Coolingtechnique's liquid cooling lab and reviews

[CoolingTech.]

Hello all!
First of all we would like to say thanks to Vapor and to XS admins to allow us to make this thread, we really appreciate it.
We are writing here because we followed this forum for years especially with regard to the reviews performed by martinm210 and then by Vapor and Skinnee, but also for other tests conducted by some manufacturers, such as those of gabe (Swiftech).
We are an Italian group that deals exclusively with the cooling for personal computers born two years ago; during these two years the tests carried out have led us to the use of a strict testing methodology and appropriate measuring instruments and we would like to share some of our efforts and investement also to the international users that have our same passion.

After a year of testing, hardware-based, joint with the help of some members of our forum, we landed on a systems that allows us to bring out more objective data thanks to the live controls that allow us to know in every moments the exactly watts that we are giving to the product;no more unstable cores, strange thermal curve and variables, just the pure heat.
The laboratory actually can count on different instruments:
-conditioned test room
-2x DC stabilized power supply
-Minebea,yateloon and scythe ultra kaze fan series
-Digital differential pressure gauge
-Two 4 way digital precison thermometer(k thermocouple)
-Digital laser rpm reader
-Fan blade reflector
-Sanso PD31 pump
-Dwyer flow meter
-professional voltmeter
-security watt reader
-High end Intel i7 based computer for data logging(liquid cooled off course!)
-Ek Flow trap
-flash,trigger,lens and a lot photo stuff
-Thermal charge system
This last one is based on a load copper block power up with a maximum power(upgradeable) of 400W with a thermal live log that give to us the product's thermal curve.
http://www.abload.de/thumb/base_simulatore3ps5.jpeg (http://www.abload.de/image.php?img=base_simulatore3ps5.jpeg) http://www.abload.de/thumb/black_boxvo3a.jpeg (http://www.abload.de/image.php?img=black_boxvo3a.jpeg) http://www.abload.de/thumb/simulatore_di_carico_ysgl.jpeg (http://www.abload.de/image.php?img=simulatore_di_carico_ysgl.jpeg)

Here you can find the whole article:
http://www.coolingtechnique.com/guide/metodologie-di-test/570-raffreddamento-a-liquido-simulatore-di-carico.html (top left is a tool for translating the article into English).

But what about the lab?
http://www.abload.de/img/2011_01_12_924865kv1rbh.jpg

In these following links you can find all the test methodologies; they are quite long to be fully reported, so for the users that are interested into the "how to", they can be find all the instructions here:
Performance test:
http://www.coolingtechnique.com/guide/metodologie-di-test/569-raffreddamento-a-liquido-test-prestazionali.html

Pressure test:
http://www.abload.de/img/wb_con_manometrocttj.jpg
http://www.coolingtechnique.com/guide/metodologie-di-test/555-test-di-portata.html

The Flow test closed loop is quite simple, and pressure readings are guaranteed by a gauge PCE-P30 (http://www.coolingtechnique.com/guide/metodologie-di-test/555-test-di-portata.html) connected to the computer for reading averages.

http://www.abload.de/img/software_manometroupyn.png

The loops on the test performance of radiators or water block are very similar.
The reading of the temperature of air, water and the simulator is performed by two thermometers VOLTCRAFT K-204 with K probes connected to the PC for the average of test results.

http://www.abload.de/img/termometri_k2043t1c.jpghttp://www.abload.de/img/sonde_acquaftlj.jpg

http://www.abload.de/img/screen_termometromg1e.jpg

The fans and pump(Sanso PD31) are connected to two DC stabilized power supplies displaying voltage and amperage of the connected components; the fans rpm are provided by a laser digital rpm reader thanks to a reflector phisically put on a fan blade.

http://www.abload.de/img/misuratore_rpm3cuu.jpg

The section on test data is collected from Snowfortrick, while the mathematics is curated by AndreaBZ that for surelly you already know for the development of the liquid and air cooling spreadsheets.

The liquid cooling laboratory was definitely started four months ago, we are developping some other instruments,that will coming during 2011, to make everything more accurate as possibile thanks also to our air cooling lab that will see in the next months a complete reconstructon with wind gallery and thermal climate chambers.
all the spreadsheet are available for free in our download section or following the thread already into XS posted by AndreaBZ:
http://www.xtremesystems.org/forums/showthread.php?t=206542&highlight=liquid+cooling+spreadsheet
with these spreadsheet you can simulate every change in you system interpolating the data coming from our lab and the products tested by Martin.
---------------------------------------------

[CoolingTech.]

Here (http://www.coolingtechnique.com/recensioni/76-waterblock/687-recensione-ek-supreme-hf.html) you can find an example of a review about the EK Supreme HF, we will keep this thread updated with the next reviews, sharing the data that we will obtain from the products.
EK Supreme HF:
http://www.abload.de/thumb/componenti_ek_supreme_xzxw.jpg (http://www.abload.de/image.php?img=componenti_ek_supreme_xzxw.jpg) http://www.abload.de/thumb/panoramica_finale_ek_sybbz.jpg (http://www.abload.de/image.php?img=panoramica_finale_ek_sybbz.jpg) http://www.abload.de/thumb/specchio_base_ek_suprepbpn.jpg (http://www.abload.de/image.php?img=specchio_base_ek_suprepbpn.jpg)
http://www.abload.de/thumb/base_ek_supreme_hf2xda.jpg (http://www.abload.de/image.php?img=base_ek_supreme_hf2xda.jpg) http://www.abload.de/thumb/interno_top_ek_suprememxj5.jpg (http://www.abload.de/image.php?img=interno_top_ek_suprememxj5.jpg)
Data:
http://www.abload.de/thumb/comparazione_portata_eoh54.jpg (http://www.abload.de/image.php?img=comparazione_portata_eoh54.jpg) http://www.abload.de/thumb/curva_prestazionale_jpkii5.jpg (http://www.abload.de/image.php?img=curva_prestazionale_jpkii5.jpg) http://www.abload.de/thumb/portata_ek_supreme_hf_zc4g.jpg (http://www.abload.de/image.php?img=portata_ek_supreme_hf_zc4g.jpg)
http://www.abload.de/thumb/prestazioni_laing_top_sx0u.jpg (http://www.abload.de/image.php?img=prestazioni_laing_top_sx0u.jpg)

Actually on our website you can find also the aquacomputer kryos XT
http://www.abload.de/thumb/base_interna_kryos_xttzwn.jpg (http://www.abload.de/image.php?img=base_interna_kryos_xttzwn.jpg) http://www.abload.de/thumb/dettaglio_foro_duscitakyum.jpg (http://www.abload.de/image.php?img=dettaglio_foro_duscitakyum.jpg) http://www.abload.de/thumb/presentazione_kryos_xtpla5.jpg (http://www.abload.de/image.php?img=presentazione_kryos_xtpla5.jpg)
http://www.coolingtechnique.com/recensioni/76-waterblock/670-recensione-aquacomputer-cuplex-kryos-xt.html

