New block (Hysterical post)

[Cathar]

Hmmm, surely spreading resistance can't be 0. If it were zero, then the source of the heat flux would have to be the same size as the (flat) convectional area.

[Les]

Originally Posted by Cathar

Hmmm, surely spreading resistance can't be 0. If it were zero, then the source of the heat flux would have to be the same size as the (flat) convectional area.

For a 50mm diam x 5mm thick disk (with Edge Coeff of 10kw/m^2*c and zero End Coeff) on a 10mm diam source.
Waterloo predicts a Spreading Resistance of ~ 0.00000167 C/W
Dunno whether this is reasonable - as said above "Have difficulty "getting my head round" all the edge cooling answers"
Maybe ask the Waterloo crowd.

[Incoherent]

Originally Posted by Cathar

Hmmm, surely spreading resistance can't be 0. If it were zero, then the source of the heat flux would have to be the same size as the (flat) convectional area.

Well, my thinking is as follows. Take a very small heat source at the end of a very thin wire length L, a 1D situation with L/k*A . Zero Spreading resistance.

Then add a 10mm thick plate. Imagine the plate is comprised of a large number of thin wires radiating from the heat source, each getting longer as their angle from the original thin wire increases in order to reach the surface of the plate. (Each one is also increasing in CS area, which offsets the increasing L somewhat) and the sum 1/Rspr =1/R1 + 1/R2 + 1/R3... + 1/Rlarge number is lower than the 1D resistance.
This impact from the increasing L is what I think of as the spreading resistance. Wrong?
In a spherical situation the L remains constant, 1/Rspr = 1/R1 x large number = very small number, in a hemisphere, 1/Rtotal = 1/R1 x large number/2. A very small number x 2.
Am I being stupid?

( R=k*A/L Edit: Oops that should be R=L/kA)

[Les]

Interpretation of "Spreading Resistance of Circular Source on Circular Disk with Edge Cooling" may well hold the answers.
Think a deeper understanding(than I have ) is needed :
"Spreading Resistance of Circular Source on Circular Disk with Edge Cooling" (when applying zero Edge Coeff ) appears to give different results than Isotropic Disk with Convective Cooling .I would have expected to be the same(maybe silly me but.....)

[Cathar]

Maybe my problem is in the interpretation of "spreading resistance".

The way I see it is this:

1-D resistance is as Les defined, travelling a fixed distance through some material where the (flat) convectional surface area is the same size as the heated patch. Heat doesn't move sideways in any fashion. There's no x-y movement of the heat. It only moves in one dimension, ala 1-D resistance.

Spreading resistance is naturally going to be the cost of moving heat in the other 2 dimensions when the (flat) convectional patch is larger than the heat flux going in. I mean, it's more difficult to move heat through the initial 10x10mm section of copper, than it is to move that same heat through the full size of the convectional patch (assuming the convectional patch is larger than the heat flux entry patch).

I utterly fail to see how given a 1D resistance as how it is being defined can ever result in a spreading resistance of zero when the heat patch and the convectional patch are of mismatched sizes.

[Incoherent]

Well it's leaving the knowledge level that I'm comfortable with.

I do now believe it (low/zero spreading resistance) is the reason that a block with 12mm between die and water can perform on a par with one having less than a tenth of that much copper in the heat path. With an average convection coefficient that is probably less than half, I think a third.

I have been swinging back and forth towards your way of thinking too Cather, over the last few weeks. To the point where I almost scrapped the block, but real (if somewhat lacking in credibility) data says it's true.
I now believe it.

Remember "...reduce the thermal gradient of copper"? I do this by increasing A, with L constant, not decreasing L with A constant.

[Cathar]

Yeah - I did simulate on my little proggy a stepped pyramid kinda deal for the block. i.e. imagine an Aztec pyramid shape. Not quite a sphere, but for these sorts of dimensions - close enough.

It did work out pretty well, being about 2C behind the White Water (from memory) but then again my thermal simulator is also not exactly a model of sophistication, being amateuristic at best.

