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

I have read many posts from this forum (as well as the [H], AMDMB, and Overclockers) about the advantages of larger tube sizes (everything 1/2 ID). Yet the only comprehensive benchmarks I've seen regarding Waterblocks has been at Overclocker's .

If you look at the top rated WB, it is the Innovatek which uses a block that has only 10mm OD (8mm ID) tubing, has an Eheim 1046 pump, and a crappy radiator (Heatercores and even BI rads are better).

So I have two questions, First, does anybody know of a review on the newer Maze 3 (uses 1/2 ID tubes I believe)? I know according to Birrman54 in the [H] forum that he is going to do a Waterblock roundup with both the Eheim 1250 and 1048 pumps (not sure about tubing size though), which will include this block as well as several other popular blocks.

My Second question is tubing size, has anyone done a performance review of various tubing sizes as related to performance? I know that the WB and Pump would probably affect this kind of review, but might be worthwhile to everythone to actually see some numbers (instead of some of the ranting I have seen done in all of the forums on this subject).

Thanks for the help.

Unaclocker (who used to post here) swapped his 3/8 for 1/2 and saw ZERO difference. The extra flow in the waterblock was negated by the fluid passing too fast in the rad.

I have yet to see a Maze3 review.

honestly, i would need more than one review / roundup to make a decision as to what block is best, etc. unfortunately, i havent found any reviews other than what Overclockers and ProCooling has posted. PC is supposed to continue their WB roundup with the AMD waterblocks (Maze2/3, spir@l, etc etc), but Joe doesnt seem to have time to do that at the moment.

so, nope, i dont know of any articles.....

and as far as what Ben said, unless una had flowmeters testing the flow rate befor and after the tubing swap, id take that comment with a grain of salt (not that i dont believe him, but we need hard numbers here)

I agree CD, we need more numbers.

Now that you mention it, simply swapping hose size shouldn't make a d*** difference, simply because most of the resistance is within the WB.

Una (not here anymore) said that his flow rate increased 400% (how did he know?).

OC has numbers too...

You don't need more low quality numbers and testing. What is really needed is a little more theory on heat transfer, fluid mechanics, and flow through pipes and fittings. That will give you some perspective on "the big picture"; there is no way to fully optimize flow and water velocity in every person's setup with a waterblock (or radiator, fitting, pump, etc) roundup. It is just too complicated a problem. However, to some extent the numbers are there if one looks at them in Joe's waterblock roundup. For the most part, flow rates of 0.5-1 GPM are typical with ~200GPH rated pumps and 3/8" setups. For 1/2" setups and the same pumps then 1-1.5GPM is more normal. Look at Joe's temperature vs. flow data and for each block you can get a reasonable idea as to how it will handle lower or higher flow rates. You can also see for most of the blocks that there is a 1-3C improvement in the range of 0.5 to 1.5 GPM.

Here is some old data of mine when I first got a flowmeter. This data doesn't have the proper resolution (I am nearly ready to start testing blocks now though with much better temp resolution and decent water temperatures finally), but it shows the general trend:

http://phaestus.procooling.com/gpmvary.jpg

Comparing the block vs water temp would be better, but since the pump puts more heat in at lower flow rates (it was throttled with a ball valve) then this graph might perhaps give a better estimate of "overall system performance". I personally am more interested in block performance only at the moment...

Has anyone ever checked out all the data here:

http://www.dansdata.com/coolercomp.htm

I find it quite interesting, and all the tests are done the same.

Thanks PHaestus, I've always been a proponent for more theory.

I see your graph, and it seems to indicate what every reasonable OC'er has been saying (more is not better). I've read MANY articles, and I plan to read them over again a few more times.

Test conditions are very important, and in my opinion, if one can't evaluate the margin of error in a setup, then the data has little value. (but the theory might be)

From your graph though, I would advance this theory: an increase in flow rate will generate better heat dissipation from the waterblock, but MUCH less dissipation from the radiator. I'd like to see someone test different flow rates with a Big Momma, then try the same test with TWO Big Mommas in parallel. (maybe I should ask Miss_Man?!?:D )

I'm also starting to consider copper tubing (instead of silicone/vynil/Tygon), or even better, finned copper tubing...

WebMasta33, yes, Ive seen Dan's data. The numbers aren't reliable (although they are consistent), but the ideas that he advances are great. (Ooh check out that OCC!). The heat source is what makes Dan's data relevant.