The Phobya CPU waterblock:
http://www.abload.de/thumb/base_lato_cpu_phobya_c1al4.jpg (http://www.abload.de/image.php?img=base_lato_cpu_phobya_c1al4.jpg) http://www.abload.de/thumb/componenti_phobya_cpu_jxzn.jpg (http://www.abload.de/image.php?img=componenti_phobya_cpu_jxzn.jpg) http://www.abload.de/thumb/interno_base_phobya_cpqxl2.jpg (http://www.abload.de/image.php?img=interno_base_phobya_cpqxl2.jpg)
http://www.coolingtechnique.com/recensioni/76-waterblock/722-recensione-phobya-cpu-waterblock.html

The Dimastech Nethuns
http://www.abload.de/thumb/confronto_top_vista_surlxv.jpg (http://www.abload.de/image.php?img=confronto_top_vista_surlxv.jpg) http://www.abload.de/thumb/nethuns_montato__frontrafn.jpg (http://www.abload.de/image.php?img=nethuns_montato__frontrafn.jpg) http://www.abload.de/thumb/panoramica_base_nethunhz5v.jpg (http://www.abload.de/image.php?img=panoramica_base_nethunhz5v.jpg)
http://www.coolingtechnique.com/recensioni/76-waterblock/657-recensione-dimastech-nethuns.html
and some others wbs.

---- ---- ---- ----
Just a little note concerning our work and this thread, the objective of this discussion is intent to share in another continent and with different persons the work that we do spending a lot of moneys for the passion that unites us; obviously we are open to any type of cooperation,criticism and suggestions with really no problems.
Thanks for reading an stay tuned for the next release.
CT.com Staff.

[PatRaceTin]

interesting

[Eddy_EK]

Hi,

This does not simulate real CPU cooling.
The surface of the heat flux just too big, as the CPU heat density is much higher.
http://i.neoseeker.com/neo_image/162213/article/nehalem_core_i7_review/Core_i7_txt72_thumb.jpg
The heat should be going through the surface as big as the core is and then covered with a real CPU heat spreader to simulate real CPU thermal resistance and cooling.

[Alexandr0s]

Good to see there's another tester on this forum. More independent tests are always great :D.

[CoolingTech.]

Hi,

This does not simulate real CPU cooling.
The surface of the heat flux just too big, as the CPU heat density is much higher.
The heat should be going through the surface as big as the core is and then covered with a real CPU heat spreader to simulate real CPU thermal resistance and cooling.

Hello Eddy, nice to see you also here.
Indeed our main objective isn't make a pure simulating of a cpu(why simulate a cpu if we can use a real one?)
the main object is to standardize all the testing procedure removing all variables from the testing section.
We made thousand test using a 775 cpu and 1366 i7 cpu; the main problem is that the "full load" of the cores is really unstable changing every second the load wattages and having different results every times(please note that i'm not talking about the dts slope or precision); this mean a lot of small variations that, if added, should be to affect the final result.
I'll post some data that we collected in this way with an high precision pt100 probe(deltaohm (http://www.deltaohm.com/ver2008/uk/HD2108.htm) thermometer) under the cpu and another probe inside the circuit to understand if we have the same heat released into water(same cpu/main/ram/pump etc...).
following some pics from the testing sections:
http://www.abload.de/thumb/dsc045822ysk.jpg (http://www.abload.de/image.php?img=dsc045822ysk.jpg)
http://www.abload.de/thumb/dsc04594ta2b.jpg (http://www.abload.de/image.php?img=dsc04594ta2b.jpg)http://www.abload.de/thumb/dsc04595v9af.jpg (http://www.abload.de/image.php?img=dsc04595v9af.jpg)
The result was that every time we have different heat into the system, you can test it also with your own instruments.
I personally also make a different versions of the load block and one have, off course, a similar solution on what you are saying with two core and an original 775 ihs on the top.
In this case the main problem was the same ihs , indeed the thermal compound that we put beetween the block and the ihs(to remove the contact problems beetween the two materials) was subject to degradation and the data change beetween every thermal test; the solution was soldering the ihs to the block but in this case we haven't again a perfect cpu simulation because, of course, welds have different thermal propriety compared to a thermal compond and the ihs was subject to change its flatness due to the many mont/umnount.
The result was to set the entirely methodology to an unique block that for sure don't will change its characteristics allowing us to compare data years later.
Please also note that the performance of the wbs aren't taken directly by the temperature of the load block, its function is merely to give heat to the wb, but is obtained after interpolating the data that we obtain from all the sytems using mathematical equations that give us the behaviour thermal curve of the product.
With a thermal camera you can also see how an ihs displace the heat(ref. pictures).
http://www.coolingtechnique.com/img/guide/metodologie_test/raffreddamento_a_liquido_-_simulatore_di_carico/Foto_termica_CPU_1.jpghttp://www.coolingtechnique.com/img/guide/metodologie_test/raffreddamento_a_liquido_-_simulatore_di_carico/Foto_termica_CPU_2.jpg

I'm free to talk with you into msn or skype and sharing some private data if you nedd it.
Cheers from Italy.

[Martinm210]

Awesome, glad you guys are up and running, looks great!!

[Church]

CoolingTech: it's just that with artificial heat source with differences from real cpu noted by Eddy (different core size/IHS) some blocks might extra gain or loose because of them being more or less "focused" on cooling core only area. And that's why block cooling results might difffer from ones users will get on real hardware. +Jumping temps of different cores under full load imho can be simply worked around by averaging during longer testing time. I advise also to take another weapon from skinnee's stash :) - do several mounts per each block, to eliminate errors introduced by misapplied TIM paste, eliminate best and worst results, average the rest. Averaging is powerful weapon if used wisely.

[CoolingTech.]

Awesome, glad you guys are up and running, looks great!!
thank you Martin, if we are here is also thanks to you and your data;)


CoolingTech: it's just that with artificial heat source with differences from real cpu noted by Eddy (different core size/IHS) some blocks might extra gain or loose because of them being more or less "focused" on cooling core only area. And that's why block cooling results might difffer from ones users will get on real hardware. +Jumping temps of different cores under full load imho can be simply worked around by averaging during longer testing time. I advise also to take another weapon from skinnee's stash :) - do several mounts per each block, to eliminate errors introduced by misapplied TIM paste, eliminate best and worst results, average the rest. Averaging is powerful weapon if used wisely.

