OMG I can't believe it. Everybody is copying Cathar!!!

I ran across this site

Tell me what you think? seems to me it came out earlier then cathar ro soemthing. Whatever it is, just want to let you know.

in any situation.. this day and age... there is always someone that had 'your' idea.. before you even thought of it....
no biggie....

I think micro-channels were around before I was born, or pretty close to it.

Still, they're quoting some pretty low convectional rates there. The White Water design is more than micro-channels.

That takes micro to a new level, it's like a skived jobbie is'nt it?...

thats only copying cathar if this is his second comeing... like he said, microchannel concept has been in the books for a long time and is used vastly in all different forms of manufacturing.

yup, its not an old thing at all, but Cathar is the first Overclocker i've seen to quantitatively reason out the value of microchannels. Those things in that mentioned article are way smaller than what is present in the whitewater

Relax guys. This is NOT a copy of Cathar. The WW is, in fact,
not even a micro-channel heat exchanger. It is a mini-channel.

What Cathar did was bring the lessons from a well-established
branch of heat transfer studies and optimally apply them to
what we as overclockers need.

You'd be surprised what else is being developed behind
closed doors.

Cathar's blocks use tiny milling bits wheras those use co2 lasers or the 'electrode-on-a-string' way of cutting (electrified metal wire 'vaproizes' metal in a solution, takes hours).

This technology isn't nearly as cheap as Cathar's techniques. The one pictured plate would probably make the WW look like a 486 cooler...

Tecumseh said:

Relax guys. This is NOT a copy of Cathar.

Click to expand...

did you even read most the post?
we're saying it isn't.... even cather say it isn't........

Originally posted by Cathar
Still, they're quoting some pretty low convectional rates there. The White Water design is more than micro-channels.

Click to expand...

How do you figure 35,000 watts/m2-°C is a low value? let's use the formula Q=h*A*(Ts-T0) to do some (very) rough calculations. assume the fin surface is at a constant temp, and has A=0.00087 m^2 (that's the surface area of the fins in 1/2" on an Evo) and a heat load of 100W. (T0 is water temperature).

At h=10,000 and T0=25C, the CPU temp would be about 36-38C, which sounds about right for convection coefficients since the standard CPU waterblock has more surface area.

At h= 35,000 and T0=25C, the CPU would be about 28-30C.
I somehow doubt that any waterblock today has convection coefficients higher than that.

And doing that with only a 1-2 psi drop is somewhat amazing.

Neo, a few things.

1) It's just plain wrong to make the assumption that the fins are of a constant temperature. That is an incredibly bad assumption to make, even for a "rough" calculation.

2) Convective transfer efficiency of 10-35K W/m^2C does not translate into a 100W 1cm^2 CPU being just 3-4C above the water temperature. That value does not take into account the thermal gradients through the metal (from base-plate to fins), nor does it take into account the thermal goop layer resistance.

3) Les has pretty much determined a convective transfer efficiency of around 30-50K W/m^2C for the White Water. Les can correct me if I've misunderstood/misremembered something, or if I'm looking at the wrong end of the graph.

As I said, those were very rough calculations, done with values pulled from memory. They were meant to see if we were in the ballpark. When I hear, "... they're quoting some pretty low convectional rates there..." I think Orders of Magnitude difference, not 0%-40%. As we all know, heat transfer isn't very exact when it comes to matching experimental results to theoretical calculations.

The assumption that the fins are of a constant temperature was only for a 1/2" length of the fins, directly over the core. The other 1.5" of the fins were neglected (swaying results towards higher convection coefficients because of the smaller surface area.) Although this is not an accurate description of the actual temperature gradient along the fin, it does allow for approximations to be made on the system, especially qualitatively.

The 3-4C difference between Fin Surface Temperature and CPU temperature was another rough ballpark figure. This was taken from some of my convection testing using a flat plate 0.25" thick. At high flow rates (convection rates), the difference between Die temp and surface temp in contact with the water was about 3-6C, and that was including the "thermal goop" interface. At low flows, there is a significantly higher gradient.

