Hoses, WB's, Pumps, and Performance???

[BillA]

pHaestus

you're making it too complicated

use the wb heat throughput to cross-calibrate
you need only the inlet and outlet temps, and the flow rate (all quite accurately, eh)
since heat in = heat out, and assuming that the goop interface's thermal impedance is the same (???), using that Q equation from the rad article will enable the setting of 2 different systems at the same point wrt the wb

a typical CPU will have secondary losses, my heat die has close to none (HIGHLY insulated), JoeC's has a real problem due to the copper slug being exposed to the wb bp (addl heat transferred due to the close proximity radiation and convection outside of the die area)

I disagree re the inherent accuracy of a simulator
the AMD/Intel silicon chip heat dies are exact (but still do not replicate secondary path losses)
and my heat die is extremely accurate given the power supply, voltage and current measurements, and virtual elimination of secondary losses

EDIT: you and I could easily derive a correlation between 'true Watts' and any software generated 'Wattage'
-> so long as the CPU's heat generation was not limited by ANY ofher element of the couputer
(is this the case ?, I know nothing about such programs; more RAM, different chipsets, etc. change the 'heat' ??)
[I have heard they do]

[pHaestus]

Bill:

Was referring to resistors on copper blocks (since that was mentioned by Bigben2k). Should have been more clear. Your simulator and test methodology seems to be quite good: I looked at your older flow rate vs heat load table at o/cers and noticed your %efficiency was around 98%. Not sure what you were meant by %efficiency; isn't that basically a heat balance? If you are accounting for 98% or more with your numbers then your simulator is working exceptionally well.

I would assume that the memory/fsb/settings dependency will vary from application to application. It may be necessary to combine several different operations at once to really get close to "maximum" load. I was also thinking about trying K7Burn in realtime mode from DOS, as high priority in windows already heats up the processor a good big more than Prime95 or foldiing or the others. I will have 2Hz sampling with the Maxim kit and I was planning on trying all of the common "100% CPU load" programs and see how they responded. I can easily try adjusting FSB as well with same MHz (Durons are easy to unlock).

[BillA]

I guess we'll keep this going 'till we get run off

as I mentioned to Les, don't go too much by the old data
I was initially going to title the rad article "A Poor Man's Calorimeter", until JoeC told me I had to define calorimeter.
that 98% business is fine until one of the calcs comes in at 102% and it is apparent that something is not properly calibrated;
but for sure, I can always compare the applied power to the cartridge heater to that heat being transported by the coolant

this is the point I was making; if you calculate the wb Btus what you have is the CPU generated heat LESS the secondary losses
- so if I adjust my power supply to the same wb Btu/Watt level
BINGO, we now know the secondary heat losses (for that board, conditions, etc)
-- nor would it be difficult to generate a curve related to the power level
and then to check that against other mobos

did you see saba's combined burn-in program ?
quote
Although it sounds rather drastic, try perhaps installing a simple web-server program and using Intel's IOMeter program to literally continuously hammer your system. Although it wont tell you specifically whats dead or not working, I find it really does put a huge amount of stress onto the system, which is always what stability tests are all about!

If I personally wish to stress-test my system, I normally install IIS (from your windows 2000 / XP cd's) and share some largish files (DivX movies work well). I set up IOMeter for 16 continuous WWW testing threads.

Starting a full recursive defragmentation within OO Defrag 2000 (www.oo-software.de, damn good defragger!) atop this is also very good. I also run two concurrent sessions of Quake II in a window, and let them demo cycle. Finally running 3dmark over the top of all this chaos.
unquote

I dunno, not my field

be cool

[bigben2k]

unregistered (I wish you'd just register!) has a point, in more ways than one.

I had to read unregistered's comments three times, to understand it. I think (correct me if I'm wrong) that you're proposing testing a CPU load, letting the system settle, and note the water temp. Then, replace the CPU by a plate mounted resistor, and adjusting the power applied to that resistor to match the water temp, in an effort to find out the ACTUAL power applied, as calculated by P=VI.

Then, calculating the power emitted by the CPU, using some software or calculation, you'd substract the power applied to the resistor, and you'd be left with the power lost by the other components of the rig.

Pfew... Ok, but I don't think that'll work.

The problem is multiple:
1-The CPu heat comes from a very small square (i.e. the die) and so, is concentrated in the center of the bottom of the waterblock. I'd like to be able to say that it's not important, but I believe that it may be. That's why I mentionned that cooling a Pelt may be performed better by a different block design, such as a Maze 1C, since (I'm assuming) a Pelt radiates heat in a uniform pattern.