Sorry for the delay; the ihs/dimension of the top block surface is the same of an intel i7 but this isn't the point.
The main point is that the cpu change every time the watts that the core/s put in water, this is due to the "full load program" and background programs that affects the cpu working making different results.
Also if you change systems you will have again more differents values;obviously you can made the values averages but you still have errors and if you use dts you can have potentially some large margins of error(according to intel dts shall be used just to have an indicative value).
What we are doing is to put watts into water through the wb(we test with 300W), this allow us to obtain the thermal curve of the products that can be applied in every situation(reffering to the spreadsheet) and the study of the equation allow us and you to optain the temperature also at 20/40/100/200 watts.
I repeat, our purpose isn't simulate a cpu but is discover the thermal curve; with that we know the thermal dissipating capacity of the product and we can obtain the value with every watts load(knowing also you rad,fans and tamb); simply physics and mathematics applied to the thermodynamics.
If you obtain the thermal curve on a cpu you will have the thermal curve on that specif cpu, if you change the cpu( with another with same tdp) you surelly will have different results(differente flatness,core,thermal paste etc...) so why made the average just on one cpu? you should made the average on 2-5-10-20-50 cpus.
We just standardize all the variables:)

[prava]

The main point is that the cpu change every time the watts that the core/s put in water, this is due to the "full load program" and background programs that affects the cpu working making different results.
Also if you change systems you will have again more differents values;obviously you can made the values averages but you still have errors and if you use dts you can have potentially some large margins of error(according to intel dts shall be used just to have an indicative value).
What we are doing is to put watts into water through the wb(we test with 300W), this allow us to obtain the thermal curve of the products that can be applied in every situation(reffering to the spreadsheet) and the study of the equation allow us and you to optain the temperature also at 20/40/100/200 watts.
I repeat, our purpose isn't simulate a cpu but is discover the thermal curve; with that we know the thermal dissipating capacity of the product and we can obtain the value with every watts load(knowing also you rad,fans and tamb); simply physics and mathematics applied to the thermodynamics.
If you obtain the thermal curve on a cpu you will have the thermal curve on that specif cpu, if you change the cpu( with another with same tdp) you surelly will have different results(differente flatness,core,thermal paste etc...) so why made the average just on one cpu? you should made the average on 2-5-10-20-50 cpus.
We just standardize all the variables:)

Mate, you can't analyse cpu blocks without using a cpu, period. Yeah, I know its a PITA to make it with a cpu because you need to do it a ton of times in order to bring statistics so you can have some sort of logic results.

BUT, if you do like you say, you are modifying greatly the heat source and, thus, biasing your review, because you are no longer using a cpu and thus the results will never be useful because they will differ with a normal cpu.

:down::down: Get a cpu, and find a program that loads the cpu in a way you load. Besides, its stupid to think that a cpu will get loaded evenly when it has discrete parts that get loaded.

Otherwise you are a biased review because you don't use a cpu when your studying how to dissipate its heat.

[CoolingTech.]

ok, lets change how i'm trying to explain the methodology.
I woul dlike to ask you a question:
what change when a wb put in water 100W from a cpu and 100W from another source?

[Church]

CoolingTech: IF another source is of same area and topped with similar IHS, probably almost nothing. To be on the safe side i'd cross-check with few of different cooler or waterblock models, to see if delta of cooling performance is same be it on artificial heat source or real cpu.
As for different loading of cores you were speaking earlier - i'm guessing it was about turbo dynamic clocking? .. my two cents would be: 1) liquid cooling often comes hand in hand with overclocking - imho it's worth to test not with stock cpu putting out eg. 100W, but with o/c-ed && overvolted one (which might rise heat dump even up to 250-300 depending on clocks/voltage), with side-effect being bigger differences between blocks/easier to measure ; 2) often higher overclocks involve disabling dynamic turbo mode, energy saving or keeping within TDP features, all of that should even out core loading/heat dump.
EDIT
Oh, almost forgot about another feat that one should simulate for artificial heat source too. - Mounting system & mounting pressure. These days when almost all top waterblocks are 0.5-2C from each other, it's mounting pressure of particular mounting sys that often might divide winner from looser. One also has to incorporate into that artificial heat source some pseudo socket, or otherwise how can you tell if mounting pressure is same as to what block's mounting system provides on real cpu/real motherboard, and without it you can have very different winners in tests/roundups.
Hmm, but if it starts to get that much overcomplified .. imho real cpu testbed is much simpler to use then.

[prava]

ok, lets change how i'm trying to explain the methodology.
I woul dlike to ask you a question:
what change when a wb put in water 100W from a cpu and 100W from another source?

If we talk about how to dissipate that heat, aka rads, no problem whatsoever, but if we talk about cpu there is a TON of difference.

You said on your own that heat spread is not even in a cpu...which means that blocks will perform different on your model thatn in real world, which means your model would be useful for even-spread heat but not for uneven heat-spread, such as cpus....which are your target of study.

So, its not about XXXX heat, its about the difference between how it is distributed in a cpu and in your model. They NEED to be the same...and they are not.


PS: also, you are factoring out the fact that a cpu is not flat nor a perfect straight surface.

PSS: also, you need to make the more remounts the better. You need to do the same test like 5-6 times AT LEAST in order to have some sort of consistency.

[Waterlogged]

We've had ppl use die simulators in the past and they've all stopped testing because they been harassed to no end about how flawed the testing method is. I see history repeating itself yet again. :rolleyes:

[Alexandr0s]

So I guess we need a comparison test on die simulators vs. real CPU's?

[nvmc]

Bottom line is that you can't get a consistent level of heat output from a real chip. While OP may not be an exact representation of a real world situation, it's a good objective measure of comparative performance. Which is more than can be said about the rest of the testers out there, especially when the apparent difference in performance between blocks often falls within the margin of error for processor heat output fluctuation.

[Church]

Waterlogged: i understand his wish for very repeatable/consistent heat source for tests. But if results from such test bed differs from how blocks perform/compare in real life, is it worth to do this way? And different die area/different mounting pressure CAN screw up results. Remember Kryos XT review? With just different thickness O-ring it was matter of loosing or winning over HF. So will these two blocks perform same on this simulated die as on real CPU, if mounting (and pressure) differs between testbeds? If he manages to do that, hats off & congratulations. If not, imho wasted effort.

[prava]

Waterlogged: i understand his wish for very repeatable/consistent heat source for tests. But if results from such test bed differs from how blocks perform/compare in real life, is it worth to do this way? And different die area/different mounting pressure CAN screw up results. Remember Kryos XT review? With just different thickness O-ring it was matter of loosing or winning over HF. So will these two blocks perform same on this simulated die as on real CPU, if mounting (and pressure) differs between testbeds? If he manages to do that, hats off & congratulations. If not, imho wasted effort.

The problem is that to do it properly one must do TWICE the effort: one needs to calculate stuff with the artificial test bed and THEN with the real cpu in order to cross the results...and as every block is different it should be done on each block.

No, really, one has to use a real cpu.

PS: with that said, It would be interesting if Skinee/vapor checked one of those load-creators under their system to see what kind of off-ness you get.

[Mazlov]

We've had ppl use die simulators in the past and they've all stopped testing because they been harassed to no end about how flawed the testing method is. I see history repeating itself yet again. :rolleyes:

Alternatively, we could discard any and all tests that have been done on first gen Core i or AM3 cpu's. They're of no use for the coming year anyway!

Seriously, I believe having standardized testing methods will only add to the mix. Have some test on real cpu's and others on die sims. We can oly benefit from the total amount of results.