Like most things I do in my life, I tend to focus around the 95-99% rule. I tried to analyse where 95% of the heat was going. i.e. as one moves vertically away from the die heat flux area, and then one attempts to draw a centered bounding shape around where 95% of the heat flux is moving through that bounding shape, the shape that I was simulating was something akin to a classic venturi funnel, where the gradient of the venturi curve is dependent upon the thermal conductivity of the metal. i.e. the bulk of the heat moves straight up for a while, really only slowly radiating out. After a vertical distance of about equal to the significant dimension of the heat source, only then did it radiate out fairly evenly as per a spherical sort of pattern.

For a point-source, indeed it is a hemi-sphere, but for a rectangular source that really is quite large in proportion to the vertical distance travelled, I don't think the model you're trying to use applies very well.

Of course, my theories/models on it could just be total crap as well....

Still - very interesting what you're doing here. Glad to see some research/innovation. It's been a while.

[Incoherent]

Originally Posted by Cathar

For a point-source, indeed it is a hemi-sphere, but for a rectangular source that really is quite large in proportion to the vertical distance travelled, I don't think the model you're trying to use applies very well.

You are quite right there. I do think however that the rectangular die has most of its heat generating components in a small section(s) of the overall die, more closely approximating a point heatsource. But of course there is heat spreading in the silicon itself too.
But if the isothermals directly above the die are like an inverted half globe, I think the model works.

Could all be crap.

[Incoherent]

Originally Posted by Cathar

Still - very interesting what you're doing here. Glad to see some research/innovation. It's been a while.

It's all been regurgitations of the Cascade/Whitewater theme hasn't it. Still, sincerest form of flattery...

Edit: With the notable exception of a few, MCW6000, R-type for example. ... Sorry Bill, Jaydee.

[Cathar]

Originally Posted by Incoherent

It's all been regurgitations of the Cascade/Whitewater theme hasn't it. Still, sincerest form of flattery...

Edit: With the notable exception of a few, MCW6000, R-type for example. ... Sorry Bill, Jaydee.

I was more referring at the enthusiast level. Jaydee has also been one of the few active people here.

Nice to see some attempt at rationalised theory behind it all too though, rather than people simply cutting up copper without any backing theory. That's what I was really getting at.

[Les]

Originally Posted by Cathar

Maybe my problem is in the interpretation of "spreading resistance".

The way I see it is this:

1-D resistance is as Les defined, travelling a fixed distance through some material where the (flat) convectional surface area is the same size as the heated patch. Heat doesn't move sideways in any fashion. There's no x-y movement of the heat. It only moves in one dimension, ala 1-D resistance.

Spreading resistance is naturally going to be the cost of moving heat in the other 2 dimensions when the (flat) convectional patch is larger than the heat flux going in. I mean, it's more difficult to move heat through the initial 10x10mm section of copper, than it is to move that same heat through the full size of the convectional patch (assuming the convectional patch is larger than the heat flux entry patch).

I utterly fail to see how given a 1D resistance as how it is being defined can ever result in a spreading resistance of zero when the heat patch and the convectional patch are of mismatched sizes.

Was dealing with a 1-D case.
When 3-D, the Spreading Resistance is the difference between the 3-D resistances

Originally Posted by Cathar

Yeah - I did simulate on my little proggy a stepped pyramid kinda deal for the block. i.e. imagine an Aztec pyramid shape. Not quite a sphere, but for these sorts of dimensions - close enough.

It did work out pretty well, being about 2C behind the White Water (from memory) .............................................

Possibly misleading without the stipulation of h

[Cathar]

Originally Posted by Les

Was dealing with a 1-D case.
When 3-D, the Spreading Resistance is the difference between the 3-D resistances

Oh, I totally agree. I'm just expressing some reservations about the terms being used by the Waterloo calculator, and attempting to wrap my head around the definition of the terms and the corresponding output for spreading resistance. The way they've defined the model just seems odd to me, even though the results that it produces at the end may be just fine.

Originally Posted by Les

Possibly misleading without the stipulation of h

True. Was a long time ago - but I was basically playing with 25K and 50K values for h. I make no claims to being accurate on that one though. It was an approach that I possibly erroneously discarded.

I actually broke my simulator a while back. I tried to make some improvements to the model, and got something wrong. It now just feeds out junk after a few cycles. Sadly/stupidly I didn't save a version of the working copy. I simply haven't had the time to dedicate to debugging it of late.

[Groth]

In my unfinished (interrupted) simulation experiments, I was seeing some combinations of bp/IHS/heff that favored edge cooling and some center. Combine the best parts of each and the result is... your block? Once I polish off a few of the current tasks, I plan to pick that up again.

