major advance in wb mounting for testing

[BillA]

JoeC has done a super job, article here

[jaydee]

I am going to think on this for a while. It seems to me a hydrolic jack might work to raise the unit/scale. Like one of the $15 1-2ton bottle jacks. Bolt that to a metal plate for stability then weld another plate ontop of it to hold the unit. Make the bolting system from the jack to the lower plate adjustable and use a level to level the top plate. Then you can use the lower plate for the support to mount the water block to keep it from raising. 4 threaded rods tacked or screwed to the bottom plate and you have something to suppor the water block mount....

I will ponder this some more.

Or instead of the bottle jack you can use one big peice of all thread with nuts for more precision mounting. The jack might be to sloppy. I think one center mounted chunk of all thread will make adjusting a lot simpler...... I got tons of the stuff at the shop.

[BillA]

no jd
alignment is the key

[maxim]

what about the WB's what use the stock socket clip? will you just rely on it to provde whatever pressure it makes?

[jaydee]

Originally posted by unregistered
no jd
alignment is the key

Yeah, I understand that and that will be one of the challenges. I will work on a concept and start a thread about it and we all can pitch in some thoughts. I will also run it by friend that is a design engineer. He is pretty talented with stuff like this.

[pHaestus]

that's pretty clever

[BillA]

big bucks here for the Zig Align ZD3

[Blackeagle]

Originally posted by unregistered


big bucks here for the Zig Align ZD3

A pretty impressive use of what he had on hand to reach his goal.

And his 8 mount average, WOW!

It will be very interesting to see how this does for him in testing of blocks.

[Groth]

Very impressive reproducibility, and the addition of the digital scales is clever advance.

I'm not sure I see the point of spending the time with the Zig Align. Is the scales perfectly parallel between corrugated top and rubber-footed bottom? Is the die-sim?

It will be interesting to see how he adapts his block hold-down for center inlets or for uneven/irregular tops.

Edit: spelling and clarity

[jaydee]

Originally posted by Groth


It will be interesting to see how he adapts his block hold-down for center inlets or for uneven/irregular tops.

I was just thinking about that while coming up with my design. Also the center inlet blocks should be offset some so the inlet is directly over the die of the CPU. That should be considered and probably measured on some blocks.

[Les]

Originally posted by Groth
...
I'm not sure I see the point of spending the time with the Zig Align. Is the scales perfectly parallel between corrugated top and rubber-footed bottom? Is the die-sim?
.....

I, also, do not see how the alignment of the base and the crossbar helps to achieve mounting forces normal to the TIM interfaces.
Yes at a constant angle but not necesarily normal.

[Les]

Bill
I note that at 1gpm(3.785lpm) the PSID of 0.57(0.4mH2O) is about double your value*(eyeballed at 0.2mH2O).
This discrepancy ,repeated in the subsequent article**, makes correlation of the C/W data somewhat difficult.
Any info/thoughts on the anomaly.

* http://thermal-management-testing.com/slitedge.htm
** http://www.overclockers.com/articles873/

[BillA]

"C/W"s are specific to a particular bench setup,
and will change also for that bench if components or procedures are changed

JoeC and I have worked together for some years, notwithstanding this collaboration I (continue to) have 'issues' with some of his equipment
- specifically his heat dies

consider the SlitEdge results:
Joe's "C/W" is considerably below that from my setup
what are the components of the "C/W" ?
°C, Watts, and the TIM joint
- I will assume that the difference between Joe's and my temps is not so large as our equipment is similar (Fluke 2180A)
- I will again assume that the effect of differences in the TIM joint is not so large
- which leaves the power characterization and the die temp

for Joe to have a lower "C/W", his Watts have to be 'greater' than mine and/or the die temp lower
possible ? - sure, even probable
1. his die has less insulation resulting in greater secondary losses
2. as I understand his setup; while the die pedestal is in contact with the wb, the whole of the heat die body is in close proximity to the wb bp and its top surface uninsulated
- but this should increase the actual heat picked up by the wb
and if so will shift the die temp down a bit
-> but it is also quite possible that where he is reading the die temp may be yielding a lower temp than with my setup ???

none of this will be too clear until Joe has posted 3 or 4 wbs that I also tested
the shame is that Joe is going to put up only the "C/W" at 1gpm

- though I too am mystified by the 'freeflow' and 'waterblock' head loss values
no idea why they should be so different than mine,
but I suspect he has not subtracted the 'connection/pressure tap' losses

[Cathar]

JoeC's heat die is 120mm^2, as opposed to Bill's 100mm^2.

