Direct Die Cooling Project.

[Cathar]

Originally posted by BrianW
Whats an esky? A portable insulated storage box, which we call a cooler here in the states?

BrianW

Yes. Esky is just the Australian term for it. I think it was a brand name for a portable insulated storage box back in the 70's and has basically embedded itself into the list of Australian euphamisms. Everyone here calls them Eskies. If you said "cooler" to most Australians they'd have to think twice before realising what you meant.

[nicozeg]

Originally posted by jaydee116
I think the biggest gain this might have over a standard block is the +-8C gain from having no TIM joint . If I can get the volocity high enough through the jets it should work well. Water shouldn't have any problem cooling the die as long as it is moving fairly quick.

That number can't be real, the real gain is a small fraction of that for a well lapped base.

[t00lb0x]

JD: are you gonna build the chiller custom or convert it from an old air conditioner.

[jaydee]

Originally posted by nicozeg
That number can't be real, the real gain is a small fraction of that for a well lapped base.

I assumed it was as that is what Bigben2K was using in another thread. Not sure where it came from though.... I thought it was high myself and it might be. :shrug: Maybe that was the TIM with the thickness of the block added in. I don't know...

[jaydee]

Originally posted by t00lb0x
JD: are you gonna build the chiller custom or convert it from an old air conditioner.

I am probably going to use a couple 226watt TEC's. I am far from that stage though.

[bigben2k]

It was from Cathar's testing. 8 deg C is the gradient through the TIM joint, given an 80W source.

I still have to test those Thermagon samples... but I don't expect much.

I really gotta put a testbench together.

[t00lb0x]

Originally posted by jaydee116
I am probably going to use a couple 226watt TEC's. I am far from that stage though.

For a peltier to work properly it has to have pressure on both sides. You know this, now are you going to use the peltier under the waterblock, or use the peltier to cool the water going through the waterblock?

[BrianW]

He is gonna use them to make a chiller.

BrianW

[Cathar]

Originally posted by bigben2k
It was from Cathar's testing. 8 deg C is the gradient through the TIM joint, given an 80W source.

Actually not my testing but by way of BillA. Bill had forwarded me a very interesting graph once that plotted the thermal junction difference betwen various points using thermal sensors placed at different levels between the heat source and the water coolant.

Roughly speaking it was showing around a 0.08 to 0.10 C/W for the thermal interface junction (when using goop) for a 100mm^2 heat die.

[Les]

Originally posted by Cathar
Actually not my testing but by way of BillA. Bill had forwarded me a very interesting graph once that plotted the thermal junction difference betwen various points using thermal sensors placed at different levels between the heat source and the water coolant.

Roughly speaking it was showing around a 0.08 to 0.10 C/W for the thermal interface junction (when using goop) for a 100mm^2 heat die.

Originates from an inconclusive discussion:
http://forums.overclockers.com.au/sh...threadid=67401

The Off-forum,rather inadequate analysis of the data should be treated very cautiously.
To assume the same TIM Joint resistance for different wbs is dangerous and can lead to possibly erroneous conclusions.For example would suspect that the fly-cut(?) surface of "The Cascade" produces a lower TIM joint resistance than the hand-lapped " White-Water".Think it is inconclusive that the reduction in temperatures for "The Cascade" is due to lower Convective Resistance and not due to a lower TIM Joint resistance.

[gone_fishin]

Originally posted by Les
Originates from an inconclusive discussion:
http://forums.overclockers.com.au/sh...threadid=67401

The Off-forum,rather inadequate analysis of the data should be treated very cautiously.
To assume the same TIM Joint resistance for different wbs is dangerous and can lead to possibly erroneous conclusions.For example would suspect that the fly-cut(?) surface of "The Cascade" produces a lower TIM joint resistance than the hand-lapped " White-Water".Think it is inconclusive that the reduction in temperatures for "The Cascade" is due to lower Convective Resistance and not due to a lower TIM Joint resistance.

In other words, there is no conclusive evidence that the design itself is truly a step forward as implemented. Wow Les, that's layin some smack down.

[Cathar]

Originally posted by Les
For example would suspect that the fly-cut(?) surface of "The Cascade" produces a lower TIM joint resistance than the hand-lapped " White-Water".Think it is inconclusive that the reduction in temperatures for "The Cascade" is due to lower Convective Resistance and not due to a lower TIM Joint resistance.

Both blocks were hand-lapped to the same degree using the same technique.

The fly-cut is an initial step that removes the "dirty" section of the extruded copper that typically has streaks of air-bubbles just tiny fractions of a millimeter under the surface. On the White Water's I had to "lap" though that section. On the Cascade's that is just fly-cut away. The fly-cut doesn't produce a particularly flat surface at all. It is ridged enough such that you can feel it with your fingernail, hence the hand-lapping of the block after being fly-cut.