[Church]

prava: well, imho they can do cross-testing only once per time when new cpu-s with different core size and different socket come out. After measuring difference with few different enough waterblocks or coolers (if mounting system is similar enough to real cpu-s), one can later reuse it with whatever other waterblocks/coolers without cross-testing anymore. Of course, if artificial testbed results differ not just in relative temps, but also in "preferability" of particular types of waterblocks or coolers over different types and make their results arrangement by performance different from arrangement on real cpu-s, then that's it, game over. It will result in yet another synthetic test not usable for actual component choice, or as i wrote in previous post - wasted effort.

[Biffa]

OK so most of the cutting edge blocks are designed to absorb and dissipate heat from a small central core under a wider heatspreader. I think the fact that a CPU fluctuates its heat output isn't really relevant to testing a bunch of blocks to see how well they dissipate heat.

So in theory its a good concept to use a controlled heat source, but a big block like that doesn't radiate heat to the block in the same position as a real CPU would, as Gabe said.

So these guys, they just need to change the "artificial" heat source to resemble a real CPU, that way they get the physical characteristics right and they get the constant heat load they are after.

Just make the contact point the same size as a real CPU and put a heat spreader on it.

[Eddy_EK]

Why the simulator?
You said it yourself, the heat dissipating from the CPU core is not stable.
And you never know what is the real wattage of the CPU when fully loaded.

That is why you need to make it like I explained in my first post.
You will have real CPU simulator.

With this testing your data of thermal performance are not representative for CPU cooling.
To cool your block of copper, totally different design of the block that is are currently on the market would be ideal.

[CoolingTech.]

Never think that the test will have attract all this attention:)
However lets give some answers:
@churchy: the block is already studied to fit into an i7 processor(socket 1366) indeed we test each wb with its retention system; we just change the M3 screws with longer ones and yes the data that we collect are already crossed with the data that we collected at the beginning on an intel 920.
If you download the spreadsheet you can see the equations that we are using.
http://www.coolingtechnique.com/download/doc_details/29-water-cooling-spreadsheet-v061-2010-by-andrea.html

@prava the data are already taken 5 times(5 mount/unmount).

@Eddy right, two years ago we made a load block similar on what are you saying, the temperature was 2,7°c higher than the block that we are actually using but the thermal curve of the product was the same or a little bit different(aka same performance); for this reason we decided to give power to an unique block.
Unfortunately i don't have any pics here(probably they are inside an external hdd, i nedd to check them on my free time) however i've found the th on our forum that talk about this project with some pictures(some are just for reference):
http://www.coolingtechnique.com/forum/showthread.php?tid=275&highlight=what

Now, if you guys would like to have a comparison beetween a real cpu, actual block and another block with the same die dimension of an i7( if i'm remember correctly is 263mm2) and a real ihs on top, I can personally made it and make some test; just give me a couple of weeks to take some electrolytic copper for the work.:)

[Alexandr0s]

But if results from such test bed differs from how blocks perform/compare in real life, is it worth to do this way?

Do they? That's the problem. I know there is a difference in heat dissipation between those dies and CPU's, but I'd really like to know how much of a difference there is. Hence the suggestion to compare die simulators to real CPU's.

My suggestion: use a non-overclocked CPU, so the wattage should be close to the given TDP. Run CPU blocks on both and test according to Skinnee-Vapor testing (five mounts, ignore worst and best results, average the remaining three). If there is a large difference between the two, try tinkering with the wattage of the die.

I know this is a pain to do, but I feel like no one will fully buy any results since they're 'not according to a real life situation'. One other solution would be to just switch to using a CPU instead of the die...

[SpuTnicK]

FIY rads should be tested with heat provided by real CPUs not some water heaters since they are "not according to a real life situation".

The main idea is not to come as close as possible to a real CPU operation (no-equal heat generation etc.) but to use a constant heat source, i.e. to creates equal test conditions, to test different waterblocks, like Skinnee does with rads. The only thing that should be done/checked thoroughly is the size of the contacting base of the heat loader to math the one of CPU IHS and the pressure of the mounting system.

just my http://www.overclock.net/images/smilies/2cents.gif

[Zio Sp_Hammer]

here I m ! ... I waiting for other results ...

[NaeKuh]

So I guess we need a comparison test on die simulators vs. real CPU's?

was done.. and didnt show great.

Its because of the focus cooling jets we use in specific spots which also correlate with where the DTS is located inside the die.

Using a uniform heat plate vs, one where specific heat spots are is not an accurate test, because the entire IHS does not heat up.


RAD Testing bud... thats where u use dedicated heaters... not cpu.

[theseeker]

Thanks for the reviews and we can all benefit from another third party testing source. The one thing that I find strange is that every review I have read basically comes to the same conclusion, the Kryos XT comes out on top just like CoolTech's results.
I have a few engineers that work for me and they all have different opinions when it comes to testing methods. I believe it best to test differently in order to simply look at the final results and compare the final data. Typically, the final results all point to the same conclusion. There is never a perfect test method and like I say to the engineers on occassion, are we picking fly sh*t out of pepper?
Just my 2 cents.

[SpuTnicK]

RAD Testing bud... thats where u use dedicated heaters... not cpu.

dedicated for what?, watercooling?. yeah right:rolleyes:

[penguins]

which CPU should he use to test it?

a 775 dual core? Quad ? an i3 i5 i7? am3 ? am2? all of them? Phenom x2 x4 x6...

you can't make a test for all those cpus with just an intel. SO it isn't 100% Accurate for everyone's cpu... Neither is a test with any one of those procs Vs. the other.

should he be doing 5 mounts per every processor that the blocks are compatible with?? cmon no one else does that. Yet we use the info from Skinnee's results (i7) and don't tell him Why didn't you test an amd! or a core 2 duo! I want to know if the XT is better for that, or the supreme HF.

use the same generality for this eh?

[relttem]

I like the tests. but, I didn't not know waterblocks were CPU design specific..eg..this water block is only for an AMD phenom II black edition. I just thought they were socket specific..more or less. Still, if I am testing waterblocks with a heated plate and find that one works better than the other, I would bet money that it will work better on a CPU too. The nice thing about a heated copper block is that the heat in can be controlled more accurately than can be with a CPU running bench marks. But, Naekuh's point about a CPU not heating uniformly is valid in terms of the block design. One block might be poorly designed for a certain spot of the specific CPU that happens to get hot during a benchmark run, which would make that block perform poorly on that CPU. Still, I like the tests.

rock on

[NaeKuh]

dedicated for what?, watercooling?. yeah right:rolleyes:

Dedicated inline heaters are used for radiator testing.

And i would know, because thats how they been tested.

Either a in sump or a inline heater are used to generate the exact wattage for rad testing deltas.

[Waterlogged]

Dedicated inline heaters are used for radiator testing.

And i would know, because thats how they been tested.

Either a in sump or a inline heater are used to generate the exact wattage for rad testing deltas.

I prefer sump, like that you know were all the heat is going.

[SpuTnicK]

doh! That copper heater of CT generates the exact wattage for CPU waterblock testing. So i see no major problem with that. :)

[NaeKuh]

doh! That copper heater of CT generates the exact wattage for CPU waterblock testing. So i see no major problem with that. :)

not with variable wattages.

dude were back in first base.

if you dont understand why a variable wattage cpu =! a set wattage load tester, then you obviously missed the past 5 yrs we shifted in the hobby.

