Maximizing Surface area where it is needed.

[jaydee]

One common thing I see from newbie water block designers and makers is they are not concentrating on maximizing surface area and convection over the area where it needs to be. I will try to explain it the best I can which isn't perfect, but hopefully some more educated people will jump in here and make this a tutorial about the subject?

Well here is an example I just drew up:


This is an example of what I am about to get at. IMO you should concentrate on the area right about the CPU die. The red square in the pic is the CPU die. I like to go about 15% further on each side to allow for moderate heat spreading. If you make your base thin enough you should be able to keep that heat in the area you want it, which in the pic is the pins and dimples. I got 2 sets of surface area in the pic. One is the dimples which if I where to build this block I would have a middle piece modified that had holes directed at those dimples, maybe pipes instead like in Cathars design here: http://forums.procooling.com/vbb/sho...threadid=6666. Then the second set of surface area is the pins. My theory here is the water will first utilize the dimples surface area and then the water expelled from those would run around the pins which is the second set. Should in theory be maximizing the surface area above the heat source and should result in efficient cooling.

The deisng in the pic was made to have a single big inlet over the middle and a single outlet. The channel winding around the block really isn't needed, but might add the extra capability of using a TEC/pelt if so desired in the future.

So if anyone want to expand on this please do. A design tutorial is badly needed around here as people seem not to interested in searching for designs to get ideas off to make their own designs. And we might be able to add the theory of the design which might be missed in whatever thread they might be looking at. Well it is late here so if this doesn't make since then sorry.

[8-Ball]

The way am beginning to see waterblock design is as follows.

It isn't sufficient to think just about surface area or "turbulence". You need to consider the two together, and in fact, an understanding of the conductivity through the copper would help as well.

OK, so how does this work.

For a given surface, and a given "flow" of coolant, the heat exchanger will have convective heat transfer coefficient (CHTC).

What is this?

In a heat exchanger, you are often considering the transfer of thermal energy from a solid surface, (in this case, the surface of the fins/pins) to a coolant fluid (water in this case, but air is also a "fluid").

Why convective heat transfer?

There is a layer of slow/sationary moving fluid adjacent to the surface of the heat exchanger, through which the thermal energy must travel, in order to reach the fast moving stream of coolant, where it can be transported to the radiator.

For a given coolant, the CHTC will be determined by the flow velocity and the geometry of the heat exchanger.

What does the CHTC mean?

Essentially, this is an efficiency, (the inverse of a resistance) so W/C. So for a higher CHTC, the temperature difference required between the SURFACE of the heat exchanger, and the flowing coolant fluid can be lower for a given heat load.

However, we also need to consider the surface area fo the "furniture".

If the CHTC were constant, then it would be simple to say that you want the maximum surface area possible. However, it is never simple.

The "trick", is to maximise the surface area, with the best possible convective heat transfer coefficient (normalised, per unit area of surface), and do this just above the core.

BUT, you also want to do it in such a way, that the thermal energy does not have to conduct very far to the convection surface. It's all very well designing a waterblock with furniture all across the base plate, but if there is insufficient thermal gradient to get the thermal energy out there, then it was a wasted effort.


I hope this makes sense.

8-ball

[Cathar]

Waterblock design is all about balance, but ultimately, for a high performing waterblock design, it is about maximising convective thermal transfer (or CHTC as 8-ball points out above) from the metal surface to the coolant. Actually I believe it's better to refer to it by its density, being the convective heat transfer density (CTHD), expressed as W/m^2K.

Maximising surface area where it's needed most is implicitly accepting that it is not possible to maximise the CHTD to a high enough level whereby the heat may be dissipated using just the surface area immediately above the copper die.

This gets into base-plate thickness issues as well. The thicker the base-plate, the wider the thermal spread, and therefore the less of a need for CHTD to be maximised to deal with the wider heat distribution.

The issue with thicker base-plates as that they create a larger temperature differential from the heat source to the convective surface. Thick base-plates are essentially bad if we want to move forwards with block design. For a 10x10mm heat die of 80W, there's about a 2.5C difference to move through the first millimeter of pure copper. By then the heat has spread out a bit further laterally and the next millimeter "costs" less. To move through the next millmeter to a flat convective surface "costs" about another 1.8C. The 3rd millimeter costs about another 1.4C. The 4th millimeter costs about 1.1C extra. The 5th millimeter costs about an extra 0.8C. By the 5th millmeter though the bulk of the heat has spread out to roughly 4x the surface area of the 10x10mm die, consequently requiring a lower CHTD to deal with the heat.

So we want a thin base-plate, but this is not so easy, because this hinges solely on the CHTD. If CHTD isn't high enough to deal with a too thin base-plate, then the rise in temperature of the metal surface above the coolant may very well be greater than the cost of conducting heat through the copper for a slightly longer distance.