Under ideal circumstances, one would figure out a way to supply water at a fixed temperature (maybe using one of those thermally controlled valves used in photo labs), so that the testing of a waterblock wouldn't depend on anything other than itself (i.e. there would not be any external factors affecting the performance test). Of course this means dumping the water out during the test, so unless you have one of these, you're looking to run up quite a bill...

Careful; you are reading a lot into a graph that has low resolution and perhaps isn't the best anyway. The thing I would really feel comfortable with saying for most of the blocks we use is that if your flow rate is below 1GPM (which it probably is using 3/8" tubing) then there is a fair amount of performance that you can gain by switching to 1/2" fittings. You could also probably get nearly the same effect by just drilling out your existing 3/8" barbs as much as you can.

Thanks for all the replys...
 

Nice to see all the support in this forum.

As to the debate on testing, I don't think there will ever be one clear solution on this. To be honest, I would like to see an overclocked AMD system (maybe at 2.1v) with just standard watercooling stuff. Having something that generates gobs of heat to test the system isn't realistic. As is using a constant flow of the same temperature water. Both of these would give a best worst/best case scenario and nothing real world. My Ideal test setup would be something like this...

1. A system with a AthlonXP overclocked and overvolted which should generate some good heat (as well as performance;) ).

2. Installed in a common case type (Antec would be good I suppose, but really any good Mid-Tower should work), with standard fans installed.

3. In a room where the temperature could be monitored, but with the case closed up (not opened which could skew the numbers).

With this setup you would have a fairly common setup (similar to the various Heatsink roundups you see done on the various websites).

Now you could setup a baseline system (say a Maze3, with 1/2 ID tubes, connected to a heatercore, and an Eheim 1250 pump).

Block Testing:
Once you have the baseline numbers, you simply can swap out different Blocks to test them with this setup.

Pump Testing:
Once the various blocks are tested, you could change the pump to a 1048, or 1046 and perform the tests again (which would give you a good baseline on pump performance).

Tube Testing:
Now simply change the hoses to say 3/8 ID (or smaller) and do the same tests again.

Radiator Testing:
Just as with the other tests, you could just change the Radiator (BIX, BI, DangerDen, Heatercores, etc), and record what the benefits are for each one.

This is my opinion, but I think this test would be more realistic to a lot of people. Just a thought, what do you all think?

If you can't control the variables and you can't measure them with any certainty then you will end up seeing whatever you want to see. I have a testbed that is about as good as one can have that is still "real world", but I would still prefer to eliminate the radiator from the equation and use something to provide a constant inlet temperature. Simulators are used in place of processors when people want to actually quantify the power (and heat) that is going into the system. That is really the only way to compare different heatsinks or blocks.

If you look at AMD's technical documents, they use die simulators that are very advanced (monitor heat in several places) and do not bother with socket probes to draw their conclusions. Why wouldn't you expect the same from an "expert" reviewer?

PowerHouse: I completely disagree with you, and on more level than one! But I do agree that testing on a typical computer is necessary.

Now, here's why I disagree:
1- The existing line of CPU's constantly change, so to keep the same testing rig throughout all tests would NOT give us a result for the greatest/latest computer. You know how fast they go out of style... I still use a Pentium 166!!! (among others)

2-CPU temps can vary, depending on the load that is put on them (regardless of over-volting). You can argue that one can use some benchmarking software, but the problem is that every computer configuration is different, and I've seen too many people run the test, then read the temperature right away and report it, when in fact, a completely assembled system could take many hours to stabilize. (One factor is the volume of water)

3-That being said, given the choice between using a $750-$1000 system to test, which has to be dedicated for that purpose, with the hope that it will never break down
VERSUS
using a $25.00 high-power resistor, a power supply, and some measuring instruments (flow-meter, temp probes, volt meters, etc...)
Well, you do the math, and honestly tell me which one you would trust the most.

4-Given the above info, and what you propose, you need to understand that the time required to test all those configurations would take days, if not weeks, and that sometimes, the guys reviewing that stuff, don't have the item in their hands all that time.

5-What you are proposing comes down to testing a complete configuration, not an individual component.

6-The pump: it's very simple, use an industrial (or big) model, slap a flow restrictor, and a flow sensor, along with a pressure gauge. This would allow for a lot more usefull data, and would be much simpler than trying to install different pumps.

