A TIM C/W number.

[Incoherent]

Les you might be interested in this.

Some early numbers from my evolving test setup.
Quite preliminary at this point but I thought I'd share this with you chaps.

A copper die, 140mm^2, nominally 12x12mm but is a little smaller in reality.
A matching fluxblock with 3 thermistors. Again nominally 12x12x13mm but a little smaller due to lapping and so on. I do have exact measurements.
The "sphere" waterblock.

Calibration of the primary fluxblock channels is good, the middle sensor has a rough cal, the Die sensor is calibrated. Water Ts are roughly calibrated, they are high resolution so their curves are linear, I don't expect much error with these.



~0.039 C/W for a 140mm^2 die. No correction done for heat shadow effects etc.

A logfile here for those interested.

Many things going on that need to change, a power supply for the DAQ independent from the heater for a start.
I am seeing a few rather curious things happening. Fun to have the resolution to see them finally.

What is good to see is that heat flux through the fluxblock is linear. The deviation is minute. I will not use three sensors in the next 100mm^2 fluxblock, I do not think it is necessary based on the data from this one.

[Groth]

You make me feel all tingly.

[redleader]

How much mounting pressure? I'm curious if these figures are comparable to what a typical bare die CPU would experence.

[Les]

Originally Posted by Incoherent

Les you might be interested in this.

Yes, very much....
Have been waiting for the electronics thread(s) to emerge from the darkness(my electronics died with valves(tubes))
Looks sweet and does not clash with your CPU measurements (Link for redleader).

[Incoherent]

Originally Posted by Groth

You make me feel all tingly.

I think you have every right to feel tingly Groth, you've contributed a hell of a lot to this little project.

Redleader, Mounting pressure I have no real idea. I am using springs which are almost hard down but I don't know. A wild guess would be 8-12kg based on the spring hardness. I will measure this.

Right now I have a strange effect:



The water temperature is oscillating. Two pumps, a sick 110v eheim at 50hz and a Hydor Seltz S30. The temperature oscillation remains with either and both on. It is not an electrical/interference problem, the temperature stabilises when I restrict the flow and the period is too long I think. It is some kind of pump output power variation or a system time constant I think. Groth I am reminded of our pump/fan water PWM generator discussion.
I wonder if anyone (Bill?) has seen or can explain this.

Edit: I beg your collective pardons. I have discovered that it IS in fact an electrical issue. Measured the water (not distilled at this early stage), its got a 40VAC level relative to house ground. The eheim pump is not grounded. Grounding the radiator removes the effect totally but grounding the pump has no impact.

Effect of grounding the radiator:




You live and learn.

Edit 2: This is actually quite interesting. The signal I was seeing must have been induced into the water. It was there with either pump off, so both pumps must have been contributing. This means that this could be a generic problem, I wonder if this has any effect on anyones overclocks??? Use distilled water!

[Groth]

Capacitive coupling of the water to mains via the pumps? Inductive coupling with the tubing as a loop antenna? Cool either way. Got a DC pump? Can you do the tubing equilivent of twisted pair?

It's a shame the parallel port isn't fast enough for you to sychronize the reading to the mains phase. Perhaps a slight adjustment of the sample period?

[BillA]

re the pump(s)
yes, there is a cyclical pressure, and flow, variation
but the tc of your wb is greatly larger
use a moving average

like your sensitivity, but the value seems low
want some Shin Etsu for a comparison ? (for which I 'know' the value)

[leejsmith]

great work
It would be interesting to see what effect a p4/a64 style heat spreader has.

Are you going to try a range of tim's too.

[Incoherent]

Groth, it's pretty cool, I can frequency modulate that waveform by positioning the hoses. A "twisted pair" arrangement gives a quite high frequency...
I think I'll just ground the radiator.

Originally Posted by unregistered

re the pump(s)
yes, there is a cyclical pressure, and flow, variation
but the tc of your wb is greatly larger
use a moving average

like your sensitivity, but the value seems low
want some Shin Etsu for a comparison ? (for which I 'know' the value)

I have the log output as a .csv file so I can look at it with any kind of averaging.
Shin Etsu TIM data? very interested to see what others get.
As regards it being low, I can't comment. I think the method is sound (the linearity of the three series sensors is compelling) but I am willing to be convinced otherwise. It is always a little uncertain when inter/extra-polating.

I have just finished calibrating the water sensors after making them properly rather than being taped to a barb with foam tape.
I am also checking the correlation between electrical power, fluxblock power and power to water. The power based on voltage (which I am logging) and resistance (which is nominally 2ohms) , W=V^2/R , compares rather well with the fluxblock calculation (W=k*A*DT/L). At the moment I can not log current (missing a 0.01ohm shunt) so I am not certain of the electrical power number, the heater resistor has a tolerance of 1% and a T coeff of -0.002 - +0.008%/°C, so this can vary a little bit.



