I'm impressed. Wish I'd come up with that solution, Incoherent:) (definitely a misnomer)
Looking forward to more graphs…including P/Q!

give that man a chocolate ceegar
yup

Nice work pHaestus!

Not to put a damper on your success, but to what do you attribute the variation on the CPU power (I have a fair idea, I'd just like to hear your version, and I think a lot of people would be interested).


Bill, before I wander off thinking about this graph for two or three weeks, then bring it back up again (!), is there anything else that you can share about it?

The variation in power from last night/this am's data? At very low flow rates it's difficult to perfectly insulate the hoses and I suspect that the 0.5GPM flow number is somewhat suspect for that reason. For the numbers above 2GPM the delta T is so small that better resolution than 0.01C is required (a difference of 0.01C corresponds to nearly 10W). The numbers from 0.75-2GPM are pretty much dead even at 72-74W; not much variation of CPU power with flow in that case.

If you mean the earlier test results then there was a 0.06-0.07C offset in my temperature probes that I didn't catch (but incoherent did).

As for Bill's graph, consider that we are dealing with water inside a heat pipe, and that water has some interesting properties which occur near phase transitions. Solid water floats, for instance.

Bill a question about your heatpipe:

If you go low enough in power applied do you see a flat C/W vs power region? I'd expect that you'd hit a point where there isn't enough heat to cause a phase change in the pipe. Then if you keep boosting power over the heatpipe's working range far enough does it return back to that same flat line? Where you just add so much more heat into the system that the heatpipe doesn't work because the cold side is too hot?

I am visualizing a situation where at really low and really high power (temperatures completely outside the heatpipe's operating range) that you get performance that is pretty similar to just the heatsink alone, and that only within mfgr specs is there a big improvement by adding a heat pipe.

I have never personally used a heat pipe though so I could be way off base about how they supplement convective cooling...

pHaestus identified it, a heat pipe
why the curves are the way they are is pretty obvious upon reflection
- not so obvious is the 4 days of testing (at 12 hrs/day, eh) to obtain repeatable results, and my initial disinclination to believe what I was seeing

pHaestus' results and his quandary expressed in this thread are a typical 'obstruction' encountered in this type of thermal testing, the smallest oversight can consume huge amounts of effort to identify the discrepant element
- this is not a criticism, pHaestus will do this many times more until he follows a written setup procedure - and establishes his own means of 'zeroing' the test bench prior to the collection of data
-> if the first data point does not replicate something already known there is always the potential to be heading off on a wild goose chase

take the time to ck everything out first
look at the Digitec photos, see those colored dots on the thermistor connectors and the face plates ?
(that was part of a setup control system)

Bill's exactly right; wild goose chase fairly accurately describes this thread. I feel bad about wasting lots of Les's and incoherent's time because I had a problem with my data collection. On the other hand I have learned a good lesson, and my testing is better now than it was yesterday at this time.

If you are concerned this offset in temperature probes did NOT affect earlier waterblock tests that were posted; it was an issue with waterblock outlet temperature measurement and that wasn't used in those reviews.

//edit: Another good lesson; looks like my 8K3A socket lugs are a bit dogeared and that the MCW5000-A is a good bit better than the above results would indicate when extra care is taken in mounting.

I too would guess that at extremes, a heat pipe is a lump of copper,
and a 'good' plot would show the hysteresis curves;
but really at extremes

excess local heating will cause the 'vaporising area' to dry out, and that is the end of the heat pipe until it is cooled and re-condenses

not too sure I would agree with "supplement convective cooling", while convection is one of the mechanisms utilized, a 'heat pipe solution' would utilize the heat pipe as the primary transport mechanism

But if only transport is involved then shouldn't C/W remain fixed? It has to serve as another repository for CPU heat to lower the C/W, correct? Or does the C/W have a different meaning now as you have basically "lowered the thermal resistance of copper"?

I'll jump in. Please correct me if my understanding is wrong.