2-I think that it is clear that the power generated by the CPU is the core of all tests, but I don't believe that knowing the actual CPU's power output is important, other than to know what the minimum, and maximum power outputs are. Here's why: the CPU will generate a certain amount of power, depending on the work that it is doing: running a word processor, calculating PI to the last decimal, etc... There is an infinite range of combinations.

3- (I'll get into that tmo, gotta go!)

[ECUPirate]

Jesus Christ, these are the longest posts I've ever seen in any thread, anywhere. I was going to read them, but then I said f*** it.

[BillA]

bigben2k
nope
go to the rad article and look at the equation for Q (work done)

on a computer with wb; set 'Radiate" Watts of whatever, and calculate Q
then on my bench setup adjust power to achieve a Q, with the same wb, of the same value
since the CPU heat generated, less secondary losses, equals the 'true' bench measured Watts;
the difference will be the secondary losses

slick, eh ?
and yes, all measurements always at equilibrium (only)

ECUPirate
thanks for letting us know you're dead from the belly button in both directions
(hint: less effort with the 'Back' button)

be cool

[pHaestus]

Stumbling blocks on my end:

No laboratory chiller to control water inlet temperature. I can still get numbers but this will result in constantly changing boundary conditions. Possible to get around this?

Flow rate measurement. I have a relatively inexpensive flowmeter that is +/- 6%. This turns out to be the major stumbling block I think.

Still have to decide upon a way to assign watts. I'll most likely go with the Duron 1200MHz at 1.75V so that I can pull the typical max load number from the AMD spec and just go with that. Comments?

Good news: CPU die measurement possible with reasonably good accuracy and resolution, wb inlet and outlet temps are also I think ok (you tell me hehe).

[BillA]

"No laboratory chiller to control water inlet temperature. I can still get numbers but this will result in constantly changing boundary conditions. Possible to get around this?"

no problem so long as you can hold a temp long enough to reach equilibrium (say +/- 0.2^C, but NOT trending)

"Flow rate measurement. I have a relatively inexpensive flowmeter that is +/- 6%. This turns out to be the major stumbling block I think."

may have a soln, looking for a link

"Still have to decide upon a way to assign watts. I'll most likely go with the Duron 1200MHz at 1.75V so that I can pull the typical max load number from the AMD spec and just go with that. Comments?"

you do need to run that matrix described earlier, too much not known

you've got the coolant nailed

be cool

[pHaestus]

No concerns about getting run off (I have a set of keys). Getting better estimates of secondary heat losses (and bad numbers from calculators) turns out to be a little more interesting than testing waterblocks even (from my perspective); I will get right on this.

I can verify Prime95 heavily depends upon memory, as it will crunch all night with my Crucial DDR at 150 fsb conservative settings but crashes within a minute at "fastest" settings in bios. That is something I don't believe I have ever heard anyone point out before in relation to temperatures. RAM at CL3 means lower temps? Hrmmm.... Different stress programs and different FSB post haste! I will see if I can get i/ometer downloaded and installed as well.

Hrm a link, eh? I can just feel the cash leaving my wallet. Oh well; no turning back now (my interest level is too high).

[BillA]

HEY, goddamnit pHaestus
I was absolutly content as a hot stank, feel free to it change back

a prediction:
after jerking yourself around for several weeks you will find that this entire CPU-as-a-heat-source idea is worthless
(in terms of repeatable cross-platform results)

a suggestion:
start a list of each computer hardware variable that changes the heat output
then the same for software related items

gonna be a long list

still looking, but I'll find it

[Joe]

Mr. Adams that was me btw not pH people kept on asking who the hell you are. You can't say its entirely inaccurate

[pHaestus]

Don't look at me (I have a set of keys but I don't have a parking spot).

Can't argue with cross platform issues; your simulator is much better for that. I will eventually move to a quality die simulator. No question about it. Even with your setup though you could argue about effects that choice of die size have upon cross-platform extrapolation (that area affects the limiting baseplate/die junction, right?). Also don't know that lesser simulators (resistors on a piece of copper) are all that much better than a CPU (would be if it were simply a constant % lost, but is this the case?)

[BillA]

not gonna win an argument with the owner of the sand box
but stank was ok, intolerant perhaps more accurate

Joe
as one who also tests wbs, take a look at this thread
comments ?
it would be helpful to establish a level playing field

[Joe]

Make'n me reg on those forums.. damn you!!! I have avoided it all this time!