Insulation at the moment is...

... crumpled toilet paper.
Works fairly well. I need something easy to remove, neoprene blocks probably.

Does anyone have any suggestions for an insulating plate to go under my heater. Something rigid and with a high compression strength but with good insulating properties. I cannot seem to find anything which is much better than the Lexan plate I am using, all the phenolic resins and such have a thermal conductivity of about 0.2 W/m.°C - the same as Lexan. ?

[BillA]

make it with closed air gaps, minimize physical contact
phenolic is the best considering the possibility of high temps

no, not Shin Etsu data; what would data from a different test setup and using a different material illustrate ?

do you want some Shin Etsu G-751-S grease ?
I also have some 765 but no 'known' value for it (lower viscosity I believe)

[Groth]

Originally Posted by Incoherent

Groth, it's pretty cool, I can frequency modulate that waveform by positioning the hoses. A "twisted pair" arrangement gives a quite high frequency...
I think I'll just ground the radiator.

Tuning a resonator, eh? A bit of free body capacitance, a bit of loop inductance. It's good that it doesn't seem related to the pumps.

Faraday cage for your whole lab?

[Incoherent]

Originally Posted by unregistered

do you want some Shin Etsu G-751-S grease ?
I also have some 765 but no 'known' value for it (lower viscosity I believe)

How is the 'known' value defined? As a C/W for a given area?
Damned good of you to offer it, I would be happy to try it. It was not an original aim to test the differences between the different types of greases but it would actually be quite interesting, this is an ideal way to do it IMO.
My general attitude towards TIMs has been that anything works, as long as it doesn't evaporate or harden but I would welcome any education to the contrary.

Some more numbers today after 24 hours settling. BTW this is"M.G. Chemicals Cat. No. 860-4g Silicone Heat Transfer Compound"

TIM: 0.0359 C/W

And a power comparison from my newly calibrated water temperature sensors.

Power in: 72.61W (V^2/R)
Heat flux: 72.26W (k*A*DT/L)
Heat to water: 70.256W (FR*Cp*(Tin-Tout)) At a flow rate of 3.22lpm and 4186 heat capacity.

Seems reasonable.

[BillA]

you have an excellent way to test greases
e-mail me your address
I'll forward it to Colin or Nevin and I'm sure they'll send some Arctic Silver (which is what I thought you were using)

the value is 'known' wrt a processor/package, it is an integral part of all thermal load/temp calcs
given the similar effects of an IHS, the 'known' value is used pretty much across the board - 0.07°C/W for Shin Etsu G-751
note the value is to only 2 decimal places, this reflects the uncertainty due to application and not the equipment's capability
and since these calcs are to ascertain compliance with thermal design limits (or goals in development), the value is conservative

I am referring to TIM2, TIM1 under the IHS is a different kettle of fish

[Les]

Originally Posted by unregistered

you have an excellent way to test greases
e-mail me your address
I'll forward it to Colin or Nevin and I'm sure they'll send some Arctic Silver (which is what I thought you were using)

the value is 'known' wrt a processor/package, it is an integral part of all thermal load/temp calcs
given the similar effects of an IHS, the 'known' value is used pretty much across the board - 0.07°C/W for Shin Etsu G-751
note the value is to only 2 decimal places, this reflects the uncertainty due to application and not the equipment's capability
and since these calcs are to ascertain compliance with thermal design limits (or goals in development), the value is conservative

I am referring to TIM2, TIM1 under the IHS is a different kettle of fish

Perhaps an "across the board conservative" value of 0.07 is in line with Incoherent's 0.039 for a Silicone paste.
0.07 for a 83.5mm^2 T'Bred B die(with a zero spreading IHS) corresponds to 0.042 for a 140mm^ die

Agree the flux-block is a powerful tool

[Incoherent]

Originally Posted by leejsmith

Are you going to try a range of tim's too.

Only planning a 10x10mm at this point. I have many other things to do before I would call this data reliable yet so the small die will probably have to wait a bit.

Bill you have PM.

Damn, I thought I was going to be able to dispense with the fluxblock having once established the TIM value and confirmed the power readings...

[Groth]

Nope, in fact you need to upgrade your fluxblock slightly. Add a boss to both it and the die sim so that you can get repeatable micrometer readings and tell us how thick your TIMs are (as a function of pressure, please ).

[Incoherent]

Originally Posted by Groth

Nope, in fact you need to upgrade your fluxblock slightly. Add a boss to both it and the die sim so that you can get repeatable micrometer readings and tell us how thick your TIMs are (as a function of pressure, please ).