So, even with these gross generalizations that I made, I was approximating numbers between 10K and 30K for the Whitewater and Evo. Since these did not seem significantly different than PNLs numbers, I figured I would ask to see what values you had calculated for the WW.

I'd say we're playing in the same ballpark as PNL researchers. I wouldn't say that their numbers are "pretty low" until we're at h=75K-100K. Then we'll have something to brag about.

Neo, the thermal gradient vertically through the fins changes dramatically. It doesn't matter if you say only for 1/2" over the core or not, it doesn't apply.

"pretty low" - let's define this pragmatically since you seem to want to make a point of it. I want to buy something. Everyone seems to sell that item for $90-100. I find someone who's selling it for $60. That's a "pretty low" price and only a 40% difference. Or how about like this. Everyone's CPU temperature with waterblock X sits at 20C above water temps in a certain setup. Everyone comes to expect this. Waterblock Y comes along and the CPU temperatures seen are now 12C above water temps. That's "pretty low" and only a 40% difference.

You can bet that 35K W/m^2C is an extreme upper-end range, and I've come to expect that when someone says 10-35K, they really mean 10-15K almost all the time, and in some special extreme scenario, 35K, which you won't ever see in normal use.

Call me a cynic. Seen 100 too many nVidia marketing campaigns in my time.

All depends on how you want to view it, and whether you want to take a comment and write a rant about it...

Cathar said:

.
..........
3) Les has pretty much determined a convective transfer efficiency of around 30-50K W/m^2C for the White Water. Les can correct me if I've misunderstood/misremembered something, or if I'm looking at the wrong end of the graph.

Click to expand...

I am suggesting between ~30k and ~37k* for the same PD of 1-2psi(5-7LPM).
Lower pink(as defined by Excel) line on graph.
http://www.jr001b4751.pwp.blueyonder.co.uk/WW4.jpg

*Edit: Changed from ~35k

NeoMoses said:

. . . . . .
The 3-4C difference between Fin Surface Temperature and CPU temperature was another rough ballpark figure. This was taken from some of my convection testing using a flat plate 0.25" thick. At high flow rates (convection rates), the difference between Die temp and surface temp in contact with the water was about 3-6C, and that was including the "thermal goop" interface. At low flows, there is a significantly higher gradient.
. . . .
Then we'll have something to brag about.

Click to expand...

this is absolute 'trash talk' NeoMoses
you are tossing off numbers, and changing the story line when you are corrected

ok, stand and deliver - you have 'quoted' the numbers
you are being asked (somewhat politely) to substantiate them

describe your test setup, in detail
the heat source and how you measure the die temp
the measurement equipment and method utilized to determine the internal bp surface temp

be cool

Heat source: XP1600+
Die temp: measured with 8k3a+
All other temps measured with type K thermocouples. Thermocouple "A" is touching the side of the core. Thermocouple "B" is mounted 0.010" below the top surface where water hits. It is not in direct contact with water.

For the particular test that I am quoting numbers, a 1/2" inlet pipe was used and the height of the inlet opening as well as flow rates were varied.

A work order has been turned in for an instrumented die simulator, something that will give us control over the amount of power dissipated, and allow us to measure die temps more accurately. Right now, unfortunately, our only heat load comes from a processor (yes, I do understand the limitations...)

NeoMoses said:

1)Heat source: XP1600+
2)Die temp: measured with 8k3a+
3)All other temps measured with type K thermocouples. Thermocouple "A" is touching the side of the core.
4)Thermocouple "B" is mounted 0.010" below the top surface where water hits. It is not in direct contact with water.

(yes, I understand the limitations...)

Click to expand...