[BillA]

pHaestus
and here I was thinking 'at the right hand of god, et all . . '

one could say mine is a resistor as well, just with a WHOLE bunch of attention paid to the details
the die area relates to Watt density, P4s (or those other big Intel things) are, like TECs, different
how different ?, no data
but I chose 10x10mm as a middle ground for AMD

getting some bad news re the new setup
it takes FOREVER to come to equilibrium, as in 1.5 to 2 hours per change
s**t

Joe
a couple of ok posters there
and a good idea knows no source

[pHaestus]

I am very concerned with the low numbers that come from JoeC's simulator since I am getting values substantially higher with the diode reader when you take AMD's estimates on W. That is, at the moment, really lowering my interest in using a similar type simulator (and I can't afford the setup you have; even the power conditioning, monitoring, and supply is out of my price range). So I figure that I can have some fun playing with how system variables affect heat production (you know you want to see it too) and then choose a suitable suite of stresses to run when testing blocks for the current AMD platform. Modest goals...

Doesn't that mean that it ALWAYS took that long to come to equilibrium, but you can only now observe it? Resolution has more prices than just the equipment cost.

[Joe]

Bill I replied, I like the idea of a standardized test, but the real fact is that there would still be hundreds of variables that people would need to figure in when comparing a test even if ambient was the same. Although it is better than nothing

Even between die simulators, testing would still be quite different between site to site and person to person.

Also... we need a real watt app for how many watts of heat these CPU's put out at the different voltages... Radaite doesnt cut it... This is not for testing but so that readers can relate the "real watts" from a calibrated die simulator to their setup.

[BillA]

Joe

re the die simulators
the better they are made, the closer they will read
good = accurate
any simulator uses a source which are all instrumented exactly the same (I use HP and Fluke stuff), and the wb is the sink
as the level of sophistication increases, the differences will decrease; got to be
a 'problem' is the die face, the silicon chips (per the Big Boys) are the std, but out of my league

I'll defer to pHaestus re the CPU stuff, don't do it, don't know it, never have
and the more I learn (from others), the less I like it
but the CPU is what people have so thats what they are going to want to use, no argument

pHaestus

is there some software program that loads ONLY the CPU ?
correlating such would be no problem
have our cake and eat it too

"Doesn't that mean that it ALWAYS took that long to come to equilibrium"
yes, and no
along with the uniform 0.01^C resolution change I also increased the die simulator insulation and added embedded RTDs, so now I'm looking at internal insulation temp trendlines that previously were simply higher losses
the wb stabilizes at 1.5 hrs, the die at 3 hrs (but only a 0.1^C change in the last 1.5 hrs), whew
gonna have lots of time to post with the nightowls

[pHaestus]

Probably have to skip programs like Prime95, Seti, Folding@home that have an actual purpose. Those all offload info to hdd and memory regularly. AMD has an internal program that supposedly is similar to Intel's MAXPWR program that only loads processor and to 100%. I suspect that such programs will have to be tailored to the cache of the processor, and then they just send it off on some endless loop. K7Burn seems to be the closest thing to this I can find, but I don't have the resolution (for another few days) to really check that.

Just like the die simulators, this is quickly going to move away from what "feels good to end users". Any monitoring equipment should be on a separate PC; don't know too many overclockers doing this though. Probably a good many more nights of discussion and thought (not many do that either).

Embedded RTDs in the simulator? Looking at heat flux through the unit? Lots of temps needed for that...

[BillA]

"monitoring equipment should be on a separate PC"
yep, I did not think of that - got to be

will/can your reader be setup as a stand-alone unit ?

"heat flux"
mercy no, just looking at the trend of the gradient through the insulation

with the test series taking so long I suspect that the AS2 could be an influence
perhaps I should change to dielectric grease (works about as well)

[pHaestus]

The eval kit from Maxim connects to a parallel port and will happily run on a pentium class notebook for logging/monitoring.

It's kind of frustrating to think that AMD already has the answers (or at the least much more insight) to all of these sorts of questions (and many more). At least Intel makes MAXPWR available.

So I assume I will have to contend with your voice chanting "oooohhhm" the entire time I am testing out different programs, fsbs, ram amounts, and the like?

Looks like the original posters have all fled for higher ground. Was this thread even originally about testing?