As for testing a pump, the specs are already available everywhere, so don't see the point in testing them, except for reliability, and mods.

The rads: setup as above (pump with flow limiter, a standard block and a standard heat source).

Tubing: tubing doesn't really need to be tested. The only thing a bigger tube will do, is add water mass to the system, and ease flow, but it's so insignificant compared to elbows, and most importantly the flow resistance in the water block (greatest of all) that it's not even worth mentionning.

You're also forgetting that regardless of the optimal setup someone puts together for a "typical" system, as you would put it, there may be other configurations that could be just as effective (i.e. a DD Maze3 with a BIX versus a Gemini HF-spiral with a Big Momma). The thing to keep in mind, is that people like to have a choice. I'd rather know that I can get a Swiftech with a Bix, using an Eheim 1048, because the optimal flow rate of each component is the same.

Now if there is someone out there than can do all that, THEN test it in a typical (at press time) system, then I would worship them!

I do agree with this, Which is why I recommend a base system. Just as the AthlonXP runs cooler than the original Athlon, and so will the AthlonT-Bred run cooler than the XP. If you have a base system, you could just compare the difference with the new CPU and adjust all you numbers accordingly.

You could run Prime95 (or folding@home, etc), which should be sufficent to heat up your system for the testing phase. And, as you said, you would probably have to run each test overnight to get accurate numbers. However, this puts a worse case scenario as the baseline (which isn't necessarily a bad thing), but some people don't work there PC this hard all the time, so perhaps a mix of Idle, Medium, and Full loaded CPU testing would be in order.

I take all the tests I see on various websites with a grain of salt, but they do give you a general idea of what's best on the market, which is what we are all interested in. (which is why I bought my SK6 heatsink, everybody gave it a great review on performance).

True, the tests would take a long time, but isn't that what you want to see anyway. Would you trust a huge test like this that was done in only a day or two, or one that spent a couple weeks working on? Besides you could release a few test subjects at a time, and drag the review out over a month or two (keeps people coming back to your site).:p

Yes, I want a complete system (but of the best stuff, not a Koolance or Swiftech system). I have yet to see any review like this, but I do believe its what people have been waiting for. However, you could break it down to just testing one component at a time also (all WB's, All Rad's, etc).

I have read too many posts about people using a larger pump and getting better numbers etc. This was just an Idea to see if different pumps actually make a difference or not. Plus you wouldn't just have to use Eheim, you could use Rio, Danner, or others. I know the pumps specs are listed, but they perform different given the tubing size, radiator type, number of elbows etc.


This review wouldn't limit the choice, on the contrary, it would provide numbers for each item, so you would at least have somekind of baseline to go on. So although your Swiftech with a Bix using an 1048 might have an optimal flow rate, if the performance of a Innovatech with a Dangerden cube, using a Eheim 1046 had better performance, wouldn't you consider switching (assuming the performance difference was enough to make you spend the extra money).:D

I know what you mean, I would love to see a review like this (even a partial review like this would just be awesome). Maybe Birrman54's review will provide us with some answer's. Any other takers on this one?

True, but there is control, and there is CONTROL. What you describe is what someone would do in a lab environment, not what the typical person would use in his home or office.

If you eliminate the radiator from the equation, then you actually skew the numbers as this wouldn't be a real world situation (at least I haven't seen anybodys watercooling rig without a radiator). So in effect you would be saying, that given a constant temperature water (hmmm, what would you consider a normal temperature for water?) WaterBlock-A can keep CPU-X at Y temperature while WaterBlock-B functions at Y+5 degrees. While this sounds like a good test, what you may miss is, that given a higer water temp, WaterBlock-B may be able to keep the CPU-X temp at it's original temperature, but WaterBlock-A actually could perform worse.

I have heard a lot of talk about birrman's roundup, but have not seen any information about how he proposes to actually test the blocks or what equipment he is using. I am soon to start on the second Pro Cooling waterblock testing, which will (due to time and money constraints) be limited to a few popular blocks. If you are interested in my testbed, I am using:

1) Little Giant 3E-12NYS industrial pump. This is a big unit (200W heat but ~7.9 PSI)

2) flowmeter from mcmaster. This is in need of upgrade; it really affects the flow rate and is not digital so think of it as a crude approximation. I am still looking for a good replacement.