We dropped the constant load block back during the Apogee GT days people.
Because EK Supreme, followed by Watercool's HK showed us, this is not the correct way, from there focused jets making a better performance impact in Real Life Applications.

Dude we even got gabe to change his testing methodology when he used a constant load tester.

[SpuTnicK]

dude, check my join year, i could not possibly follow you for the past 5 years.;)

Because EK Supreme, followed by Watercool's HK showed us, this is not the correct way
link me to this interesting thread. i must read it.

BTW how old is jet plates in CPU waterblocks, one or two years?
And how old is Watercooling, maybe three? (jk)

How can you compare two objects in different testing conditions (variable cpu wattage) and then make constructions that one is better than another? Those results are useless.
And, convincing people to change his mind does not always means that he was wrong in the first place.

[BillA]

And the cycle repeats, a new crew (like all such) that cannot dredge up and learn from what has gone before; so sad.
Italy is a swell place (more or less grew up in Firenze and Napoli), and not so far from the rest of Europe that French, German, and Dutch testing should be a mystery.
What on earth is the need to repeat the most basic cpu heat dissipation testing ERRORS that have been cussed and discussed infinitum ad nausium?
I am old and out of it, but the magnitude of the sloppy thinking displayed by all parties causes me to react.

1) Damned near anything can be rigged to measure something, and data can be collected and massaged to the number of decimal places desired - and so what?
What is the point in testing?
And what is the intended use of the test results?
-> If the goal is to gin out some numbers for component comparisons, get it on. Just what is done and how it is done doesn't matter much at all, as long as the "results" are reasonably repeatable - show error bars, eh? And do NOT use defined units whose derivation cannot be supported (the use of "Watts" is the clearest example).

2) A bare copper slug is shown as the heat source, and I stand in awe at how uninformed the responsible test engineer is of even the basics of thermo.
If the "test results" are to mean anything at all, if the tester is going to use a defined unit like "Watts", then just how has the heat input to the sink been quantified? It is the most simple thing to measure the heat energy input to the slug (with certain assumptions), what the hell does that have to do with the heat dissipated by the sink?
-> What are the secondary losses? And under which conditions? (The answer is a rather large number of variables/curves whose combined value is a single number ONLY under a single combination of the affecting conditions.)

3) How about the secondary heat losses in the rest of the system, including the sensors, (the primitive) flow meter, and ultimate sink? Again another slew of curves combining differently at each data point. Not saying this has not been done, but I will opine that that the results from the equipment shown are not accurate at all. Been there, done that.

3) Anything can be correlated (in theory) to anything else. But I can assure all, from years of hands-on experience, that correlating the heat flux from a slug to a cpu will be the final test project that an individual undertakes. They will be pensioned off long before they get it right.
-> There is a good reason the chip mfgrs use a die simulator with an IHS; for the quantification of test results, and the comparison across platforms.
Sorry guys, you can't get there from here.

This is not a debate, just some comments intended to provoke some critical thought by the tester and the readers.

Be Cool, BillA

[Martinm210]

Thanks for stopping in Bill...have always admired and learned sooo much from your past work...very much appreciated!!

[Waterlogged]

And the cycle repeats, a new crew (like all such) that cannot dredge up and learn from what has gone before; so sad.
Italy is a swell place (more or less grew up in Firenze and Napoli), and not so far from the rest of Europe that French, German, and Dutch testing should be a mystery.
What on earth is the need to repeat the most basic cpu heat dissipation testing ERRORS that have been cussed and discussed infinitum ad nausium?
I am old and out of it, but the magnitude of the sloppy (by all parties') thinking displayed causes me to react.

1) Damned near anything can be rigged to measure something, and data can be collected and massaged to the number of decimal places desired - and so what?
What is the point in testing?
And what is the intended use of the test results?
-> If the goal is to gin out some numbers for component comparisons, get it on. Just what is done and how it is done don't matter much at all, as long as the "results" are reasonably repeatable - show error bars, eh? And do NOT use defined units whose derivation cannot be supported (the use if "Watts" is the clearest example).

2) A bare copper slug is shown as the heat source, and I stand in awe at how uninformed the responsible test engineer is of even the basics of thermo.
If the "test results" are to mean anything at all, if the tester is going to use a defined unit like "Watts", then just how has the heat input to the sink been quantified? It is the most simple thing to measure the heat energy input to the slug (with certain assumptions), what the hell does that have to do with the heat dissipated by the sink?
-> What are the secondary losses? And under which conditions? (The answer is a rather large number of variables/curves whose combined value is a single number ONLY under a single combination of the affecting conditions.)

3) How about the secondary heat losses in the rest of the system, including the sensors, (the primitive) flow meter, and ultimate sink? Again another slew if curves combining differently at each data point. Not saying this has not been done, but I will opine that that the results from the equipment shown are not accurate at all. Been there, done that.

3) Anything can be correlated (in theory) to anything else. But I can assure all, from years of hands-on experience, that correlating the heat flux from a slug to a cpu will be the final test project that an individual undertakes. They will be pensioned off long before they get it right.
-> There is a good reason the chip mfgrs use a die simulator with an IHS; for the quantification of test results, and the comparison across platforms.
Sorry guys, you can't get there from here.

This is not a debate, just some comments intended to provoke some critical thought by the tester and the readers.

Be Cool, BillA

:eek2:

He's alive. . .and he still knows how to open a can of Whoop. :rofl::up:

[relttem]

When we do our tests, using a copper plate - which is actually milled to resemble a CPU and works very well, we measure our heat input, and we can also see how much load is being removed by the fluid. Then, as a result we know what is being lost to the surroundings (not picked up by the fluid). My experience with AMD and Intel in their CPU tests was that their mock CPUs were just pieces of copper. I have never seen anything special about them. They heated up like any copper block would. I did once work on a thermal head system that used dead CPUs, but that was only to make sure everything fit. But, if you do your experiments correctly and consistently a copper block works great for fluid tests..NOTE that we are NOT testing blocks. But, I would still bet money that if I used our copper block and found that one block did better than another it would do better on a CPU. All-in-all though, there is no such thing as a perfect experiment. And, as stated above, you can correlate/massage data to mean anything. We just try to account for everything and keep it all consistent.

[Martinm210]

FYI, wayback machine...prior to 2003..:)
http://web.archive.org/web/20060507072627/thermal-management-testing.com/methodology.htm

http://web.archive.org/web/20060507072627/http://thermal-management-testing.com/WBconn.jpg

[Computurd]

umm BillA i have never seen ya, but that is one of the most intelligent things i have read on the boards in a long time :) kudos!:up:

[gmat]

Be Cool, BillA
Wow, way to appear at the most unexpected moment ! Welcome back mister to this little hobby of ours.

[CoolingTech.]

I let the discussion go on without intervening to see how it develops itself:); however i would like to reply to BillA(Bill Adams) concerning two things.


