X4pro@SystemCooling.com

[Nugit]

I recently stumbled upon this review at systemcooling and found it to be very thorough and done on a decent testing bench too.

I have previously read reviews at systemcooling but they were mostly show-and-tell, this though includes c/w vs. flow, pressure-drop vs flow, c/w vs pressure drop et al.

Check it out: http://www.systemcooling.com/mcs_wc-01.html

[BillA]

Lee should be added to the JoeC, pH, Roscal group
makes 4 competitent testers (± depending on the type of testing)

progress I would say

[Les]

To allow cross-correlation of test-beds, would like the position of die's temperature sensor to be specified.

[RoboTech]

The hole for the temp probe is 0.1265" in diameter and runs thru the full length of the 32mm square step.
Temp probe is inserted 32mm (I've tried J-type thermocouple, Platinum RTD, but use thermal diode as it gives me 0.000 resolution.
The centerline of the hole is 0.161" from the top surface of the die.
The exposed die is 14mm^2 x 0.062" thick
Machined from a solid piece of C-110 copper



Hope that helps...

[pHaestus]

I am more of a non-tester. I'll get tenure soon enough though and the kids will start sleeping at night and then I'll have spare time again. Or at least I can dream...

[Razor6]

The MCW6002 is not performing very well compared to the RBX, almost the opposite to pH's results. It should be from the large difference in die size, pH 84mm^2 vs SC 196mm^2, correct?

[Les]

Thanks Robo for numbers.
Happy enough with MCW6003 v RBX
Behaviour is what I would expect from a "thick bp" v "thin bp"
eg Simulation for MCW6000 v WW

Not sure of overall "C/W" values.
Have a feeling that they are a little high.
Maybe mounting pressure on the 196mm^2 die is still(Asetek springs) relatively low(cf Incoherent's 13.8kg/cm^2.
Still playing with numbers(particularly MCW6002) in this and this fashion.

[Incoherent]

Originally Posted by Les

Not sure of overall "C/W" values.
Have a feeling that they are a little high....

They seem to be quite a close match to mine. Using my Lumiere number adjusted to 196mm^2, ((0.0515x 144mm^2)/196mm^2) and Waterloo to calculate spreading resistance in two sections (14.001x14.001x1.575mm with a 14x14mm heatsource [almost purely 1d] and 32x32x2.515mm with a 14x14mm heatsource, with h=very high number (infinity because there is no interface) and k=392, I get this:

We might even be seeing the difference between the 6000 and the 6002.

[Les]

Ah yes.
This change of X-section flux channel is part of my hesitancy.
Had paused for thought - look forward to your upload.

[Les]

Like offset numbers.
Robo's C/Wwb do seem a little high.
Don't think is MCW6000/MCW6002 difference ; I had also chosen to equate the two.
Would expect scaling with die area as indicated with your and Bill's data - still to do sums for 196mm^2 die.
Still thinking- getting slower and slower - still toying with shadowing(have not forgotten).

[RoboTech]

Originally Posted by Les

Not sure of overall "C/W" values.
Have a feeling that they are a little high.
Maybe mounting pressure on the 196mm^2 die is still(Asetek springs) relatively low(cf Incoherent's 13.8kg/cm^2
.

Thanks for your input and insight... Mounting pressure is certainly an uncontrolled variable for me. I try to use the hardware supplied with a particular block as that should be more representative of what the end user will experience. In the case of the 4XPro blocks, their springs were so weak I elected to go with heavier springs to get consistent numbers.
Now that I have a workable test stand (first to admit, far from ideal) I feel more like an apprentice with much to learn...

[Les]

Originally Posted by Les

.......Would expect scaling with die area as indicated with your and Bill's data - still to do sums for 196mm^2 die.
......

Using method outlined here get :



For info, "h(effective)" values :

[freeloadingbum]

Originally Posted by RoboTech

I try to use the hardware supplied with a particular block as that should be more representative of what the end user will experience.

Just to clarify, all the cpu blocks being compared were using their own supplied hardware and different mounting pressure. Is this correct?

[BillA]

to repeat a statement that many don't seem to want to hear:
the MCW6000 is better than the 6002
the MCW6002 was produced to afford those with 1/2" systems a means to more comfortably use the wb
any direct comparison I have made shows the 6000 always just a bit better, with a bit higher flow resistance due to the barbs

[Les]

Originally Posted by unregistered

to repeat a statement that many don't seem to want to hear:
the MCW6000 is better than the 6002
the MCW6002 was produced to afford those with 1/2" systems a means to more comfortably use the wb
any direct comparison I have made shows the 6000 always just a bit better, with a bit higher flow resistance due to the barbs

Have seen no Heat-die data ,by you, for the MCW6002.
Have only the Swiftech statement: " Refined data is due for publication soon. Meanwhile, MCW6000 data curves may be used for indicative performance levels"
I feel justified in using the same "C/W" values, at the same flow, for the MCW6000 and MCW6002.