Let's assume we have a CHTD of around 25000 W/m^2K, which is fairly typical of a single inlet above the die design on a flat base-plate like the Swiftech MCW462 blocks. If the copper base-plate is 1mm thick, we incur a 2.5C cost for the copper conduction, and have about 144mm^2 of surface area with which to cool the bulk of the heat (I say bulk of the heat because some heat does conduct laterally further, but for the ease of simplicity we'll assume that 100% of the heat is concentrated in an inverted pyramid shape projecting upwards from the edges of the heat source). This gives a net C/W of the convective surface of 0.277, or for an 80W heat source a rise of (0.277 x 80 =) 22.2C. Add on the 2.5C of the copper as we get a total of around 24.7C.

At 4mm thick, we have 324mm^2 of effectual convective surface area, and a 6.8C copper conduction cost. Net convective C/W is 0.123, for a total temperature differential of 6.8 + 0.123 * 80 = 16.7C.

Make the same base-plate 5mm thick, and our copper conduction costs us 7.6C. Over the 400mm^2 area though, the net C/W of the convective surface is 0.10. Total temperature differential therefore is now 7.6 + 0.10 x 80 = 15.6C. We could keep on going but this expresses the point well enough.

Please note - these are just all rough calculations to express the point, rather than intending to be factual statements of reality. They also don't take into account the cost of a thermal interface material layer.

Clearly the CHTD of around 25000 is not high enough to counteract the increased heat density of a 1mm thick copper base, and a 5mm thick copper base gives lower temperatures, despite the extra 5.1C of copper conduction cost.

So we have three choices available to us at this stage if we want to see an improvement.

1) Keep the base-plate thin and raise the effective surface area (ESA)
2) Raise the CHTD
3) Attempt to do both

What jaydee is suggesting is 1). By attempting to raise the surface area say 4x after the first millimeter of base-plate we're attempting to lower the cost of copper conduction while maximising surface area to the same levels of travelling through 5mm of copper base-plate. of course the heat still has to conduct up the pins and "furniture" so it's not going to be a 1:1 tradeoff. We're attempting to claw back some of that 5.1C temperature differential of the 5mm base-plate.

By adding jets we also attempting to do 2). If we increase the CHTD as well, then we win on two fronts.

CHTD may be improved through various jet impingement methods.

So a good block design will therefore minimise base-plate thickness, maximise CHTD and maximise ESA. What becomes the real trick though is deciding how to manufacture the design such that these goals are achieved. Another more subtle point for those that have followed this far is that the need for more ESA can become diminished if CHTD is raised far enough (ie. effective direct die cooling).

[Les]

Much posted previously but in my opinion a a fair summary:

[#Rotor]

I do not understand the need for a definitive path, given to the fluid when it has already passed through the flashpoint. In my mind, such a path only adds volume to the block, but does not add value to it, from a cooling point. I can agree with the inlet being on top of the core, but only in Cathar's case of design, where a jet, induces a much higher velocity and from that the turbulence induced, multiplies the effective surface area.

for the noob, trying to rap his mind around all this, think of it like so. There is a bunch of constants involved, and then there's a couple of things, that one might call multipliers....

for instance:

the total effective surface area of any give design, is a constant.
So is the actual volume of said design.
The aerodynamic characteristic, or should I say fluidynamic.... of the flow path is a constant to, a very important one.

Now obviously the flow rate is a variable, that depends on a bunch of other things, external to the block.

flow rate should never ever depend on the block's design. The block is not there to make the system flow properly. That is where so many go astray, because as soon as you start optimizing block internals, to make the pumps life easy, you loose cooling performance by the gigawatthours and a half.....


turbulence, I like to think of as a multiplier. It apparently seems to multiply the effective surface of the block. I'm saying apparently, because we all know it does not really do that, even though the amount of cool water getting in close proximity to the copper, is greatly increased by it.

So how does all of this go together

In my mind, the ideal is to be able to get every single water molecule, to touch the surface of the internals of the block, and immediately after doing so, be evacuated with as little interference as possible to the incoming molecules.

Believe me, that's a tough one.

You want to have as much surface for as little volume. The smaller the volume, the faster it can be replaced with fresh cold fluid. The bigger the surface, obviously, the more fluid molecules will touch it at any point in time. for this surface to volume thing, one needs to get into Mathematical shapes and the relation between there volume and surface area. this... might give one an idea of what it's about.

Some of the laws involved are mentioned here

[jaydee]

Excellent stuff guys. A mod can rename this thread to something more suiting if they want because it may not entirely apply to maximizing surface area.

[jaydee]

Bump for the new guys.

[WAJ_UK]

I know this is a bit of an old thread thought I would revive it. I was doing some simple calcs today to estimate the effect of the baseplate.
Assuming a 100W , 10mm by 10mm heat source and a C101 baseplate.