And the cycle repeats, a new crew (like all such) that cannot dredge up and learn from what has gone before; so sad.
Italy is a swell place (more or less grew up in Firenze and Napoli), and not so far from the rest of Europe that French, German, and Dutch testing should be a mystery.
What on earth is the need to repeat the most basic cpu heat dissipation testing ERRORS that have been cussed and discussed infinitum ad nausium?
I am old and out of it, but the magnitude of the sloppy (by all parties') thinking displayed causes me to react.
well, first of all really thank for replying to the thread; I really appreciate it and i always appreciate peoples that spend some times to discuss our job.
However I think that there is a main point that seems that i couldn't made it clear; so lets say it another time.
Our goal is NOT simulate a cpu, this is impossible like is impossible simulate all the cpu with just a single load block(different die size/position/tdp etc...)but instead, is to have some data with fewer variables possibles that can be applicated with some errors to all the cpus.



2) If the "test results" are to mean anything at all, if the tester is going to use a defined unit like "Watts", then just how has the heat input to the sink been quantified? It is the most simple thing to measure the heat energy input to the slug (with certain assumptions), what the hell does that have to do with the heat dissipated by the sink?
I don't think that me or anyone else wrote that the watts that we give(resistive load) = heat dissipated, this is just your conclusion.
If you are reffering to the charts, the values that you find inside them are explained in our test methodology.
Here there is also another discourse, as you can see inside the web there is a lot of persons with different ages and also with different instruction, a public website must make any content more easy as possibile to be understood by all and this means more simple and "fast read" charts.
There are over 10k posts on our forum concerning the test and we had wrote 9 pages on our testing methodology, please read them before said any, wrong, conclusion.
I know that them are in italian but as i wrote i reply to every question that come to me, just send a pm to this account or compile our website form.



-> What are the secondary losses? And under which conditions? (The answer is a rather large number of variables/curves whose combined value is a single number ONLY under a single combination of the affecting conditions.)

3) How about the secondary heat losses in the rest of the system, including the sensors, (the primitive) flow meter, and ultimate sink? Again another slew if curves combining differently at each data point. Not saying this has not been done, but I will opine that that the results from the equipment shown are not accurate at all. Been there, done that.
True, this is a problem on what i personally spent a lot of time.
However for do what you are saying, probably we need a figure with six zeros and a number in front, for this job however you shall call a reasearch center and not coolingtechnique; also this isn't a problem that afflict just our methodology but interest every test with liquid cooling components.
So, what is the point?
We can choose two ways:
-testing with a single cpu with all the errors that you said + the errors coming from unstable heat of the die and having the data that are reliable just on that system
or
-testing with a load block witout the errors coming from the heat source(you know the exaxt power, and that is applied mathematics) that is adaptable to every systems with the errors that you said.
we chose the second one, why?
simply because we have fewer variables and data are more reliables.
edit: please note that we start testing on copper after we tested on a cpu with the internal dts and with an outside high precision thermocouple:
http://www.coolingtechnique.com/guide/dissipatori-waterblock/322-cablaggio-e-posizionamento-sonde-di-precisione.html

Professional test require a lot of money and time, you must choose between spending x000.000€ for testing a wb with no error or spending x000.000-990.000€ for testing a series of wbs with some errors.
Persons test their cars on "power rulles"( i don't know the exact word sorry) to SIMULATE the real street but it ISN'T the real street, is just a way to have some comparable data thank to the standardization of variables and this is what we are doing.
Obviously we had spent a lot of time(2 years) to understand all the points on how our system work and, when we have found all the constant that there are in every test session, we start to make some test that allow us to obtain comparable data.
You talk about "basic" but this is the basic of testing, remove all the variables(or more variable as possible) and understand what constant have you got inside the system; this will allow you to obtain comparable data =).
Feel free to contact me for any more questions and thank again for the reply;
Damiano.

[BillA]

Martin
that photo was while I was learning (lol, the whole time), but it was early - pre Swiftech
better photos here (pg 5) http://www.google.com/translate?u=http%3A%2F%2Fwww.cooling-masters.com%2Farticles-38-0.html&langpair=fr%7Cen&hl=en&ie=UTF8, post Swiftech

http://www.cooling-masters.com/images/articles/gts240/images/banc0.jpg

http://www.cooling-masters.com/images/articles/gts240/images/banc1.jpg

instrumentation, magnetic flow meter, and chillers not shown, secondary losses/sources must be controlled AND quantified

I might suggest that the testers in this thread coin a new unit, say "Italian thermal units"; then no problems. But this will relate ONLY to their test rig and conditions.

Be Cool, BillA

cross posts
fair enough Damiano, you are correct that I did not go to your site and read your background testing info
- I just saw that slug and uninsulated lines and sensors and "flipped out"
I will go back to sleep now
Ciao

[shazza]

Nice to see BillA here.

In reference to the fact that people seem to repeat the same testing errors ... that does seem to be a problem we encounter as people come into the testing arena. I've done some research on the "old stuff" and it's fascinating and there's much for all of us to learn from. Problem is, it's a bit difficult to run down the original testing work. We did have some interesting discussions here (http://www.xtremesystems.org/forums/showthread.php?t=243982&highlight=built+liquid+cooled) about the history of water cooling. I started a bibliography/reference list of resources, but sadly got sidetracked :(

[Martinm210]

Martin
that photo was while I was learning (lol, the whole time), but it was early - pre Swiftech
better photos here (pg 5)

Yes, I just wanted to share with some of our newer members that you've tried this to an extremely detailed level.

You're the only one I know who has gone through all the die simulator experiences/challenges and it's easy to forget with the old sites no longer being hosted in their original location.

It will always be good information, even if we have to extract it using the wayback machine.:yepp:

Thanks for stopping in, please do so more often if you can find the time....:up:

[zytrahus]

Hello Eddy, nice to see you also here.
Indeed our main objective isn't make a pure simulating of a cpu(why simulate a cpu if we can use a real one?)
the main object is to standardize all the testing procedure removing all variables from the testing section.

.../...

Cheers from Italy.

Over the past years (#1) what seems to be the most important thing is the ability of a "tester" to observe/comment/review on what end users will be observing when they buy the product. (#2) Next on the priority list would be to comment on things end users may not see? (I wouldn't care much about #2 if #1 can't be achieved)

what's the point of removing what you call "testing variable" (which sounds like a good thing) if you end up with results different than what actual users are observing? I say "if" but there is no reason to observe the same results, really. You just can't simulate CPUs thermal behavior with a simple piece of metal and a resistor. Heat flux and hot spots are only parts of the reasons why you want to avoid Die Simulator. TIM joints are another very good one.

as an enthusiast if I am shopping for the best waterblock for my new $3K rig I want to see data that I can trust - I.E. data I will be able to replicate in my non-lab equipment.

