Current review project (MCWCHILL)

[Les]

Originally posted by KnightElite
Good work pH. 23ºC below ambient for CPU is not bad...

Is this CPU running at 2.0V with CPUburn at high priority?

The cpu is running at 1.6v and 2084MHz. ( As shown in his results http://www.procooling.com/users/phae...l/newtemps.jpg ) with CPUburn at high priority.
The graph of "Sub-ambient of cpu" v "Coolant Temp" probable reflects the change in Wattage as much as Coolant Temp

pHaestus.
The small dicrepncy between my version of the Billa data is due to my use of the raw data (as posted by Bill in http://forums.procooling.com/vbb/sho...5&pagenumber=5

Think C/W v Coolant Temp may show a real trend.
However feel that the use of Coolant temp as an indication of Wattage variation is optimistic.Feel the normalisation(terminology?) technique and then using VoltxVolt(cpu) values to obtain Watt variation maybe more appropiate:

Suggesting a Wattage range of 60%-100% for the 2200MHz cpu.
If ,and hopeful when, Flow-rates are available the use of all temps may give an insight into Wattage. Even the notorious bucket would be helpful.

[BillA]

Originally posted by pHaestus
. . . .
. . . .

goddamn its nice to see a scientist do testing
very nice job pHaestus

yes Les
but the flow rate is going to be a bear due to the viscosity changes
quite unsure how to 'calibrate' for that
pHaestus will need to quantify the pressure drop of a cross to cross straight connection (to deduct from the 'apparent' wb dp)
-> and this must be done over the actual temp range
? so how can the temp be controlled w/o the wb heat source ?
ain't gonna be easy

stay with it pHaestus

[pHaestus]

Yes Les and to further complicate things I am left with realizing this am that I left Bill's 20W data in the fit of the polynomial so the W are probably all wrong to begin with. It's a dificult problem.

Yes that's probably what the coolant temp vs subambient graph shows.

I have the digital pressure gauge plumbed in, but it is not putting out numbers I am confident in at all. More fiddling tonight, but I am loathe to completely unplumb the loop and start over. Maybe I'll give Bill a call for some insight on wtf could be wrong with it. Next up for testing purchases is a decent DPT setup with logging.

[pHaestus]

I don't see how I can get data for flow rates over the actual temperature range of the chiller tests Bill. Wouldn't that cause problems with the flowmeter I am using (GPI paddle wheel type)? I was thinking that the viscosity just represented another restriction to flow. So a room temperature pressure drop vs flow rate curve would still be relevant as long as it hit the range of pressure. If that isn't the case, wouldn't there be an eqn where I could take a room temp curve and come up with predicted curves as a function of temp?

[Les]

Originally posted by pHaestus
I don't see how I can get data for flow rates over the actual temperature range of the chiller tests Bill. Wouldn't that cause problems with the flowmeter I am using (GPI paddle wheel type)? I was thinking that the viscosity just represented another restriction to flow. So a room temperature pressure drop vs flow rate curve would still be relevant as long as it hit the range of pressure. If that isn't the case, wouldn't there be an eqn where I could take a room temp curve and come up with predicted curves as a function of temp?

Probably could get predictions, but would possibly come up with something like this:

From previous post http://forums.procooling.com/vbb/sho...?threadid=6347 ,linked to in post http://forums.procooling.com/vbb/sho...5&pagenumber=5.

These changes appears to adequately descibe behavior down to Temps a couple of Degrees above 0c.,but do not account for behaviour bellow.

[pHaestus]

Joe (K not C) has a swissflow flowmeter. Wonder how it would deal with the low temperatures? It connects to fan header on mobo and then convert to flow rate. Could calibrate THAT easily enough with the GPI.

//edit: Looks like it'd work well enough:

http://www.swissflow.com/html/800_5.htm

[BillA]

worth a try if you can't go the dp route
(sent you an e-mail)

[Les]

Originally posted by Les

.
Feel the normalisation(terminology?) technique and then using VoltxVolt(cpu) values to obtain Watt variation maybe more appropiate:

Completed sums using my prefered method.


Ugh or Significant ?

[BillA]

btsoom

but device C/Ws should be flat wrt inbut power

???

[Les]

Originally posted by unregistered
btsoom

but device C/Ws should be flat wrt inbut power

???

Yep.
For pHaestus we possibly return to your Mantra ( ....... and Watts are worse).
Your own I suspect flow-rate.

[pHaestus]

Not much to say other than "I'll just keep running tests". That curve seems to warrant another look for sure. FWIW the 50W number for a 2200MHz TBred seems reasonable; the 32W that comparing water temps with Bill yielded seemed a little low. But then again who knows what kind of power CPUBurn produces vs. AMD's "typical" numbers.

I am concerned that now we aren't even reproducing trends, and the idea that C/W gets better when flow rates markedly decrease seems a little unlikely.

[kai]

are you still using an 8k3a+?

and what bios are you using if you are , ive been having problems with it reading tbredb and barton diod wrongely

[pHaestus]

Kai: See somewhere on page 2 or 3 where I pulled the diode pins out of the socket and then soldered wires directly onto the CPU pins. Mobo reading isn't the issue in this case.

[Since87]

Originally posted by Les
Completed sums using my prefered method.


Ugh or Significant ?

Interesting.

MOSFET's (the basic transistor building block of CPU's) leak. This current leakage is a major contributor to 'idle' power consumption in modern CPU's. The leakage goes up nonlinearly as temperature increases. It may be that the 'step' seen in the "C/W" curve is a step in power consumption due to leakage current dropping dramatically near -4C.