C/W does not remain fixed because the transport mechanism in a heat pipe only works for a defined range of input power. This is why (in my opinion) heat pipes stink for general purpose usage. They have to be made for a specific heat load, which is fine for a manufactured system, where you match the heat pipe to a known CPU with known heat output. For consumers, you'd have to select a heat pipe for your particular CPU and then swap it out if you overclocked it differently.

Heat pipes work by depending upon a pool of liquid that evaporates from the hot end, travels to condense on the cool end, and then uses capillary action to migrate back to the hot end (repeat ad nauseum). This mechanical action of moving the phase change liquid (I doubt it is water, it is probably an alcohol or cheap refrigerant) depends upon several things.

If the hot end isn't hot enough to vaporise the liquid, then no heat pipe action occurs (no phase change).

If the hot end becomes too hot, then it "dries out", as Bill mentioned. The cool end may be able to recondense the liquid, but the liquid may re-vaporise before it reaches the hot-side.

the thermal resistance of a specific heat pipe varies with the heat flux density of the source (not the case with wbs and hsfs)

I do not yet know as a demonstrated fact, but I strongly suspect that there are several additional parameters which also influence the heat pipe's efficacy (thermal resistance), and I will be isolating and testing these in the future
- the capillary plus vapor 'transport'
- the condensation at the 'cold' end

not a DIY device in anything other than the crudest sense
Brians256 is correct

As long as you're looking for more tests to do, I'd be interested in see how the vapor pressure (and thus coolant boiling point) varies with heat-load. :)

ah Groth, gotta love it
you gonna design, build, and gift the probe to me ?

Can't afford a pressure transducer? Hmm, I guess one question that should have been asked is, are you making prototypes in-house, having them made, or are you just playing with someone else's pipe?

eh ?
the pressure transducer is not the difficulty,
it is its insertion into the heat pipe after the fact

looking only for the moment

Ah, yes, it is a bit limiting if you're only looking at complete pipes. :shrug: No playing with coolant or charge or wicking or dimensions. meh.

Don't feel bad on my account, I absolutely do not consider this a waste of my time, especially considering the amount of time (and effort, and money ...) you are spending obtaining this data for us to pick away at.
I am just happy that the errors are stable enough for them to be detected, a testament to your consistancy.

Heatpipes, and their requirement for a specific heat load. I wonder if these Shuttle SFF CPU coolers take that into account. There appear to be four seperate pipes. Could they all be tuned to different optimum heat loads? in effect creating a "wide band" heatpipe cooler...
It is a shame they are so specific. It's a pretty neat mechanism. I am very curious as to what you are working on Bill.

Cheers

Incoherent

every day is a new lesson, just chuggin' along

Yes the sensitivity and the accuracy of the probes is sorely tested by this work.
Notwithstanding,Incoherent's admirable method is a tool to determine the solution. The "0.07c systematic error" suggestion is a the value to give a best fit to an invariable W . The invariable W is questionable.
Using an offset of 0.07c is prejudging that W does not vary with Flow-rate. I don't think this can be done.

I may not be understanding correctly but pHaestus has suggested two evidential values for the systematic error - 0.06/0.07c and 0.02c.
I probably prefer the 0.02c solution which shows an increase of "Heat Absorbed" with increasing flow rate. However, if had to bet, I would choose a value between 0.06/0.07c and 0.02c as giving the true variation of W with flow-rate( this best corresponding with my preconceived notions of the variation of "Heat Absorbed(W)" with "C/W").
Some graphs:
http://www.jr001b4751.pwp.blueyonder.co.uk/pHugh3.jpg

http://www.jr001b4751.pwp.blueyonder.co.uk/pHugh4.jpg http://www.jr001b4751.pwp.blueyonder.co.uk/pHugh5.jpg http://www.jr001b4751.pwp.blueyonder.co.uk/pHugh6.jpg

Think will catch up on the Heat Pipe discussion much later. However I imagine that wb's behave similar at a high enough temp when Flow Boiling can be envisaged.