As we all know, the TIM thermal resistance is proportional to the inverse of the surface area.

The TIM thereby has about 20% lower thermal resistance than Bill's setup, so there's some of the difference.

More of the difference arises in thermal density and its effects on the waterblock. The surface area presented to the block in Bill's test scenario is smaller. Now a smaller die "footprint" equates to a smaller concentration of heat that is presented to the water coming into contact with the wb. This is offset a bit by the spread of heat through the wb base-plate though, but given an assumed fixed value of h, being the rate of thermal convection, measured as W/m^2K, we are again left with a situation where a smaller die has less effective cooling "effort" acting upon it, resulting in a higher C/W for the block itself (which Bill has coined as T/W), even after discounting the TIM.

Now the thermal density effect is not going to be linear, again highly dependent upon the wb base-plate thickness, but we given the bp thickness that seems apparant on the SlitEdge, one could probably predict a 10-15% higher block T/W.

Weighting out the predicted proportion of TIM C/W and block T/W, we can predict that we're roughly looking at something a bit more than a 15% higher C/W for BillA's test-setup over JoeC's test setup for any particular wb that gets tested, again dependent upon bp thickness. Thicker bp's will incur a less significant relative "cost" for smaller die sizes.

The rest of the differences (~5%) can easily be accounted for by Bill's rationale above.

At least that's my take on the situation. Where it gets confusing for the common reader is when they see Bill's and Joe's C/W data presented on the same page, and people mistakenly assume that JoeC's naturally lower C/W figures means that certain blocks are better than those tested by BillA, as witnessed by someone sparking up this thread over at OC.com forums.

[leejsmith]

i feel like i am a little fish swimming with the big sharks but here goes.

i dont know if this is possible but you could improve on this by only using 2 hold down screws or even better one.
Like an old pulp press with one one big screw on top. this should allow even pressure to be applied in one motion instead of one at each corner.

just something that i thought after seeing the pictures at oc.com

[Les]

Originally posted by Cathar
JoeC's heat die is 120mm^2, as opposed to Bill's 100mm^2.

As we all know, the TIM thermal resistance is proportional to the inverse of the surface area.

The TIM thereby has about 20% lower thermal resistance than Bill's setup, so there's some of the difference.

More of the difference arises in thermal density and its effects on the waterblock. The surface area presented to the block in Bill's test scenario is smaller. Now a smaller die "footprint" equates to a smaller concentration of heat that is presented to the water coming into contact with the wb. This is offset a bit by the spread of heat through the wb base-plate though, but given an assumed fixed value of h, being the rate of thermal convection, measured as W/m^2K, we are again left with a situation where a smaller die has less effective cooling "effort" acting upon it, resulting in a higher C/W for the block itself (which Bill has coined as T/W), even after discounting the TIM.

Now the thermal density effect is not going to be linear, again highly dependent upon the wb base-plate thickness, but we given the bp thickness that seems apparant on the SlitEdge, one could probably predict a 10-15% higher block T/W.

Weighting out the predicted proportion of TIM C/W and block T/W, we can predict that we're roughly looking at something a bit more than a 15% higher C/W for BillA's test-setup over JoeC's test setup for any particular wb that gets tested, again dependent upon bp thickness. Thicker bp's will incur a less significant relative "cost" for smaller die sizes.

The rest of the differences (~5%) can easily be accounted for by Bill's rationale above.

At least that's my take on the situation. Where it gets confusing for the common reader is when they see Bill's and Joe's C/W data presented on the same page, and people mistakenly assume that JoeC's naturally lower C/W figures means that certain blocks are better than those tested by BillA, as witnessed by someone sparking up this thread over at OC.com forums.

"h"(Convection Coeff.,W/m^2K)changes with flow rate.
"h"'s relation with Waterloo's "Film Coefficient" is wb dependant.
Waterloo's "Film Coefficient" can be used to calculate the Spreading Resistance*
A previous post** took into account the change in "h" for a Slit-Edge model
My previously posted guess at the relationship for the Slit-Edge :-
Assumed JoeC's Die was 128sq mm
Heat Spreading calculations using Waterloo and guessed relevant Dimensions (not all given in article). Used "C/W(TIM)"=0.1 and scaled to 0.078125(1/1.28) for JoeC's(no Spreading).


* http://www.mhtl.uwaterloo.ca/tools.html#
** http://forums.procooling.com/vbb/sho...5&pagenumber=4

[Cathar]

JoeC's information said a T'Bird sized die, which is 120mm^2.