Oh I have already done that Groth.

32596.7 Å





Actually that's a real number, the calculated thickness based on the conductivity spec of the grease. = 3.26µm

[Groth]

Ahh, but it's those conductivity specs I want verified.

[Incoherent]

Originally Posted by Groth

Ahh, but it's those conductivity specs I want verified.

OK, I'll get to it. But I think that interferometry might be more appropriate...

[Les]

Originally Posted by Incoherent

Oh I have already done that Groth.

32596.7 Å





Actually that's a real number, the calculated thickness based on the conductivity spec of the grease. = 3.26µm

Assuming a simplified picture of uniform TIM layer and no contact resistance(TIM/Cu).
The usual picture is of TIM filling voids and then including Cu/Cu and TIM/Cu contact resistance
Have no authoritative links to hand (long time since looked at - put on hold - a can of worms)

Yes interferometry and load cell data would be useful but wouldn't expect a simple interpretation.

God, am struggling with the sheer volume produced so far in the Excels.
Converting to beermat format is a challenge.

[Incoherent]

Originally Posted by Les

Assuming a simplified picture of uniform TIM layer and no contact resistance(TIM/Cu).
The usual picture is of TIM filling voids and then including Cu/Cu and TIM/Cu contact resistance
Have no authoritative links to hand (long time since looked at - put on hold - a can of worms)

Yes interferometry and load cell data would be useful but wouldn't expect a simple interpretation.

God, am struggling with the sheer volume produced so far in the Excels.
Converting to beermat format is a challenge.

I agree about it being a can of worms, I was being tongue in cheek with regards to TIM thickness.
Load cells. Hmm you've given me an idea for measuring contact pressure. Does anyone have any links/sources for load cells, pressure sensors?

The data volume does get pretty heavy alright. For even more data I can actually sample at 1ms, as I have just tested, although I am concerned about the loading of the channels at this high rate. CPU load goes to 100%, logfiles assume massive proportions very quickly and we get saturated. But, it lets me perform some meaningful Fourier analysis in the search for mains hum and other periodic interference.

[Les]

Originally Posted by Incoherent

I agree about it being a can of worms, I was being tongue in cheek with regards to TIM thickness.

Yes, did notice the smile.
Thought would post before others got carried away.

Slowly accustoming myself to the Excels
However, even now, I realize your manipulation techniques are beyond mine; but will plod on.

On a more mundane point , yes, Kcopper=393 is probably a little high, dunno about Cp(water)

[Les]

Had a play, nowhere near as rigorous as your analysis
It is fascinating stuff
Have taken a random snapshot of 100 data sets(out of several thousand) from the calibrated log2 .
My manipulation skills not up to dealing with all at one sitting.
Averages :



Gives:
Power in (145.226601/R) Watts (cf 72.61W} (V^2/R) not found a value. for R
Heat Flux 72.098 Watts (cf 72.26W) (k*A*DT/L) k=293
Heat to Water 70.090 Watts (cf 70.256W) ( (FR*Cp*(Tin-Tout)) At a flow rate of 3.22lpm and 4186 heat capacity)

C/W values (140 sq mm die)
TIM : 0.0356 C/W (cf 0.0359 C/W)
WB : 0.0973 C/W
WB+TIM : 0.1329 C/W
WB+TIM+Offset : 0.1581 C/W (1.4mm Offset)

The (WB+TIM+Offset) C/W is usually quoted(e.g JoeC) as the C/W for the waterblock.
It would be fascinating to compare results with JoeC's(140 sq mm die)
Do you have any of the wbs he tested ?
Most preferred would be one also tested by Billa (on a 100sqmm die) and pHaestus(on a cpu).
Sorry, I am dreaming up work for you whilst I go to the pub.

[Incoherent]

Originally Posted by Les

It would be fascinating to compare results with JoeC's(140 sq mm die)
Do you have any of the wbs he tested ?
Most preferred would be one also tested by Billa (on a 100sqmm die) and pHaestus(on a cpu).

I don't have any commercial waterblocks, have always preferred to make my own of everything where possible, for both financial and personal reasons. At the moment I can't justify the cost of buyng one just to test it.

But you are quite right, a direct comparison is definitely necessary, an MCW6002 is probably a good choice so when the setup is a bit more verified I'll start saving pennies for that.

Originally Posted by Les

The (WB+TIM+Offset) C/W is usually quoted(e.g JoeC) as the C/W for the waterblock.

I would like to know the details of the Offset for JoeCs setup. I don't frequent OC very much so I have certainly missed that somewhere. Sensor location and dimensions primarily.

[BillA]

I'll send a 6000 with the grease, what mounting system ?