we screw with a part it will have secondary effects. Are goal is to see the real world performance

Also note that we want to controll contributing variables. If a die is scratched and chiped (but still works), do you think you will see a measurable difference compared to a perfect one? If it effects performance .1C i'll eat your shorts.:eek:

Should we lap the blocks? I am thinking no, but not commited. My thinking is if a manufacture lappes his block should he not show as better than others who don't? I don't think we should do any work on the blocks. Let the cards fall where they may. At least tests should be done both way's.

In fact we should not even remove the goop pad. And we should use the supplied goop on a clean die. If one man buys better goop then he gets better performance, if not then no.

Just a couple ideas, but i got to go. I'll get back.

[BillA]

lennytiger - distilled water + Water Wetter

UserName - not really

"We would want to as close as possiable duplicate the CPU as it itself is a whole system. if we screw with a part it will have secondary effects. Are goal is to see the real world performance"

what I am doing is bench testing a single component under defined condidions to ascertain it's performance capability
I am not testing a system, in fact I'm not even going to be testing the mfgrs hold down hardware
(I calibrate each spring, then install each to the exact height for the required force)

if I wanted to use a CPU I would, but I certainally do not
and yes, AMD and Intel use a thin film resistor, last I heard they were about $1400 - JUST for the resistor, not including all the other pieces

the purpose of the foam shim is to block a secondary heat path, just like the insulation around the body of the die
the overall goal is the control of every variable, in a reasonably realistic manner

wbs will be tested as provided, as they are sold to the public (except for the TC hole in the bp)
then
as I see fit (based on my experience), they will be modded and lapped - and retested

remember the purpose of the article - to describe a waterblock test methodology

reviewers do shootouts or roundups, I do testing

be cool

[Thelemac]

If this is just for comparing waterblocks to each other (which is what I've gathered, I could be wrong) then you could probably eliminate the thermal goop as if none of them have it, it doesn't make a difference. I only mention this because you mentioned that you can't think of a way to standardize the application of it. :)