myv65,

By efficiency i ment the (lower) cpu temperature under given heat produced by cpu and given ambient temperature. But my question was rather about idealized equations (proportionality of heat transfer to delta-T's) than about real life situations.

It wasn't sarcastic. Glad you didn't take it that way. I guess you'll have to pardon my mannerisms.

Very educational. :D

Yup, the can's open! Lowest CPU temperature for a given ambient will vary from system to system and chip to chip. At one extreme, you could have flow so low that convection in the block was awful. In order to carry "X" watts, the water would need to go through a large delta-T and due to the low flow/convection coefficient, the chip would need to be a lot hotter than even the warmed fluid. At the other extreme, you could conceivably pump far more energy into the water just to pump the volume than what energy the chip would add. Delta-T in the fluid would be extremely low, but at some point you may have rising temperatures simply because of the energy needed to pump the water around.

Call it cheating on my part, but there is no single answer that covers every system. There is, however, a general rule that most systems will provide a lower cpu temperature as flow rates increases.

More practically speaking, performance from one system to another will depend far more on how you set things up than anything else. Put your radiator in a spot where it has trouble getting air, bad news. Get a block that refuses to clamp down properly, also bad news. Do everything right and you'll be hard pressed to find more than a few degrees C between decent systesm.

My only real beef with designers of this stuff is that they don't do things as efficiently as possible. Given water's specific heat, you don't need much flow to get the job done. What you *do* need is high convection, aka high velocity in the block. With the right porting, you could cool the CPU with <20 gph easily. Why this matters is that it would allow smaller pumps and smaller radiators. I'll admit that the delta-T in the fluid would be higher than a system running ~75 gph, but so what? The cost, both purchased and operating, would be lower. I guess that's why some things ya gotta do yourself. It's also why I won't be running a pre-built system any time soon.

Thanx myv65, you replied all my questions.
To resume :
1 - static or ambient pressure plays no role in heat transfer. That means the order of pump, rad, block would be determined by local deltaT rather than local pressures.
2 - Higher flow produces better cooling.
3 - consider that we take care properly of other problems (air flow through rad, elements position, etc..)

Besides i'm considering building a custom radiator. Every small bit of valuable engineering info is welcome.

Moreover i've yet to see an efficient "low-flow" system. Yes money can be a consideration but i dont see any other interest. Given a fixed system, increasing the flow will always increase heat transfer (up to the point where the pumps dumps too much heat - in our situation it's not the case).


Sirpent: "myv65, is it true that as long as we think in terms of delta-Ts for the radiator and waterblock (ignoring frictions in the system), changing the flow rate does not make any difference in the cooling efficiency?"

i'm not sure of what you mean, but if you mean "heat tranfer" the Fourier formula Q=UAdeltaT answers it. Q=heat tranfer, A = contact area, U = heat coeffficient, which depends on such things as turbulence and flow. U will get higher if flow gets higher...

As I see it higher flow is better but for the people selling this stuff to the masses they don't want the word to get out. And I say why not! The simple fact that people are willing to buy watercooling when it is really not needed for a computer to function proves that they are willing to pay for the absolute best.
Opinions?

Thanks, myv65!
I think I agree with most of what you wrote. But, still, my question was slightly different. Probably I just didn't use the right words. I'll try to be more precise. I'm building a simple mathematical model of the situation (trying to keep worms in their can).

So assume that we have an ideal situation, where the cpu generates a given wattage, all the heat goes to water, the air temperature in the case always stays the same, and we can vary the water flow in the system without generating extra heat. Suppose that the heat flux F1 from cpu to water equals C1(Tcpu - Twater-wb), where Twater-wb is the "averaged" temperature of water in the waterblock, and the heat flux F2 from water to air equals C2(Tair - Twater-rad), where Twater-rad is the "averaged" temperature of water in the radiator and Tair is the "averaged" temperature of air (the arithmetic mean of the in-case and exaust air temperatures). For simplicity, we can even assume that Tair is the (supposedly constant) in-case temerature. In a stabilized system, we have F1 = F2 = (heat generated by cpu); and we can also think that Twater-wb = Twater-rad. Two biggest difference between this simple model and the real life situation are, imho, the following: (1) in reality, C1 and C2 are not constants, they depend on the flow (or, really, local velocity, as you wrote, but the flow is our only input parameter) - higher flows mean higher numbers; (2) higher flows generate more heat even if the pump is 100% effective (and it's not). The two correction terms work one against another and it seems that the first one usually wins in the typical range of applications (higher flow -> lower cpu temperature). But my question is: is it true that in the simplified model (I think it's really the simplest possible one) the cpu temperature does not depend on the flow?

