Current review project (MCWCHILL)

[Les]

Completing series of Graphs .
Possibly worthy of note that "Deleted 20W value" is more akin to "6th Order Extrapolation" than "2nd Order Extrapolation".

[pHaestus]

Well I am back at home and will try to get the swissflow flowmeter up and running over the next couple of days. It seems to me that there are 2 possible exmplanations for the increase of C/W at low temperatures: (a) losses from the chilled wb not being insulated that become more and more important as temperature drops and (b) lowered flow rates due to increased viscosity/resistance to flow.

It looks like you can model (b) reasonably well, and I do know that I was having problems with the coolant freezing and had to increase glycol levels as I was first setting things up. That would indicate to me that the fluid indeed is getting pretty viscous even now. Hopefully there will be no surprises with the swissflow and I can post some definitive test results and sort this out.

[BillA]

won't the swissflow get smaller with a decrease in temp ?
lol

[pHaestus]

It's rated to -20C but who really knows. It should be better than nothing I would think.

[BillA]

the Swiss are very good engrs
ask them for a temperature correction factor

[pHaestus]

No data today due to a configuration issue (I moved case around slightly to change video cards and affected the seating of waterblock). I DID test out performance vs. flow rate over a range of MHz and it looks like the flow rate does drop off but not nearly as extreme as you are modeling it Les. More like 1.24 L/min at -5C coolant and 1.17 at -9C coolant (the extremes of temp I ran). Hopefully some numbers tomorrow that:

1) Reproduce some of the earlier datasets with flow rates
2) test for a given CPU wattage as a function of flow (0.3-1L/min or so)

[Les]

Originally posted by pHaestus
... and it looks like the flow rate does drop off but not nearly as extreme as you are modeling it . ....

Originally posted by Les
After much pissing about.
The most consistent calculations do appear to be the "Billa Delta Hot-Cold" based that you are using.Whether they are meaningful is perhaps another matter.

or the "purple circles" have some credibility?

Looking forward to numbers.

[Les]

Originally posted by Les
or the "purple circles" have some credibility?

I think not.
More likely is non-equilibrium and/or Heat Gains in loop between CPU and Chiller:-

[pHaestus]

Ok I still am playing around with the data a bit but there was an anomoly that stood out:



There seems to be a flow rate range that results in coolant temperatures substantially lower than one would expect. No big deal; sweet spots are not unheard of after all. But WHY would the delta T between CPU and coolant change too? It seems like it should stay constant (as it does in the other flow rates). I am struck by the possibility that this might be related to the increased C/W at low coolant temps seen in previous data. Any ideas WHY this would happen? I can only speculate that something in the wb's performance or something in the CPU heat production (???) may be the cause the increased delta T. Puzzling.

[pHaestus]

Ok I am moving along now. One thing that was confusing to me a bit was that the estimated W (from Bill's swiftech data) increased as flow rate was decreased. I thought about this a bit and decided it could go either way.

1) this is a real effect due to extra heat (max of 10W) being dumped into water by pump throttling.

2) this is weirdness due to imperfect insulation and the actual heat input is constant (I guessed 62W for 2200MHz 1.83V TBredB).

I graphed both cases:



The data with W estimated from Bill DOES seem to suggest improved performance at really low flow rates (as Bill was advocating). Not sure if 10W extra for the throttling is reasonable though.

[pHaestus]

To further muddy the waters, here is today's data graphed on top of Bill's Swiftech numbers and my previous data:



Take that as you will.

For you Les:

http://www.procooling.com/~phaestus/...lownumbers.jpg

[KnightElite]

Interesting though that the actual Delta T from CPU to coolant increases 1ºC, when the coolant gets 2ºC cooler.... That doesn't seem to make much sense to me. But that would seem to be the ideal operating point, at least in terms of coolant temperature.

That's odd, nonetheless. Good data there pH.

[pHaestus]

yea I went back over it a couple of times and you can see that the flow data was collected "out of order" as I tried to describe the dip in temperature vs flow. I can't for the life of me figure out WHY that it would result in a higher delta T for CPU-coolant though.

[Les]

The "DeltaT(chillin - Chillout) v Flow-rate" plot suggests more heat is entering the system in the range of flow rates showing enigmatic cpu Temp rises.
Translated to Watts:-

This heat will enter through both "cpu/mb" and "the rest of coolant loop"
In steady-state would only expect heat entering through the cpu to effect DeltaT(cpu-coolant).
If steady-state I would intepret the enigma as a change in cp/mb Heat imput.

Will post more later. Now going to pub for hair of dog.

EDIT. In the pub I realised that had miscalculated "Heat Removed by Chiller". Have corrected graph

[pHaestus]

No worries Les. It's very puzzling to me (WHY it would happen) but as I mentioned it seems pretty reproducible. Your data seems to back up that it's a real additional heat source too. NO IDEA why myself.