[SpuTnicK]

Well looky here (http://vr-zone.com/articles/vr-zone-coolers-testing-methodology/11089.html). Another CPU simulation stand dedicated for processor coolers.

you can start throwing stones at it right about .......now! :D

[Waterlogged]

Well looky here (http://vr-zone.com/articles/vr-zone-coolers-testing-methodology/11089.html). Another CPU simulation stand dedicated for processor coolers.

you can start throwing stones at it right about .......now! :D

http://img59.imageshack.us/img59/3129/impliedfacepalm.jpg

[CoolingTech.]

You just can't simulate CPUs thermal behavior with a simple piece of metal and a resistor. Heat flux and hot spots are only parts of the reasons why you want to avoid Die Simulator. TIM joints are another very good one.


...... so lets say it another time.
Our goal is NOT simulate a cpu, this is impossible like is impossible simulate all the cpu with just a single load block(different die size/position/tdp etc...)but instead, is to have some data with fewer variables possibles that can be applicated with some errors to all the cpus.
.....
:)
and also please read the last post of AndreaBZ(Andrea 22) on our forum:
http://translate.google.it/translate?js=n&prev=_t&hl=it&ie=UTF-8&layout=2&eotf=1&sl=it&tl=en&u=http%3A%2F%2Fwww.coolingtechnique.com%2Fforum%2F showthread.php%3Fpid%3D28363%23pid28363

All the data are worked with some precise mathematical functions derived from tests performed on real CPU; the tests are more complex than they seems and for sure they are totally different on what vr-zone is doing(just heating up).



as an enthusiast if I am shopping for the best waterblock for my new $3K rig I want to see data that I can trust - I.E. data I will be able to replicate in my non-lab equipment.
the data that you have from a cpu test are valid just on that cpu, on your 3k$ rig that test are more unreliable than the data that you can obtain from a standard testing procedure.:)

@BillA PM for you!

[zytrahus]

I understand what you are saying but I disagree with what you are trying to do.

The big picture is: You basically play with a real CPU (and its non-perfect thermal probes). Then you play with almost perfect Die Simulator (yeah I know it's not what you want to call it). Then you mix it all up in Excel and extrapolate correcting laws that transform results from your Die Simulator into Real CPU results.

Result:
You will go ahead and install heat sinks, water blocks on your Die Simulator.
You will come up with results thanks to your "precise mathematical functions derived from tests performed on real CPU".
You will assume those tests are correct.
But they will not be correct.

Because your extrapolations cannot take into account all of the phenomena that happen because of the certain mechanical/thermal contact you have between a real CPU and a water block that you haven't tested yet. And... it's just an example (mechanical/thermal contact).

In case this is too subtle I'll give you an obvious example, say tomorrow someone comes up with a waterblock with a 20x20mm pod on its base. Suddenly the heat flux is no longer the same. Oh crap, you have no "precise mathematical functions derived from tests performed on real CPU" to correct your numbers. Then what?



All of this complication and potential for errors for what exactly? to be able to use more accurate thermal probes? To have a more stable heat source (assuming you have a decent way to regulate your WB inlet temperature because your heater resistance will vary quite a bit with the resistor temperature...)?

Many have been through this. And in the end all of this is pointless.. unless you are really fond of playing with power supplies and pushing buttons.


PS: what I meant to say with my example of end users shopping for a block is: OK I have a i7 920 and I am looking for a new CPU block: I am going to look for a waterblock review that was done on a i7 CPU and if possible a 920. All in all, numbers will be accurate enough for any end user even if, granted, on some occasion 2 similar CPU could give pretty different results. But oh well, you're also using real CPU to define your correcting factors, so what the heck, huh?

[BillA]

posted on coolingtechnique

Hi guys
I'm going to write in English as that is most comfortable for me. Then I'll look at it in Italian to see if it needs clarification. (Yes, I lived in Italy some 50 years ago.)

Damiaon posted on xtreme and I responded bluntly, as is my custom - but without having visited this thread to better understand how you fellows were testing. No, I have not read all 50+ pages; but I've read enough to generally see the methodology. Let me share something of my experience testing watercooling components.

For those not familiar with me, I started testing in my garage and posting on procooling, eventually I was the VP of Engn at Swiftech, and then CTO at CoolingWorks. I am now retired and live in Drake, Costa Rica. Other than a million plus shares of worthless Swiftech stock, I have no economic interest in anything related to watercooling (no horse in this race).

It would appear that your 'test bench' consists of a heat source (copper slug), the DUT (device under test), a pump, and a heat sink (air cooled radiator - heat exchanger); plus assorted sensors and gauges to instrument the system. What I have just described, less the heat source and instrumentation, is what is sold in the industry as a "watercooling kit". The implicit/explicit assumption is that all elements of the "kit" are known such that the specific performance of the DUT can be characterized and described under a variety of conditions (with different pumps, radiators, fans, and heat loads).

Well, you are wrong. What you are testing are 4/5 items whose individual performance is interrelated to the others'. You may believe that you have a (valid/correct) correction factor for each component, but each component measurement/correction will - indeed MUST - have an associated error. What is the error for each component? How will you determine such when testing a system (kit)? Error is additive so the error for 5 components will be vastly larger than for one.

I will not dwell on accuracy as this is a function of the equipment used and its calibration; but if its not calibrated using a (tested and calibrated) reference standard, the measurement is - by definition - NOT accurate. If you were to be audited, the first item is test equipment calibration, then individual sensor calibration. But checking for calibration stickers confirms only that the item "is within the manufacturer's specifications", this is not sufficient for the testing we are discussing. What is needed is the actual deviation curve within the intended range of use. The calibration lab will provide the data points (over a defined range) from which a "measurement correction curve" can be generated and applied to the raw data. I used to calibrate my PRTDs "in house" and trust me, each one had a different curve even though purchased in groups.

Let's look at the parts of your "watercooling kit" test bench.

You have a copper slug heat source with its power supply, voltmeter, and ammeter. This will tell you nominally the heat generated within the cartridge(s). There is a thermal interface (translate as gradient/thermal resistance/temperature step) between the cartridge and the slug, the thermal resistance of the copper itself to the DUT interface, and somewhere between the cartridge and the DUT interface, another interface with the temperature sensor. Knowing the actual dimensions, the interface temp can be calculated (calorimeter); but I would not imagine you have done this.
(A possibly better solution is a ceramic heater as made by Watlow, about any size and wattage - custom made.)
So how much heat is being input into the DUT (a very basic question)? A blind man on a galloping horse understands it is not the nominal heat generated by the cartridge, that might be called the gross heat input. With watercooling that question can be answered by calculating the heat picked up by the coolant, the temperature rise and flow rate. For this calculation to have any meaning the measurements must be extremely accurate.

You have a pump, flowmeter, and probably another voltmeter and ammeter. The pump is inputting heat into the system, how is it being measured? The flowmeter you are using is inadaquate; it does not have the necessary accuracy, too much flow resistance, and excessive heat loss. Tell me, what is the parallax error for each reading? And between operators? Do you have a flow resistance curve? What is the flow meter's heat loss vs flow rate vs temperature (ambient and coolant)? Thermal calculations without these values will be fantasy.