I don't know how big or how small are changes of U in a typical system (any data?). So I am asking what are the answers for the "first approximation" model where U=constant.

very lucid comments myv65

having several hundred pages of obsessively detailed measurements, I would add several observations

while the design basis of wbs could be different, the necessary pumps to obtain the higher convection rates are hugely more expensive
I'm looking at an Iwaki mag drive gear pump that has nominal 'ratings' of 36 gph @ 45 psi; list over $650;
and another 'big boy' that delivers 92 gph @ 75 psi lists for $1050
(max flow, max pressure; not both together, eh ?)
continuous duty positive displacement pumps are not cheap, and never will be

the smaller rad is quite 'doable' in terms of dumping the heat (with LOTS of air/noise),
the problem lies with the resultant system equilibrium temp of the coolant
- with the present wbs, a higher coolant (inlet) temp at the wb will always result in a higher CPU temp

looking forward to your article

be cool

Dunno whether this approaches the description but:

I have made a simple 50 x40 x 0.8mm chambered Copper waterblock ,with 6mm thick base which should give a velocity of ~ 1.7 m/s at a flow rate of 50gph.(hope my sums are correct)[Edited sums on Oct 30th(changed from the wrong ft/s to correct m/s).

http://www.jr001b4751.pwp.blueyonder.../P0001076a.jpg

Unfortunately this easy to construct "flat" geometry has a high flow resistance.
Flow rate ( radiator in separate cooling circuit) 25-30 gph with 600 lph (unknown max head) pump,and 45-50 gph with Eheim1250(1200 lph, max head 2.8psi ).
Cooling a Morgan1100 and stressing CPU with Jouni Vuorio's Stabilitity Test got( 3 seatings):

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

Intend to disassemble wb and fit with central water inlet to look-see at the infleunce of "Die Area Impingement".

Well this is OT but i'll answer by another question. I want a total silent (aka fanless) cooling solution for my ultra-high performance PC (max OC'ed AMD, ultra160 drives and so on). I want it to be portable and "not too big". My answer is watercooling. Got another solution ?

Sirpent,

You're really asking a question on radiator performance since a block will alway yield a higher convection coefficient with higher flow. Rather than get into gory details here, I'll try to make certain I give this topic its proper due in my next article at AMDMB. Actually, I've got a short one on fluids due to go up very soon, but that's already finished on my end. The next one I'm working on is radiators and blocks and will hopefully address a lot of the questions posed here and elsewhere.

Les,

1.7 ft/s is actually a very low flow velocity. Is this the "average" velocity in your block? Most of the pumps we use will crap out with peak velocity in our system around 10 fps or less. If we keep velocity in our tubing and radiator relatively low, with a spike entering the block we'll generally be close to "optimum" performance for a given setup.

gmat,

A few months back a guy wrote me from Europe seeking input on a pump-less water cooling system. His intent was to cool water to ~4°C and allow natural flow to develop based on the fluid losing density via heating in the block. I was skeptical, but upon running the numbers figured he could cool his chip with ~1 gph and a 20°C delta-T in his block. If that's enough density variation to generate 1 gph, he may be onto something. It is definitely not an over-clocker's solution, but can't be beat for quiet operation. I did not inquire about his source for getting 4°C water, but would hazard a guess that he'd use a TEC with large passive heat sink on the hot side.

Unregistered (BillA :p),

I owe you an apology. Based purely on second-hand comments, I took a couple of shots at some of your work. Not sayin' the comments won't be warranted, LOL, but I should say nothing without being certain of the true content. I have no excuse other than a natural ability to stick my foot in my mouth occasionally. I've read some of your stuff and found it to be of much higher caliber than most of what's available out here. Indeed, I'd like to discuss your radiator article in relation to my own prior to completing my work.

In the future, I'll try to get all the facts straight from the source and discuss one-on-one rather than going off half-cocked.