Finite element analysis, GHz compensating for lack of theory.

[Incoherent]

Why it works

70W load, TIM C/W 0.063, CPU T 1mm below surface, k for this 392 W/m*C, k for block the same



Edit: This chart primarily to demonstrate the shallower gradient in the sphere.

[Cathar]

I suppose the followup question would naturally be this:

Can you sustain an effective h of 100K (very very difficult I might add) across a spherical surface of high surface area as opposed to a flat surface of smaller area?

As a followup consideration, flat plate blocks (Cascade included) typically engage, though surface/wall structure, two to three times the convectional surface area than a pure flat-plate model. Let's assume that the spherical model is doing the same with its available surface, how does that change the outcome once the 1-D conductional costs are factored in?

One further consideration - the thin-base flat-plate model's heat-spread pattern really isn't affected so much with changes in the size/area of the heat-die. With a thick-base spherical model, what happens as the die size approaches the size of the hemisphere and spreading resistance becomes a factor again?

Am just musing over the implications of the physical implementation, rather than the simplistic theory of it. Not intending to be negative at all - just to promote further thought.

[Les]

A little confused
Have only begun analysis but seem to have discrepancy.
Dealing with 24mm diam disk and hemisphere
Disk C/W(conv)= 1/3.142x0.012x0.012x100000 = 0.0221
Hemisphere C/W(conv) = 1/2x3.142x0.012x0.012x100000= 0.01105
Giving for a 100w load


Will continue tomorrow.

[Incoherent]

Now we start getting into the real details, limitations to the idea.

Originally Posted by Cathar

I suppose the followup question would naturally be this:

Can you sustain an effective h of 100K (very very difficult I might add) across a spherical surface of high surface area as opposed to a flat surface of smaller area?

I would agree, to the point where I'd say it is difficult to get this number up above 50000 for the sphere. I do have a few little tricks which are not visible in the pictures though. Anyway, in my simple model there is little benefit in trying to go higher, the sphere has very little headroom above about 115000, if you could generate an h of 1000000, it would have very little impact. The Cascade type on the other hand benefits immensely from an increasing convection coefficient.
I see it beating the sphere above 110-120,000 in the simplistic model. Quite within the Cascades reach.
The difference is, the sphere can outperform the thin base anywhere below this, at possibly less than half h.

Originally Posted by Cathar

As a followup consideration, flat plate blocks (Cascade included) typically engage, though surface/wall structure, two to three times the convectional surface area than a pure flat-plate model. Let's assume that the spherical model is doing the same with its available surface, how does that change the outcome once the 1-D conductional costs are factored in?

Another good point. By simply putting in an A multiplier for the Cu-H2O interface area into the model I see the C/W vs 'h' crossover point move. In the flatplate's favour. x2 makes the crossover point ~60000, x3 gives ~40000. The sphere does not benefit from a surface area increase much either.

Originally Posted by Cathar

One further consideration - the thin-base flat-plate model's heat-spread pattern really isn't affected so much with changes in the size/area of the heat-die. With a thick-base spherical model, what happens as the die size approaches the size of the hemisphere and spreading resistance becomes a factor again?.

I am not sure. Perhaps the ideal shape would become less spherical, heading towards a flat plate. In this scenario your first point (hard to maintain high 'h' over large area) starts applying to all blocks.
Edit: with a gaussian power distribution over the die I think the sphere remains valid.

Originally Posted by Cathar

Am just musing over the implications of the physical implementation, rather than the simplistic theory of it. Not intending to be negative at all - just to promote further thought.

I welcome it. It is good to probe the disadvantages, I started this thread hysterically excited that something I had come up with seems to actually work in a manner in line with the theory that generated it. As I said earlier, I have put a lot of effort into the mathamatics of this one. One of the goals was the perfect low flow block, I believe this, or something like it, is it. Another was to extend the "C/W vs 'h' crossover" point
Now to tweak and test further to try and refine the concept. It has a lot of copper there to provide thermal inertia. Might be a dark horse advantage.

I think that if one were to summarise the spherical concept it would go something like this.

1. In a low flow scenario it is unbeatable. (according to my interpretations)
2. It could be dirt cheap if made on a lathe. (but two pieces)
3. Can be made with a low pressure drop with very little performance loss. Not really sure about this, need emperical data.