I'm also assuming that the heat spreads at 45 deg from each corner of the heat source. Do you guys think this is a reasonable approximation for how the heat spreads? If not is there a better way of doing it?
Just thought I would throw the graph in as well

[OvcA]

Originally Posted by WAJ_UK

I know this is a bit of an old thread thought I would revive it. I was doing some simple calcs today to estimate the effect of the baseplate.
Assuming a 100W , 10mm by 10mm heat source and a C101 baseplate.

I'm also assuming that the heat spreads at 45 deg from each corner of the heat source. Do you guys think this is a reasonable approximation for how the heat spreads? If not is there a better way of doing it?
Just thought I would throw the graph in as well

I am aware that this link will not do you much good since the forum is in slovene but it shows a rather nice analysis in Ansys DesignSpace of the effects of baseplate thicknes on thermal transfer.


Ambient temp: 30C
Heat current: 80W (an air cooler is simulated)
Convection coefficient: 50W/m^2K

[Cathar]

Originally Posted by OvcA

I am aware that this link will not do you much good since the forum is in slovene but it shows a rather nice analysis in Ansys DesignSpace of the effects of baseplate thicknes on thermal transfer.


Ambient temp: 30C
Heat current: 80W (an air cooler is simulated)
Convection coefficient: 50W/m^2K

Where's the link?

Water's convection coefficient, depending on the design, can be >50000W/m^2K, or about 1000x more than air. This has a dramatic effect on just how thin a base-plate can be, and still be effective.

[OvcA]

Silly me, forgot the link.

Here it is

[Incoherent]

Originally Posted by Cathar

Where's the link?

Water's convection coefficient, depending on the design, can be >50000W/m^2K, or about 1000x more than air. This has a dramatic effect on just how thin a base-plate can be, and still be effective.

Cather, do have any idea what kind of effective convection coefficient you are getting with the Cascade? 50000W/m^2K seems a bit low to me, based on the kind of performance you are getting from it. After some playing with numbers I'd say that if you were getting an h=~50000, the baseplate would have needed to be at least 5mm thick to get near the numbers you are getting...

Dunno, thinking...

Incoherent

[Cathar]

Originally Posted by Incoherent

Cather, do have any idea what kind of effective convection coefficient you are getting with the Cascade? 50000W/m^2K seems a bit low to me, based on the kind of performance you are getting from it. After some playing with numbers I'd say that if you were getting an h=~50000, the baseplate would have needed to be at least 5mm thick to get near the numbers you are getting...

You're quite right. I was just throwing a number out there at 50K as a comparison to water having around 1000x the convectional coefficient of air.

In another thread here I speculated that the Cascade is in the vicinity of 80-90K when flowing at 10LPM.

[Incoherent]

Originally Posted by Cathar

You're quite right. I was just throwing a number out there at 50K as a comparison to water having around 1000x the convectional coefficient of air.

In another thread here I speculated that the Cascade is in the vicinity of 80-90K when flowing at 10LPM.

These seem like impossibly high numbers even allowing for the fact that it is an "effective h" for the cup region rather than a strictly accurate calculation of surface area, turbulence, viscosity bla bla.
Not doubting it though, it's just that conventionaly speaking this would require that the liquid be boiling or something. Could it be? Some localised low pressure in the cups...

...the cogs are turning slowly these days...

Cheers

Incoherent

[BillA]

said it before, I'll try it again
you guys are playing with yourselves

you are attempting to characterize "h" as an apparent average value;
this will not yield a useful result
and the experimental technique is insufficiently developed to measure (back-calculate) actual values of h with any effectiveness
-> a value of "h" is always associated with a specific velocity and direction, AT a specific location

noobs
disregard this thread, it is confused

[9mmCensor]

Originally Posted by unregistered

noobs
disregard this thread, it is confused

too late, my brain has melted down

[Cathar]

Originally Posted by unregistered

said it before, I'll try it again
you guys are playing with yourselves

you are attempting to characterize "h" as an apparent average value;
this will not yield a useful result
and the experimental technique is insufficiently developed to measure (back-calculate) actual values of h with any effectiveness
-> a value of "h" is always associated with a specific velocity and direction, AT a specific location

noobs
disregard this thread, it is confused

Bill, am very much aware of this. I am no way fooled that h is a consistent value across the surface area. People do keep asking what h is though, for which I can only guess is for running some ANSYS simulation of some kind....so I do comply with an "average" value.

Had been playing with some singular jet designs and explicitly exploring bp thicknesses required due to variations in h across the impingement region. Very interesting stuff, mostly focused around what to do at a smaller scale.