No comments from me yet (busy for next 4 days) but a quick noted I posted/found here:
http://forums.procooling.com/vbb/sho...4238#post34238

Like i stated, i want maximum overclocking.
Yet i wont go through the hassle of installing a chiller or a TEC. For one thats way too expensive (money doesnt matter, up to a point). For two it has too many drawbacks (reliability, moisture, weight, power supply problems...) to be a convenient solution.
If you want "low performance" but totally silent cooling the solution is heat pipes (passive phase change cooling).
But in my case i'm cooling just every hot part (CPU, GPU, NB, HDs, maybe RAM...) so encumbrance is a constraint. And i want to be able to carry my box :)

(edit) for those who don"t know them heat pipes use passive phase change convection to "pump" heat out of a tiny surface, to bring it to a big rad. This, without any moving mechanical part.

Yes. Calculated "average" velocity..
I am afraid calculating the velocity profile is beyond me.
Was hoping for 3x the average velocity( ie 5m/s for Eheim1250)[Edited Oct 30th (getting units mixed)]
Higher velocities were hoped for because with the Eheim1250 could get 150gph(US gal) in a Maze2.2 with a guessed X-sec of 36sq mm vessus the the 32sq..mm. Sums were very iffy in the first place not to mention failing to take account of the higher drag of the "flat" geometry.
Maybe the maximum velocities ,over the die area, approaches these values.
Thanks for taking the time and trouble sharing your knowledge expertise.

Les

wrt the die area:
open up the flow area to, and from (in the top plate)
then 'pinch' the flow from the sides (you're not trying to cool the corners)

and look for a low volume/high pressure pump
then you can put a good 'squeeze' on it

be cool

Bill.
Yes/No to both profile and pump.

It depends on which direction I take.
Have 3,enthusiasm dependant, plans:
1) Look-see at DAI by fitting a central inlet (keeping 2 oulets and same internal dimensions)
2) Obtain some Diode temps to compare with my "In Socket" Peltier temperatures - initial reason for getting a Diode reader.
3) Look-see at the inclusion of an intermediate shim and a crack at that TIM analysis nightmare - have you any intentions of a revisit?

this gd pos forum software won't hold a post while the window is being reused
bigass time waster

Les

1) my suggestions were to reverse the cross-sectional flow area 'ratio' over the die area when flowing corner to corner (narrow-wide-narrow to wide-narrow-wide, relatively of course)

this, across an optimized surface, will test the turbulent parallel flow scheme

jet impingment normal to the bp will certainally produce turbulence, the problem is that at our 'level' of testing we cannot easily demonstrate which is better
and cannot 'calculate' anything at all relating to such

I have all the wb test pieces to quantify flow rate/nozzle velocity/wb 'thermal resistance'; just doing other things right now

2) not my bag, good luck

3) doing goop testing 20hrs a day (Cooling Flow is better for testing), and just added another
http://www.evertech.com/category.cfm?Category=60
typical confusion (about testing) here
this review was fairly well done
(looks like I need to do an article on goop testing, jk)

be cool

myv65,

I am waiting to read your new articles on amdmb (the old ones are very good, imho!). I have one question (maybe, for your upcoming rad/wb article): My impression is that a waterblock and a radiator are very similar devices, from the heat exchange point of view. The only difference is in size. So if you write What you *do* need is high convection, aka high velocity in the block, doesn't the same apply to radiators as well? Implying BI original is better than BI pro...

Sirpent,

They are certainly similar, though the few differences are dramatic. A block must receive a large amount of energy through a relatively small area. It must then transfer this heat via conduction where a fluid picks it up via convection. A radiator also has three processes. It's convection from the liquid to the tubing, conduction through the tubing to the fins, and convection from the fins to the air.

So each takes heat input from a small area (region contacting the chip, ID of radiator tubing contacting fluid) and conducts it to a much larger area (surface area inside block passages, surface area of radiator fins). Due to vastly different properties from air to water, the convection across the radiator fins easily can become a bottleneck. I'm going to talk about each of these thermal resistances in a whee more depth during the article.

and to pick nits a bit,

there is an additional gradient across the soldered (or, worse yet - just mechanically swedged) tube/fin connection

be cool