1) Wattage somewhere betwwen 50% and 110% of calculated(perhaps I exaggerate).
2) Uncalibrated?
3) Not sure what Thermocouple "A" is measuring . If ,being used as a check on the 8k3a+ diode temp( I presume the 8k3a reads the diode), I suggest it is not a very good one. I suspect that the temperature recorded will be just as closely related to the CPU's circuit-board and secondary heatsink as to the die temp.
4) The temperature recorded by Thermocouple "B" is also far from simple.It is positioned in a compound heat flux-channel of varying crossection and composition.
I do not believe the numbers from this set can be used even to calculate ballpark numbers.Interpretating the numbers is a nightmare before any consideration is given to accuracy.

Notwithstanding the above,the calculations(to my understandings) are also incorrect.
Even for the simple case of a flat Cu plate with perfect TIM,calculations have to account for the spreading resistance of the Cu plate.This will vary with Convection Coeff,die area, convection area(in this case plate area),and plate thickness.Posssibly the simplest way to estimate the spreading resistance is to use one of the Waterloo calculators*.Some examples of these calculations are shown in the graph.For the flat Cu plate case the x-axis is the Convection Coeff.


* http://www.mhtl.uwaterloo.ca/tools.html#

EDIT: Edited as a result of probable misunderstanding of the position of Thermocouple "B". Deleted "It will be closely related to bp/TIM interface temp, but how close?"

NeoMoses said:

Heat source: XP1600+
Die temp: measured with 8k3a+
All other temps measured with type K thermocouples. Thermocouple "A" is touching the side of the core. Thermocouple "B" is mounted 0.010" below the top surface where water hits. It is not in direct contact with water.

For the particular test that I am quoting numbers, a 1/2" inlet pipe was used and the height of the inlet opening as well as flow rates were varied.

A work order has been turned in for an instrumented die simulator, something that will give us control over the amount of power dissipated, and allow us to measure die temps more accurately. Right now, unfortunately, our only heat load comes from a processor (yes, I understand the limitations...)

Click to expand...

NeoMoses
I am going to presume in my following comments 2 things:
- that you understand what I am saying, and
- that your interest in testing is genuine,
-> and not the proliferation of psuedo-science as part of a self-promotion or product promotion scheme

the above quote is your response to my request for a detailed description
it lacks specificity in many areas, it is inadequate to enable another to evaluate the results you described
what are you hiding ?
(do you recall my comments about 'testers' unwilling to describe their equipment and/or procedures ?)

I believe there are 2 operative aspects:
you do not (yet ?) understand what you are trying to do, and
your equipment is not capable of doing what you are attempting

first to your understanding:
look at some of the papers by Aavid to see how they set up a calorimeter to quantify the TIM joint resistance
and google can provide you with hundreds of calorimeter descriptions
from an understanding of calorimeters you will be able to understand just how meaningless is a type K TC 0.01" reading 'under' the bp's inner surface

consider what you are doing:
taking a differential temp measurement
so the very FIRST thing that must be done is to cross-calibrate the temp sensors
-> they must read the same (or the offset KNOWN), and any non-linearity between them must be quantified

and if this is not done, you don't know sh*t about any actual difference
and if this is news to you, you don't know much of anything at all about testing
and if you already DID know this, then you are posting fraudulent data, aka bullsh*t

moving right along to your equipment:
Ill not waste words on using a CPU as a heat source, Radiate to your heart's content
what is an 8k3a+ ? - got an accuracy/uncertainty statement from the manufacturer ? (and why not ?)
but this is too rich ! type K thermocouples
wtf ?
do you know what the best possible accuracy is with a type K ?
why don't YOU tell us ?
THEN, what is the accuracy of your thermometer ?
so this makes the measurement's uncertainty what ?
(note that knowing the measurement's uncertainty has nothing at all to do with the measurement having any significance - see above)

NeoMoses
the burden is yours to demonstrate the validity of your numbers
otherwise . . . .

I would speculate about why you chose to get into this kind of a discussion with Les, Cathar, and myself

be cool

reading your counterposts always gives me a good chuckle BillA

I too enjoy reading these kind of threads. I'd be afraid to add technical comments.....I like my butt well cooled.....hate gettin' flamed.