Basically Les's graphs are showing a similar sort of relationship though between JoeC's and BillA's test-die sizes, and accounts for a very large percentage of the differences in results between the two setups for the same block.

[Les]

Originally posted by unregistered
[B

1. his die has less insulation resulting in greater secondary losses
2. as I understand his setup; while the die pedestal is in contact with the wb, the whole of the heat die body is in close proximity to the wb bp and its top surface uninsulated
- but this should increase the actual heat picked up by the wb
and if so will shift the die temp down a bit
-> but it is also quite possible that where he is reading the die temp may be yielding a lower temp than with my setup ???

3. none of this will be too clear until Joe has posted 3 or 4 wbs that I also tested
the shame is that Joe is going to put up only the "C/W" at 1gpm

[/b]

1. Hopefully pHaestus's set-up will be able to monitor this with his accurate inlet/outlet temps.
2 Have not considered either of these.Further thought required on leakage from Shoulders of die to wb. However,would think, the effect of die probe position can lumped with "C/W(TIM)" as far as scaling with die-size.
3. Think I would put it stronger - data at other flow rates may be vital to rationalise the observations

[jaydee]

Originally posted by leejsmith
i feel like i am a little fish swimming with the big sharks but here goes.

i dont know if this is possible but you could improve on this by only using 2 hold down screws or even better one.
Like an old pulp press with one one big screw on top. this should allow even pressure to be applied in one motion instead of one at each corner.

just something that i thought after seeing the pictures at oc.com


[/IMG]

Damn, thats a lot like what I had in mind. Sit the scale on the base that is adjustable on 4 points for alingment, the die sim on the scale, and then lower the block from the top to get your pressure.

[bigben2k]

I gotta side with Bill here: the placement of the temp probe within the die is more than likely creating a difference: it'd be hard not to. What's missing is a compensation/adjustment factor, for actual CPU temp readings, and that's where pHaestus' excellent work comes in.


I like that idea Les. I also fail to see the relevance of the Zig-Align. The only problem is that if either surfaces (die or block) are convex, the results may vary.

My original plan was to use 4 calibrated springs, and a "normal" 4 bolt hold down. Hadn't given the socket clamp much thought, yet. What spooks me in this whole thing, is having a clamp system that flexes against the springs, thereby lowering the actual pressure applied. Now if this clamp was calibrated before hand, maybe.


Maybe it's time for an air suspension... I'll ponder this some more.

[BillA]

google on 'load cell'
flexure is unimportant so long as the alignmeit is maintained

Joe's setup will doubtless change when he gets around to testing with a 60lbf load such as is now used with several of our products

[Cathar]

The other thing to add here is WB base-plate flatness variations. Unless Joe is testing the same block as Bill did, there are bound to be some variations here. BillA's review showed that the flatness on the block that he reviewed could be improved and he makes mention of this in his review of it.

Bill has pointed out many times that block makers should all find themselves access to a proper lapping machine. I'm still hunting for one in my area that doesn't cost an arm and a leg.

[Les]

Originally posted by Les
Bill
I note that at 1gpm(3.785lpm) the PSID of 0.57(0.4mH2O) is about double your value*(eyeballed at 0.2mH2O).
This discrepancy ,repeated in the subsequent article**, makes correlation of the C/W data somewhat difficult.
Any info/thoughts on the anomaly.

* http://thermal-management-testing.com/slitedge.htm
** http://www.overclockers.com/articles873/

Noted amended article and results(unless my memory is failing)
New PSID of 0.15(0.105mH2O) at 0.96gpm compared to an eyeballed 0.28(0.2mH2O) in Billa tests:http://thermal-management-testing.com/slitedge.htm

However PSID a 0.46(0.3244mH2O) at 0.9gpm for the PolarFLO* corresponds to an eyeballed 0.28(0.2mH2O) in Billa's tests**

The picture gets fuzzier as the results come in.

* http://www.overclockers.com/articles879/
** : http://www.thermal-management-testing.com/polarflo.htm

[pHaestus]

Not too sure of the specifics of JoeC's setup, but I know that Bill measured the pressure drop due to the fittings and hose and subtracted that from all waterblock measurements. Perhaps this wasn't done by JoeC?

[Les]

Originally posted by pHaestus
Not too sure of the specifics of JoeC's setup, but I know that Bill measured the pressure drop due to the fittings and hose and subtracted that from all waterblock measurements. Perhaps this wasn't done by JoeC?

An explanation would be that JoeC has now corrected for fitting in the SlitEdge tests but not in the PolarFLO.
Seems weird in testing articles.