I've been somewhat expecting to find that such a step occurred. (Although this is a more dramatic transition than I would have guessed. And, at a dissapointingly low temperature.) Most likely this step is attributable to something other than a drop in leakage current, but if it is leakage current changing, it should be possible to detect it, by looking for discontinuities in 'idle' vs 'load' measurements at around 4C.

Because leakage current is also affected by voltage, the way to test for this would be by varying the CPU frequency only. I don't want to design an experiment right now, but if you are interested in trying it pH, let me know. Of course a Grothammeter would be awfully handy.

[Groth]

Originally posted by Since87
Of course a Grothammeter would be awfully handy.

Yeah, yeah. Unfortunately I lost about 70% of my income six weeks ago - been a little low on cash and enthusiasm. I'll git it done at some point.

[pHaestus]

So you'd like to see several MHz at same voltage then? Can do, but I may have to add some more antifreeze

I was thinking a bit more about Bill's numbers and I wonder if his C/W doesn't go up at lower coolant temperatures because of increasing losses from the uninsulated hoses? That would neatly explain the effect and would point to why my W estimates from his hotside-coldside are so far off Les's predictions...

[pHaestus]

Ok crap. I had to add more antifreeze today, and when I did so I detected some air bubbles in the lines. I tapped out the air from chiller, bled it out to res over a few minutes, and started over. I was able to get 3 tests done tonight: Same voltage but range of MHz as since87 suggested. However these numbers don't seem to be matching up too well to last night's results. Some air in lines last night? Perhaps. My instinct is to just repeat expts over again and see. Anyway here are the numbers collected after purging air:



Very slight change in C/W with coolant temp this time, and consistently lower than before. I tested out chiller after finishing up and was able to get the water quite cold this time:

http://www.procooling.com/users/phae...chill/-10C.jpg

More numbers tomorrow as always.

[pHaestus]

Now as I said I will rerun some of the other points to see if there were problems before.

Les: Not trying to slight your W estimation but the spreadsheet is already set up for the way I was doing it last night. I wouldn't want to deprive you from any Excel joy anyway.

[Les]

Originally posted by Since87
....
Although this is a more dramatic transition than I would have guessed. And, at a dissapointingly low temperature....

The step if it occurs is at a Coolant temp of -4C(ish) and a CPU temp of 9(ish).

Originally posted by pHaestus
....So you'd like to see several MHz at same voltage then?....

Also most interested in variation with MHz : From June 3,2002
"Let me get my 6655 set up with 0.125C res and I will see what I can come up with regarding data. That may not help you directly though Les" http://forums.overclockers.com.au/sh...5&pagenumber=3 . The scaling with Mhz and my inadequate Laboratory Craft was what stopped me in my tracks.

Originally posted by pHaestus
......Les: Not trying to slight your W estimation but the spreadsheet is already set up for the way I was doing it last night. I wouldn't want to deprive you from any Excel joy anyway.

No slight taken .This is Exploratory Play. Happy to be allowed to play in your sandpit. . .Our starting point is the same(ish), both courtesy of Bill:- Coolant 0.83c, Wattage 81.5
Have not yet rationalised thoughts sufficiently to calculate a Wattage based on MHz .However will present(for my own pleasure):


Regrouping thoughts.

[BillA]

"I was thinking a bit more about Bill's numbers and I wonder if his C/W doesn't go up at lower coolant temperatures because of increasing losses from the uninsulated hoses?"

yes, I have always cited this as a deficiency in my tests of the MCW-CHILL

be interesting to see how this 'step' plays out
many have supercooled, none have described this
wonderful what good instrumentation can reveal

[pHaestus]

I'll try to repeat Sunday's tests to some extent tonight. First order of business though is insulating the CPU socket and wb as much as possible. I got water from condensation melt on the top of my video card last night. I figure I should check the CPU sooner rather than later.

I also in a moment of clarity thought about the drop in C/W that Les showed yesterday. When I was running CPU back and mobo backside temperatures too I noticed clearly that when under load at high Vcore the motherboard was getting pretty hot (30-35C) and affecting the CPU backside temps. Idle at the same speeds you could see the mobo backside being cooled. This could be the reason for the deviation in C/W at low coolant temps &W; the motherboard becomes far less of a secondary heat source and so the overall W (die+mobo electronics) goes down more than you'd expect. Just a guess though.

[BillA]

Harvard Thermal TAS (Thermal Analysis System) which I have, sells an add-on unit PC Trace - which I do not have
this permits a very accurate thermal modeling of the board in terms of heat inputs and dissipation
Since87 – it models with resistors so can also predict boardwide voltages, interesting stuff
I do not doubt that mobo mfgrs use this (in fact I know of several)

this 'variable' is why even a grothmeter will still fail to accurately characterize the heat load actually imparted to the cooling solution

all we can do is empirically hack away at it; and, as pHaestus and Les are doing, try to identify the source of anomalous results

. . . . and Watts are worse

[pHaestus]

Rather interesting because we end up with a really interesting (to us) set of results and testing that is of absolutely no interest to the enthusiast community at large.

**** it I'll slap an "editor's choice" gif at the end of article

[BillA]

lol
jeez pHaestus, damn near all of what we do is of no interest to anyone but ourselves
(I probably should exclude much of my 'job' work from that, but little of it is actually visible)

never fear, Swiftech will link to your article

[Joe]

So the info I need to know:

is it UV "reflective"? And how does it work the CCFLs? Also those are some nice abstract pictures and drawings in the previous posts, very post modern...

Oh and if its going to win an editors choice award, could you just remove all the text and just put pictures up with teh first one being "WE CAN JUST TELL ITS FREAKING AMAZING! - EDITORS AWARD FOR BEST COOLING BASED ON VISUAL OBSERVATIONS"


hehe