Les: 0.02C is the offset in my current setup after moving thermistors around and swapping one out. It is only valid for last night's (Swiftech MCW5000-A) data. In that test run the W varied from 69-75W but I am a bit skeptical of the numbers above 2GPM.

As for the earlier results I don't think much can be done to be conclusive other than repeat tests again. Shouldn't be a big issue though as many more blocks have to be tested and data collection is free.

I wondered about that, too. Plus, a heatpipe will have optimal temperature range. How wide? How does that range relate to coolant and charge? How much will the thermal impedance vs. heatload curves change over the temperature range? Could using two coolants with different boiling points give a wide-band cooler in a single pipe?

I think you have hit the nail on the head, as those have been the questions in the back of my mind for quite some. time. However, I have not seen any information about this in public literature. So, if the answers exist, they are either in proprietary care or they are in expensive books that are not in my library or on the internet.

I do think that using a multiple coolant heatpipe would greatly widen the effective sweet spot. I also think that even a single coolant heatpipe would probably have a wider sweet spot than we fear. For example, Bill's heatpipe data shows that it works fairly well for a ten to twenty watt range in heat input.

Les
just poking at the semantics here

"The invariable W is questionable." WRT pHaestus' testing, that W (net to wb) is variable is certain, only the rate of change is in question.

"Using an offset of 0.07c is prejudging that W does not vary with Flow-rate. I don't think this can be done." In a strictly absolute sense you are probably correct, at least with anything my setup can document. More loosely speaking however, I can and have repeatedly demonstrated that within 'reasonable' limits the device C/W is independent of the applied load (but NOT for heat pipes).

Once I have a testing 'peer' (get with it pHaestus !), I will be better able to identify the slight 'apparent' influence that changes in the applied power may induce. One needs multiple benches to verify that an 'effect' is not an artifact just of that particular bench.

pHaestus
I observe a very slight increase in the coolant temp rise (at the lowest flow rates where such is most visible, 0.01 to 0.02°C) between wbs with a high C/W as compared to others with a lower C/W. In these cases the bp is at a higher temp, which implies a different energy 'load'. Be interested to see if you see the same.

Groth
and you haven't even touched on wicking structures and fabrication methods, lol

Not my intention to suggest a device's "C/W" is changing with heat load. A device,here, is considered to be a particular wb with a particular coolant flow.Each point on the "Watts Absorbed" v "C/W" graphs are separate devices with different "C/W"s.

ah, ok
was looking at the 'set' of points, rather than individually

A practical observation:

When I remount the MCW5000-A (using one hand to hold clips onto lugs and other to work the screwdriver) the mobo's power circuitry is both very close and very hot. It's hard to imagine that at least a little of this heat doesn't make its way into the waterblock via the baseplate and top. And the amount would be pretty wb dependent (the Swiftechs are a lot closer to these parts whereas the long 4 hole wbs are instead closer to the northbridge).

and . . . . .
is this too not part of the reason for bench testing ?

are we testing the system (of which the mobo is a part), or the wb (independent of the effects of the other parts)?

perhaps I've not been paying attention, but why are you not using your heat die ?

heat die still involves sinking substantial dollars into it (need 2 good DMMs, PSU, insulation, etc etc). Time and money, time and money.

Plus I am a bit of a masochist; can't you tell from my hobbies?

pHaestus, have you seen those cones veterinarians put on dogs? How 'bout using a similar approach - blow some significant air on your voltage-regulators and northbridge to minimize the extraneous heat, but protect the CPU/block from spurious, non-reproducible air currents.

Bill,
Wicking? All the possibilities of material, fiber shape/size, weave (or not), bonding, thickness, and how all of those interact with all the other variables. It makes water look easy and friendly. When can we expect your data? ;)

PH would it be possible to drill into the blocks you have to obtain a temperature reading. This would allow you to easily insulate a small hole rather then trying to control a larger environment. I don’t know how feasible this would be on some blocks but I think it would provide a real insight to what the temperatures look like between the heat source and the transfer medium.