But. 4. Very little performance gain from a stronger pump or refinements to convection structure geometry. It hit's a performance ceiling.

[Incoherent]

Originally Posted by Les

A little confused
Have only begun analysis but seem to have discrepancy.
Dealing with 24mm diam disk and hemisphere
Disk C/W(conv)= 1/3.142x0.012x0.012x100000 = 0.0221
Hemisphere C/W(conv) = 1/2x3.142x0.012x0.012x100000= 0.01105
Giving for a 100w load

Will continue tomorrow.

Not quite sure what you mean Les.
The chart above is for a 70W load, sorry did not note that last night.
My "hemisphere" is actually a 34mm (ish) diameter flattened top shape 12mm high. This complicates the area calculation a bit.

does that help?

[Les]

Originally Posted by Incoherent

Not quite sure what you mean Les.
The chart above is for a 70W load, sorry did not note that last night.
My "hemisphere" is actually a 34mm (ish) diameter flattened top shape 12mm high. This complicates the area calculation a bit.

does that help?

Some.

Not much progress and ,probably, still lacking in clarity
However


Probably as far as I can get without progressing from sums to mathematics(35years of rust)
Probably will not pursue further.
Do still agree that,particularly at low h, the dome should lower DeltaT - but the numbers elude me.
In consequence concur with most of your reply to Cathar.

[Incoherent]

I've had the block settling for a few days now. Results seem excellent. Except that it has started to leak. Around the barbs in the polycarbonate. Brass next time you idiot.

Here compared to the data I have for the Crater block.



I have done a bit of normalisation against the data that pHaestus has generated for the same block. here.
pHaestus' numbers seem to have some kind of temperature compression, reason unknown. I see this as his temps being about 85% of what they should be. This could be due to me seeing an "expansion" of course.
But anyway:

Edit: I've pulled this chart for being a wild drunken claim

Data is eyeballed from pHaestus' charts. 73W assumed.

[Cathar]

Would suggest that it would be more useful to have a common point of reference (i.e. the same block tested on both testbeds) before attempting a correlation of results.

Also Phaestus is using an 84mm^2 die-size area.

[Incoherent]

Originally Posted by Cathar

Would suggest that it would be more useful to have a common point of reference (i.e. the same block tested on both testbeds) before attempting a correlation of results.

Thats exactly what I do (not) have in the Crater. This is precisely the reason I sent the block to pHaestus, to be able to compare a common point of reference.

The die size issue can have several side effects, many/most taken care of by the "compression " multiplier. Although it is a significant point I think it can be normalised away.

[Cathar]

Hmmm - I didn't recognise the line for what it was. My apologies.

[Les]

Originally Posted by Incoherent

Thats exactly what I do have in the Crater. This is precisely the reason I sent the block to pHaestus, to be able to compare a common point of reference.

The die size issue can have several side effects, many/most taken care of by the "compression " multiplier. Although it is a significant point I think it can be normalised away.

Am happy with your data manipulation and collateral assessment
However confirmation by pHaestus would reassure

Ramblings :
Correlation of data from the same platform is difficult enough(eg size and TIM)
Correlation of the CPU and "Insulated Die" platforms is a nightmare.
Dimensions,. location of Diode and "Die sensor", Heat-flux character,and of course TIM
Perhaps all "Die Sim" should be normalized to zero(0) offset to "hot-face"

A "compression factor" is an answer
First play looked good.
Applied 0.773 "Multiplier" to Bill's data for LRWW
0.773 because I understand your heat-source(Flux-Block) is 110sq mm(0.85/1.1) and Bill's is 100 sq mm.


Application of 0.773 "Multiplier" to MCW6000 data gave :

Probably incorrect to use the same Compression Multiplier on Bill's old and new data, but.....

The only thing the above suggests(to me) is that any "Compression Multiplier " may be flow(h) dependent.
My candidate for the chief cause is the position of the CPU Diode.
This takes us back to here and pHaestus's "The calibration's fine but what is the calibrated diode measuring? "

[BillA]

and this question applies to every die sim
one known answer is to groove the die sim face and mount a TC; now the data is defined and comparable

if Incoherent adds a 3ed TC at a different spacing then he will be able to state a face temp (with presumably perfect insulation)