[jaydee]

Originally posted by t00lb0x
For a peltier to work properly it has to have pressure on both sides. You know this, now are you going to use the peltier under the waterblock, or use the peltier to cool the water going through the waterblock?

I have designed several different chiller blocks just never made any yet. The two TECS will fit in between two blocks. Cool side chills the water, hot side on a seperate loop.

[gone_fishin]

Originally posted by Cathar
Both blocks were hand-lapped to the same degree using the same technique.

The fly-cut is an initial step that removes the "dirty" section of the extruded copper that typically has streaks of air-bubbles just tiny fractions of a millimeter under the surface. On the White Water's I had to "lap" though that section. On the Cascade's that is just fly-cut away. The fly-cut doesn't produce a particularly flat surface at all. It is ridged enough such that you can feel it with your fingernail, hence the hand-lapping of the block after being fly-cut.

Didn't the review by Bill show from optical surface inspection a slightly non flat area in WW? Too bad he didn't get to publish a comparison review of the Cascade before he went on to bigger and better things.

[nicozeg]

Originally posted by Cathar
Actually not my testing but by way of BillA. Bill had forwarded me a very interesting graph once that plotted the thermal junction difference betwen various points using thermal sensors placed at different levels between the heat source and the water coolant.

Roughly speaking it was showing around a 0.08 to 0.10 C/W for the thermal interface junction (when using goop) for a 100mm^2 heat die.

I doubt that numbers for several reasons: First lets make some calculation with AS3 specs.

Thermal conductivity: >9.0 W/mK (Hot Wire Method Per MIL-C-47113)

Thermal Resistance: <0.004°C-in²/Watt (0.001 inch layer)

For an 1cm2 die and 80W load that gives 2ºC temp drop. The big variable here is layer thickness; it can be even lower on a good base and high pressure.


Then there are some specific issues with BillA tests: The CPU die simulator that he have used for all his WB tests can not measure just the [block + tim] C/W; it measures [Block + tim + temp gradient to the probe location on the simulator] C/W. That's why he have a fixed 0.1 C/W added to each block own termal resistance; most of it comes fron the probe positioning, and a small part from the tim.

The cause of this is what I consider a wrong decision on the design of bill's simulator: He placed the probe some milimeters lower than the surface, and on one side. While all recomendations are to place the probe at the center and as close to the surface as possible.

Sorry if this is threadjacking, just wanted to make clear that one of the suposed advantages of direct die is not as big as the mith says.

[jaydee]

Originally posted by nicozeg


Sorry if this is threadjacking, just wanted to make clear that one of the suposed advantages of direct die is not as big as the mith says.

Seems all relevant me. Interesting stuff to boot.

[jaydee]

Duron 1.3gig arrived a few minutes ago.

DHD1300AMT1B
AHLDA 0239HPEW
94187921227

[BO(V)BZ]

On the topic of the Thermagon stuff: I tried that out on my P4 w/ IHS. I set the FSB as low as possible, as I was worried about what might go wrong if it didn't work, so my processor was running at 1.7Ghz and 1.7V. I let Prime95 run for about a half hour, and the temp settled at 67*C, aircooled using an AX478 and 80mm Tornado. Mind you, I got 60*C earlier using OCZ UltraII @ 2.8Ghz. After seeing that, I took off the heatsink, and the TIM only had one small spot that had 'smoothed' out a bit, right over the core. The rest of the TIM looked the same as when I first put it in. My thoughts on this:

My temp probe could be massively off, as it might not be hitting the needed 60*C mark to 'melt' right
I didn't give it long enough to work
The P4 heatspreader actually does its job and diffuses enough heat where it won't work well.

If anyone else has some thoughts on this, I'd like to hear them. I'm open to trying different procedures, assuming they won't wreck my P4 in the process.

BTW, it's a P4 2.26Ghz, on an Abit IT7

BO(V)BZ

[gone_fishin]

Originally posted by BO(V)BZ
On the topic of the Thermagon stuff: I tried that out on my P4 w/ IHS. I set the FSB as low as possible, as I was worried about what might go wrong if it didn't work, so my processor was running at 1.7Ghz and 1.7V. I let Prime95 run for about a half hour, and the temp settled at 67*C, aircooled using an AX478 and 80mm Tornado. Mind you, I got 60*C earlier using OCZ UltraII @ 2.8Ghz. After seeing that, I took off the heatsink, and the TIM only had one small spot that had 'smoothed' out a bit, right over the core. The rest of the TIM looked the same as when I first put it in. My thoughts on this:

My temp probe could be massively off, as it might not be hitting the needed 60*C mark to 'melt' right
I didn't give it long enough to work
The P4 heatspreader actually does its job and diffuses enough heat where it won't work well.