Incoherent, an average h of ~120-150K is well within the realms of possibility with "commonly available" pumps with water without requiring the water to boil. Not talking about the Eheim's though, but higher pressure pumps like the Iwaki's and certain 12V pumps. Again, not that h is consistent in any fashion, but it is possible to shape various aspects to achieve such an "effect". As much as it will grate Bill for me to say it (again) I believe I have successfully demonstrated an average h in the vicinity 110K, but I don't think for a second that I can then whack that atop the thinnest possible bp which that average value would theoretically support - with localised spot values for h theoretically dropping down to ~10K.

Am focusing work now around limiting both the size and the effect of those "low-point" spot values of h, although sometimes I feel like I am attempting to build the space-shuttle with an abacus and a chisel...

[BillA]

Cathar
I believe you are doing your acolytes a dis-service by catering to their desire for (over)simplification
- that you (and I too, eh ?) use a simplification is of little import so long as the degree and extent of such is understood
-> that 'science' be also described so is quite harmful as those that do not understand are going to be beguiled into thinking that they do

please review the above thread,
h is never accurately defined, and the ENTIRE 'discussion' is a sham
(all wb calcs of "h" are pure bs; the "h" of the areas of significance are completely hidden within the overall average values calc)

WAJ_UK
Electronics Cooling had an article on the 45° assumption

[Cathar]

Fair enough Bill. It's not been the first time I've been chastised for entertaining the over-simplification of inherently complex matters. Will accept that this thread has gotten off track.

[Incoherent]

Originally Posted by unregistered

said it before, I'll try it again
you guys are playing with yourselves

you are attempting to characterize "h" as an apparent average value;
this will not yield a useful result
and the experimental technique is insufficiently developed to measure (back-calculate) actual values of h with any effectiveness
-> a value of "h" is always associated with a specific velocity and direction, AT a specific location

noobs
disregard this thread, it is confused

Oh well, you're probably right Bill
Perhaps there is a more appropriate term. I am trying to quantify the effectiveness of a given design, at a given baseplate thickness, across various flowrates. "h" or at least an effective average something like a global "h" seems to me to be a reasonable approach. I'm not trying to calculate it, rather compare what other parameters should be doing when it changes. Fit this to existing data and thereby generate some kind of figure of merit for a given design.

Originally Posted by Cather

...an average h of ~120-150K is well within the realms of possibility ... ...it is possible to shape various aspects to achieve such an "effect".

I understand I think. Agree.

Late...

Cheers

Incoherent

[BillA]

"figure of merit"
a favorite exercise of mine also
but since such is dependent on the weighting assigned, ONLY the user can actually define

good topic for a contentious thread, go for it Incoherent

[pHaestus]

Ask any chemisty or physicist and they can give you the value of h to at least 3 decimal places (Planck's constant).

[Incoherent]

Originally Posted by unregistered

"figure of merit"
a favorite exercise of mine also
but since such is dependent on the weighting assigned, ONLY the user can actually define

good topic for a contentious thread, go for it Incoherent

I may just do that.

[WAJ_UK]

Blimey lots of activity here yesterday
If you look at PH's graph http://forums.procooling.com/vbb/showthread.php?t=8941

There seems to be 2 distinctly different curves. The Swiftech has a very different curve to the other blocks. Is this largely because of the baseplate thickness? It seems to make sense to me in my little world that the base plate thickness would have quite a large effect on the shape of the curve. I'll have to try and find that article Bill mentioned

Edit

Found the article quite informative. I think I'll have to crack out the Ansys. I ought to be revising Ansys now anyway for one of my exams although we are only doing structural stuff

[Incoherent]

Originally Posted by WAJ_UK

Blimey lots of activity here yesterday
If you look at PH's graph http://forums.procooling.com/vbb/showthread.php?t=8941

There seems to be 2 distinctly different curves. The Swiftech has a very different curve to the other blocks. Is this largely because of the baseplate thickness? It seems to make sense to me in my little world that the base plate thickness would have quite a large effect on the shape of the curve. ..

WAJ_UK, funnily enough this is exactly what I have been looking at. I agree.

I have been violently simplifying the whole problem and have generated a "waterblock simulator" which has 2 variables: Baseplate thickness (average) and FOM.

Without knowing the actual bp thickness of any of these blocks I'll stick my head on the block and see what comments it generates. Oh dear, bad pun.
Before I get crucified for spreading bad science, this is just some thoughts and an attempt to simplify the problem FOR MY SELF. The BP thickness is a guess. The FOM is actually based on apparent local surface area, more turbulent blocks appear to have a greater surface area. "h" is just "h", the temperatures are generated based on these values which could be miles off. The blocks are fitted to phaestus's data, the MCW6000 is a guess based on Swiftech C/W data, "h" being somehow proportional to flowrate.


All horribly arbitary.

Incoherent