You have uninsulated hoses and temperature sensors in metal fittings, what is the coolant heat loss from all this?

You have a radiator and fan(s), with presumably another voltmeter and ammeter(s). For thermal calculations you will need the air temperature and flow rate. Understanding that you will not (ever) have the air flow rate, for repeatability and dissipation calculations you MUST have the temperature. How are you measuring it? If it is not being measured per the Intel procedure, you don't know what you are doing. The air temperature varies hugely in an uncontrolled environment, to know it you need multiple, continuous, averaged measurements. Here is an instrumented radiator from http://www.cooling-masters.com/articles-38-0.html

http://www.cooling-masters.com/images/articles/gts240/images/banc1.jpg

So what do all these words mean?
-> Test components individually, not as a part of a system.
Reduce variables wherever possible, for watercooling this means an environmental chamber AND a chiller (or two). If testing a waterblock, suspend it to eliminate the hose bending moments.
Use accurate and calibrated instruments and sensors (2 decimal places minimum, magnetic flowmeter with digital readout)

Too much money?
OK; but if you test as a hobbyist, don't make fancy claims.

Be Cool, BillA

[CoolingTech.]

@zytrahus i never said that our test are the state of the art of the perfection:)
there are some problem inside all the system but fortunatelly all this problem are a constant that can be isolated, you just know that there is a xC°+kC°
Concerning the pod, if a brand chose to do it probably it's perform better than with out it, it is an intrinsic characteristic of the wb same math.
If we have a wb that reduce the flow more than another one, we do not increase the pump voltage to have the same flow, simply that is a characteristic of the products and probably use that feature to perform better.


PS: what I meant to say with my example of end users shopping for a block is: OK I have a i7 920 and I am looking for a new CPU block: I am going to look for a waterblock review that was done on a i7 CPU and if possible a 920. All in all, numbers will be accurate enough for any end user even if, granted, on some occasion 2 similar CPU could give pretty different results. But oh well, you're also using real CPU to define your correcting factors, so what the heck, huh?
the real cpu is just needed to have a "porting" to real situation with the spreadsheet.
The advantage of this testing procedure is that when we have the thermal curve concerning the wb(taken using the block) you can apply it to every situation(aka cpu) downloading our spreadsheet and simply selecting yur rig.
This will give to you also you future temperature if you, for example, would like to change the radiator,pump or something else and this for me is amazing.
http://www.xtremesystems.org/forums/showthread.php?t=206542&highlight=spreadsheet+liquid

Now obviously the data aren't perfect and i'm the first one to say this thing, i think however that the charts and the data that we are taking are already a good starting point to understand the performance of a product also on multiple systems.
There are some problems inside the testing methodology and BillA already listed them, i'm already working on(thank also to the same help of Bill) but this mean time to improve the data;I'm pretty sure that in the next months we can do really interesting things.

edit:@BillA concerning the calibration of the instruments we have a calibration certificate($) of the instruments and this actually is the max that we can do.
This problem interest all the tester,like the flowmeter, on the net( honestly i don't think that every one have a certificated instruments) and not just us, in this way we can't do actually more.
i will reply to the other points with the help of andrea in the spreadsheet discussion:)

[gmat]

edit:@BillA concerning the calibration of the instruments we have a calibration certificate($) of the instruments and this actually is the max that we can do.
This problem interest all the tester,like the flowmeter, on the net( honestly i don't think that every one have a certificated instruments) and not just us, in this way we can't do actually more.
I dont think it's precisely his point here. Of course you cant do more in this aspect, that would be too expensive.
What Bill says is (to my understanding), *try to know* precisely the error margins of each one of your equipments, and know how to add them up together, then report that error upon your results (graphically for example). And for example try with different operators so different eyes see different results, report the measurement differences, because the parallax error when reading needle instruments (again for example) is not insignificant. Once you know the limits of your instruments (and of your own eyes), do not forget to report them on your results tables and graphs.
Also the measurement response from each instrument has a deviation that is not linear related to the input parameters, indeed knowing that response factor can help improve the accuracy of your measurements.

[BillA]

Hi gmat, thanks for the input

One of the most difficult issues in testing is accepting that what one believes may not be correct, it hounded me for years until I learned humility (imperfectly, yea I know - see, there it is again !!).

A rather detailed appraisal can be read (in English) here http://www.coolingtechnique.com/forum/showthread.php?tid=274&page=52

Be Cool, BillA

[rosco]

Hey Bill, glad to see you here. Hope you're OK since the last time ;)

Same discussions about thermal considerations come again and again, it's an endless loop :D

[BillA]

Hi rosco
I used to think I was Don Quixote, now I know I am only the mule.

Be Cool, BillA

[Quagmire LXIX]

For those not familiar with me, I started testing in my garage and posting on procooling, eventually I was the VP of Engn at Swiftech, and then CTO at CoolingWorks. I am now retired and live in Drake, Costa Rica. Other than a million plus shares of worthless Swiftech stock, I have no economic interest in anything related to watercooling (no horse in this race).


lol, much of this stuff is over my head, but this BillA is a treat. Also may be the oldest join date with the least amount of posts by someone still posting.

Thanks for the great read everyone involved :up:

[CoolingTech.]

ok guys, just a little update.
the discussion continued a little bit into our forum( same url (http://www.coolingtechnique.com/forum/showthread.php?pid=28619#pid28619) that Bill post some posts ago)
we are going to making some improvement, thank to help from Bill, due to remove some errors(costant in each test) that we already have into the test, of course will be not the wayback machine(this need some time more:D ) but for sure will allow us to obtain more reliable values and allow us to obtain comparable data on multiple platforms.
I would like also to share this link of Bill, that explain some of our points and i think is really a good conclusion on what we have wrote in these three pages :
http://www.overclockers.com/the-evolution-of-aftermarket-heat-sink-waterblock-testing/
The first updates and an english methodology will come in a few days.



What Bill says is (to my understanding), *try to know* precisely the error margins of each one of your equipments, and know how to add them up together, then report that error upon your results (graphically for example). And for example try with different operators so different eyes see different results, report the measurement differences, because the parallax error when reading needle instruments (again for example) is not insignificant. Once you know the limits of your instruments (and of your own eyes), do not forget to report them on your results tables and graphs.
Also the measurement response from each instrument has a deviation that is not linear related to the input parameters, indeed knowing that response factor can help improve the accuracy of your measurements.
yes you are in right, like I wrote into CT our main object actually was reach the level of the main testers into the net and simply have a comparision with more users as possible, thank to this we can improve our test and reaching maybe a better point of view than what is already into the web.
A lot of the points that BillA listed are present also in the others "testing rig", we however will try to fix them improving our data and this was the purpose of the thread and our main point.
Keep the subscription!;)

[Martinm210]

http://www.overclockers.com/the-evolution-of-aftermarket-heat-sink-waterblock-testing/



Ah...good read!:up:

Any my hat is off to anyone attempting to do detailed thermal testing of blocks and such, it's a huge amount of work and a challenge for sure at the hobby level.

Cheers to science!...Martin