[Joe]

Originally posted by unregistered
Joe

re the die simulators
the better they are made, the closer they will read
good = accurate
any simulator uses a source which are all instrumented exactly the same (I use HP and Fluke stuff), and the wb is the sink
as the level of sophistication increases, the differences will decrease; got to be
a 'problem' is the die face, the silicon chips (per the Big Boys) are the std, but out of my league

I agree, but hardly any site that CAN afford to buy a few grand in testing gear would, and all the sites can cant, wont. I mean setting a "standard" testing level to the point that its unrealistic to 90% of the sites out there to buy is a bit impractical and would do an injustice to the readers just as much as not having a standard does.

What needs to come out is a "standard" that is economical, yet effective and accurate to a tight tolerance. I mean do readers really need .01 or .001 Accuracy? Will .02 C make a HSF that much better or worse than the other one? Not saying the more accuracy is not better, but I am saying is it really needed/practical for the hardware that’s being tested on the web?

I think a "standard" is a pipe dream that in all reality will never happen between more than 2 or 3 sites on the web... For the simple reason that any time a website posts numbers that are different from the norm, instantly readers think the review is bunk and not considered correct. Hell the owner of KDComputers questioned my testing for weeks after I showed that POS CPC 370 HSF sucked! he even bough a Extech thermometer, to duplicate the test on his side ( since it was the first time I did a test with a T/C under a core like that it gave numbers far different than the oh so wonderful under the core thermistors. ) Many of the Bigger websites are afraid of that, damn afraid, and are also afraid of being found out that they have no clue what they are doing. Also the time it takes to REALLY test a handful of HSF's and blocks is not a day thing, its weeks of work, this is another aspect they don’t seem to gather.

I am just off on a rant at this point so I am going to go to bed... I would just like to remind everyone.. I freaking HATE doing reviews... by far the most thankless part of doing anything on the web.

[pHaestus]

Gotta be careful though Joe: radiate and side mounted compunurses and C/W tables all originated from this same exact line of reasoning.

I think that you let the readers come to realizations at their own speed. I remember when I saw your "CPU back" measurements complete with dremel on the socket and epoxied thermocouple that I really realized how much hotter my TBird was than I suspected. So I soon did the same thing (with thermistor since I had a spare flat one). That wasn't a popular stance (as you mentioned), but it has to feel good to have AMD correlate diode to cpu back temps in technical documents later.

There are tons of untested ideas and theories relating to water cooling that require careful precision testing to really expose. The same questions pop up often on forums relating to system design and component choice (all forums are full of them), and there are plenty of responses but not much data and often the answers are hearsay or popular opinion more than anything else. The community needs testers who are interested and have equipment up to the task of proving (or more often disproving) some ideas. Thermodynamics isn't always particularly intuitive (especially to those with a non-technical background) and even though my background is decent the application and terminology are all different and still have a substantial learning curve. The need for reproducible numbers at acceptable uncertainty levels goes far beyond reviewing blocks I guess is my point.

[Dix Dogfight]

Whooa. That took a long time to read. But it's a very interesting topic.
Here's my take on the "Do i measure all the different setups or do I make a simulated test on each type of component"

Say you want to compare 10WB, 4 pumps, 10 rads and 2 types of plumbing. That gives you 10*4*10*2=800 different combinations.
Versus running 10+4+10+2=26 different tests for the simulated test measurement setups.

I know you guys are putting alot of time and effort into these matters and I (fore one) really appreciate it. But I would not expect it from any of you to do measurements on 800 different setups.

So from this point of view the simulation approach is the only way to go.

There are of course a number of other factors that makes simulated measurements better, reproducability, better control of the variables etc.

Just my 2 cents

[Jessfm]

Its very difficult even in a lab to keep a level playing ground (I.e equilibrium) for a whole batch of testing.
But how accurate is accurate ?
You may spend so much time testing your test idea that the point is lost ?
As Bill has pointed out, secondary thermal transfers in a CPU do play a part in its cooling. Unless the die can also simulate this, then we are back to square one.
I would really like to see some form of standard, where we at least get some real temp figures from ppl who quote how well a certain component is performing.
I purposely never release any figures(temps) for my products as the only parts of the test that seems to be consistent is the amount of errors I believe I have.
I also use Fluke meters & thermocouples(K type). What I lack is a flow meter - so how relevant is any testing I do ? Only for R&D IMO.
Still my hat off to you guys, your efforts are valiant. I wish I had the time & resources to be able to do complete methodical tests of components. Can any of what you need be hired btw ? (just a thought, since purchase is highly expensive) .
Hell as a manufacturer I *should* pay for the data you guys generate - since it may help us(us as manufacturers) improve our products, or those that still read threads like this anyway
I certainly look forward to the Method.