Testing again today; trying to get results in order so I can actually write review. Looks like reviewing the chiller has become somewhat ancillary to the testing itself though (par for course).

[BillA]

my bet is that the 'peak' is very device specific,
and is principally related to viscosity and its effect on 'h'

I am not overly fond of the 'sweet spot' notion,
but can think of no other simple explanation

this premise could be easily (lol) verified
got any other 'high end' wbs pHaestus ?

EDIT: caution
need to mount with the same hardware to preclude clamping force differences
- I could send you the old 'high deck' version of the MCW5000
(note that the present one uses stronger springs in addition to the dropped deck)

[pHaestus]

More wbs now? You are going to end up with me as a lifelong unpaid MCWChill tester if I am not careful

several other decent waterblocks (what defines high end?) but none I can modify to use Swiftech socket clamp. Yet another reason why you should make a 4 hole socket 462 block with this design

I might be more inclined to try one of the old MCW372 wbs anyway; much larger design difference than just a difft model MC5000. All out of stock of those though right?

Seriously though I wasted a day of testing because some extra vigorous tapping of the lines to verify there was no air affected block mounting on Thursday.

[BillA]

let me pick through our boneyard . . . .
found one, a MCW372-B
want me to put 1/2" barbs on it and send it to you ?

as I recall the MCW5000/5002 is slightly too big to fit between the AMD studs, but not by much
- I suspect that some slight grinding of the bp/housing would provide enough relief,
then drill the P4 adaptor plate, problem is no space for the springs
- I used to make a flat plate with cutouts for the connections
(which I could loan you ??)
but then you will get into the higher variability of the 4 spring compression

ah testing, thy name is complexity

[pHaestus]

Ok I think I am done testing for the moment. Today I ran two separate tests (varying one parameter per test):

f(Vcore): where I ran the CPU at 2000MHz at 2.0, 1.9, 1.8, 1.7, 1.6, and 1.5V Core.

f(MHz): where I ran the CPU at 1.85VCore and 2111, 1757, 1407, 1231, and 1055MHz.

f(flow) was run last night with CPU at 2200MHz and 1.85V and flow rates were varied by progressively closing an inline ball valve.

Here are the comparisons:





It looks to me like the throttling of the pump results in a system that behaves a fair amount differently than adjusting CPU voltage or frequency. There also looks to be a noticable difference in behavior between voltage and frequency; guessing it's an issue of the "W" calc at the lower end of MHz.

As usual here is the data for Les:

http://www.procooling.com/users/phae...andmhzdata.jpg

[Les]

Have corrected miscalculation in "Heat Removed by Chiller". graph.


I think the "Heat Removed" is a fundamental number in the characterization of a Chiller.
Possibly the two key parameters for the performance of a Chiller are the Degrees Chill and the Watts Extracted.
Suggest that before any review of the MCWCHILL can be contemplated there has to be confidence in the Watts Extracted values.
I do not have confidence in the values which show a threefold increase in Heat Extracted.
I suspect Inequilibrium.

[BillA]

"I do not have confidence in the values which show a threefold increase in Heat Extracted."

your plot is heat vs. flow, how does it differ from this one ?



had the flow rate been reduced to 0.3lpm I suspect the same order of magnitude diffences would be seen

of course that 'efficiency peak' is still an artifact

[Les]

Originally posted by unregistered
"I do not have confidence in the values which show a threefold increase in Heat Extracted."

your plot is heat vs. flow, how does it differ from this one ?



had the flow rate been reduced to 0.3lpm I suspect the same order of magnitude diffences would be seen

of course that 'efficiency peak' is still an artifact

Energy balance.
Your radiator system is tested with a constant coolant/air differential.
Here the heat source Wattage is fixed* and any increase in "Heat Extracted" will result in a reduction in Chilling**(EDIT).
This does not happen :-


* Assuming minimal changes in the gains/losses from enviroment and pump.

EDIT ** Wrong. Will show increase.

[Les]

I do find the the latest results more sensible.The Heat Extraction increasing with a decrease in Chilling.

[Les]

Bill,
Thanks for your gentle probing and the ref to possibly comparable radiator data.
Have edited my reply where I got my logic arse about tit.
Maybe I was a little hasty in questioning the threefold increase in Heat Extraction.
Will,of course, investigate further and probably plague you with more Excels.
Les

[pHaestus]

And I can produce another data set as f(flow). I was wondering what I was going to do with myself in the absence of loud ass axial fans anyway.

edit: I still think the throttling via valve is what is making the f(flow) data screwy. I can collect fewer points for a much longer equilibrium time though just in case. Don't see why the system would take a lot longer to respond to changes in flow rate than it does in changes in CPU power though.