If anyone else has some thoughts on this, I'd like to hear them. I'm open to trying different procedures, assuming they won't wreck my P4 in the process.

BTW, it's a P4 2.26Ghz, on an Abit IT7

BO(V)BZ

I read up on it some time ago and seem to remember that you need an initial "setting" temperature or else it will not apply itself correctly. What you describe sounds like it did not get the required amount of heat applied to it to set up correctly. Try overclocking it with high voltage and the fan off for the initial mounting and then see if your temps are better when running at normal speeds. It will throttle down when the P4 gets to its max allowed temperature so the chip should not get damaged , but then again it's your money.

[Cathar]

A few points to make on the thermal paste and base-flatness.

1) The White Water and Cascade blocks that I used for testing were both lapped by me at the same time. I actually lapped each block twice. I did not use the block that was sent to BillA.

2) The raw thermal conductivity of a thermal paste is NOT the way to predict the thermal conductivity through a thermal interface. Using this conductivity assumes a 100% molecular interaction between both surfaces and all molecules of the thermal paste.

Here's an example. Copper has a thermal conductivity of ~395W/mK, far better than the ~9W/mK of AS3, so why not just lap flat a piece of copper and stick that in between the CPU and the heat sink?

The answer, of course, is spreadability. However, and I posit, that it is very wrong to assume that even AS3 will give you anything like a 100% efficient thermal contact patch. We know this to be true because AS3 "settles" over time, sometimes picking up 2-3C for a ~100W CPU over a week or so. However, is it 100% efficient by the time it's finished settling? A better question perhaps is "is the silver micronised down to the point of individual molecules?".

I still believe Bill's value of ~0.08-0.10C/W for the thermal interface layer. Yes, Bill's thermal probe was about 1mm below the surface, but the heat flows through the copper in a fairly linear gradient when talking about distances this small, and the graph I saw was plotting the gradients either side of the measurement points and extrapolating from there. There were some small anomalies, hence the "uncertainty", but to date it is the best evidence I've seen to quantify the size of the thermal barrier, unless someone has some better real testing to show.

[Les]

[i

I still believe Bill's value of ~0.08-0.10C/W for the thermal interface layer. Yes, Bill's thermal probe was about 1mm below the surface, but the heat flows through the copper in a fairly linear gradient when talking about distances this small, and the graph I saw was plotting the gradients either side of the measurement points and extrapolating from there. There were some small anomalies, hence the "uncertainty", but to date it is the best evidence I've seen to quantify the size of the thermal barrier, unless someone has some better real testing to show. [/b]

To my knowledge Bill Adam's never gave a value of 0.08- 01c/w for the thermal resistance of a TIM joint
It was agreed privately as a "possible value" until we knew more.
Then was used .

FI Probe 2mm below surface (check orig http://forums.overclockers.com.au/sh...threadid=67401

[Les]

[i

I still believe Bill's value of ~0.08-0.10C/W for the thermal interface layer. Yes, Bill's thermal probe was about 1mm below the surface, but the heat flows through the copper in a fairly linear gradient when talking about distances this small, and the graph I saw was plotting the gradients either side of the measurement points and extrapolating from there. There were some small anomalies, hence the "uncertainty", but to date it is the best evidence I've seen to quantify the size of the thermal barrier, unless someone has some better real testing to show. [/b]

To my knowledge Bill Adam's never gave a value of 0.08- 01c/w for the thermal resistance of a TIM joint
It was agreed privately as a "possible value" until we knew more.
Then was used .

FI Probe 2mm below surface (check orig http://forums.overclockers.com.au/sh...threadid=67401

[Cathar]

Seemed to me to be a decent value to be working with, agreed though, until we know more.

I guess it's a case of what's better?

1) No information at all?
2) False (marketing) information?
3) Some rudimentary testing information?
4) Some decently obtained testing information until disproven?

If we're going to apply any form of theory as to what to expect from direct-die, it would seem to me that people would want to have at least a vague idea of what they're working with/against, rather than no idea at all?

[Les]

Originally posted by Cathar
Seemed to me to be a decent value to be working with, agreed though, until we know more.

I guess it's a case of what's better?

1) No information at all?
2) False (marketing) information?
3) Some rudimentary testing information?
4) Some decently obtained testing information until disproven?

If we're going to apply any form of theory as to what to expect from direct-die, it would seem to me that people would want to have at least a vague idea of what they're working with/against, rather than no idea at all?

Information -Yes.
But unreferenced speculation or possibly misinformation - No

[Cathar]

Originally posted by Les
Information -Yes.
But unreferenced speculation or possibly misinformation - No

I guess we'll have to agree to disagree.

I guess I place a little more faith in Bill's testing and results than to call them mere "speculation" or "possible misinformation". I personally put it into the category of a "working hypothesis until disproven".