[BillA]

no argument here Les
use the same #s for both
but in side-by-side testing the 6000 will best the 6002 (part of why Swiiftech uses 3/8"ID tubing)

[RoboTech]

Originally Posted by freeloadingbum

Just to clarify, all the cpu blocks being compared were using their own supplied hardware and different mounting pressure. Is this correct?

Yes, that is correct (except for the 4XPro as noted). I look at it two ways: (1) You can try and use one consistent mounting scheme to test all blocks that will produce the same clamping force. This would be ideal for analytical testing of blocks - primarily of interest for waterblock designers. or (2) you can use the mounting hardware supplied with each block that will not produce consistent clamping forces but will better represent how a waterblock will perform on an end users machine. Eventually I would like to be able to measure the clamping force of the various blocks similar to what Joe C. is doing...

[BillA]

I would vote for the mfgrs' hardware, thats what the user must cope with
I too like JoeC's setup, but not for consumers info because it is different

[RoboTech]

Originally Posted by Les

Like offset numbers.
Robo's C/Wwb do seem a little high.
Don't think is MCW6000/MCW6002 difference ; I had also chosen to equate the two.
Would expect scaling with die area as indicated with your and Bill's data - still to do sums for 196mm^2 die.
Still thinking- getting slower and slower - still toying with shadowing(have not forgotten).

Les and Incoherent,

Thanks for taking the time to look over my system and give it some thought...

Any ideas why my C/W numbers seem a little high? Any suggestions for how I might improve?

I don't fully understand what the graphs you generated are showing or what the offset is. Any clarification would be appreciated. Thanks...

[Incoherent]

Originally Posted by RoboTech

Les and Incoherent,

Thanks for taking the time to look over my system and give it some thought...

Any ideas why my C/W numbers seem a little high? Any suggestions for how I might improve?

I don't fully understand what the graphs you generated are showing or what the offset is. Any clarification would be appreciated. Thanks...

Robo. It's not really that I think that your numbers are too high. The C/W number that you are generating is the CW for your setup, i.e. you are measuring the C/W of the waterblock + TIM + copper between die surface and the temperature sensor.
What we are trying to do is compare your number to other tests.
We hope that the waterblock part of the total C/W is relatively constant. The TIM is a bit of a variable, however based on the measurements that I have made, it is fairly consistant depending on the compound type. Lumiere is a fairly thin grease, the number that I get for it stays pretty much the same over several mounts and pressures. It does change after a few days (settling from 0.0515 to about 0.0495). I think that applying my number to your data is not necessarily accurate but in the absence of anything else...
If we can accept this TIM value and apply it to your numbers, adjusting for the different die size we are left with the copper C/W. I did a quick calculation based on your drawing above using waterloo to try and establish a number for this. Ending up with an offset for your setup consisting of TIM C/W + Cu C/W = 0.0765 C/W.
In my setup I am able to generate, from a couple of assumptions and several measurements, a "true" WB C/W. In the chart I attached above I have added the offset for your setup to my data for the MCW6000 and overlayed it onto your data for the MCW6002 (we are considering them to be essentially identical). It is basically what my numbers say your setup should give. It is pretty close. Too close, we would expect that the waterblock should give a different (lower) number due to the larger die size although this behaviour must be waterblock design dependent. What Les is showing is that your numbers are not matching the predicted value. The culprit is not necessarily your data, it might be my treatment of it - or of course my numbers, mounting pressure might be more significant than I think. However, I would reiterate my concern about the probe active area and location.
A possible suggestion would be to make the die heat channel unstepped, i.e. make it 14x14 for a longer distance from the surface so that the sensor is in the 14x14 channel, that way we could be sure that the temperature gradient is fairly linear and hence easier to calculate what the offset should be . Pain in the arse I know.

[Les]

Little to add.
.Maybe worth digging deeper into TIM vs Area.
With pressure and probably flatness(using "1/average gap" as measure) decreasing with area,then,perhaps, TIM resistance is not linearly related..
Am taking a leisurely look at some Waterloo work.There maybe others more pertinent.

[RoboTech]

Thank you Gentlemen,

Lots I need to learn about thermophysics (among other things)...

I designed the die block with the thermal probe mounted in the larger 32mm X-section because I thought it might allow the probe to be relatively close to the die surface (reducing dT across copper) with minimal shadowing of the heat flux by the temp probe hole.

I'm also thinking about incorporating a load cell into my next rev to measure actual clamping force.

Back to reading up on some of the links posted to U of Waterloo examples and calcs! Thanks again,