3) Water and air temperatures: wb inlet and outlet temps taken with Dallas 1-Wire probes that have been sealed with silicone and placed into 1/2" NPT caps. Radiator intake and outlet and also room temperature taken also with DOW sensors. Specs are 0.5C accuracy and 0.125C resolution.

CPU temperature: AMD Duron 1.0 (Morgan) at 1200MHz 1.9V (or more if I get around to vmod). CPU under temperature taken with DOW sensor epoxied under center of core. AMD diode readings taken from a MAX6655EVSys diode reader that interfaces with a computer via parallel port. 1C accuracy and 0.125C resolution on this one.

Pressure Drop: DPT measurement setup that I still have to get running. I will use PVC crosses to put the DOW temp probe and the DPT setup close to the wb inlet and outlet.

This all sounds terribly impressive, but the temperature readings are still not really that great (I really want 0.1C ACCURACY) and it is sort of a bastard setup in that it is higher quality than most general users are willing to put together but it remains much poorer than what is really needed. In the next few months I will likely upgrade the flow sensors and add a recirculating water bath to keep the water at exactly constant temperature (we have a VWR unit in our lab that I may try to appropriate if it keeps making strange noises). The temperature monitoring is troubling and will be very expensive to do right. I am for now satisfying myself with doing the best job I can with the equipment I have (replications, careful variation of flow rates, minimizing errors) and I figure the most pressing upgrade needs will be obvious once my test methods are established.

Is there any physical basis for what you are proposing? I have never heard of such...

Why? I see no reason that radiator performance must vary with flow. Perhaps you could explain your reasoning?

Heres my theory. Everyone can afford $17 for a heatercore (or a Blackice) that has 4x the cooling power they need. Therefore radiator and pump dissapation are not particularly important.

Furthermore the thermal resistance of copper is fixed. Therefore no matter how much you flow you can only approach a given effciency with a given design. Thats why more flow will eventually not yield more performance, but not because of radiator effciency.

Here is some better data (from BillA)
 

http://www.ecom-answers.com/462U70W.jpg

That might change the discussion a little...

To redleader, radiator efficiency is much like block efficiency. However, the relationship tends to be reciprocal. Imagine a radiator in which the water exited a mere second after it entered. There would be insufficient time for the water to approach the temperature of the air blowing over the radiator. On the other extreme, imagine a radiator where the water required ten minutes to get from entrance to exit. Upon exiting, the water would essentially match the air temperature. Each of these extremes would not exist in a computer cooling setup, but illustrates the concept involved.

This is the catch-22 of water cooling. Blocks perform better with high water velocity, yet radiators perform better with longer dwell time.

In pHaestus' graph, you see the effect in a block. As flow approaches zero, the differential between the CPU and fluid grows as 1/flow. As flow increases, the differential follows this same 1/flow trend. If a similar graph was plotted for radiators showing the differential between fluid and air, you'd find a relationship that is the reciprocal of that for the block. ie 1 over 1/flow = flow. As flow rate in a radiator drops, delta T drops. As flow rate through the radiator increases, delta T increases.

There is no magic in stating that blocks do well with high velocity while radiators do well with low (water) velocity. What more folks need to recognize is that velocity is not simply a function of volumetric flow rate. Velocity is volumetric flow rate divided by flow area. Blocks that utilize (properly defined) small flow paths can meet or exceed the performance of blocks using substantially larger flow paths (and correspondingly higher flow requirements). The kicker here is that a block with small flow pathways and low volumetric flow rate yields better radiator performance that a high flow block with all other components held constant. It does this simply because the lower flow of the former block creates more dwell time for the water in the radiator.

A corollary to this concept is that you can get equal performance from a smaller radiator using a low flow block when compared against a big radiator and big block. Make no mistake, to realize such a result requires proper design of the block and proper selection of the pump, radiator, radiator fans, and tubing. Unfortunately, the majority of water cooling users seem to blindly follow the "bigger is better" theory without regard for the interactions that take place between the block and radiator.

increased flow in a radiator does decrease cooling capacity. i am not nearly as advanced as those who have spoken above, but the graphs in BillA's test show that most radiators lose cooling capability at higher flow, except for two. the big momma he tested, did lose some, but not much, dissipation capabilities. the serck he used increased dissipation as flow was increased. you can see it here:

http://www.overclockers.com/articles481/index05.asp

its the third graph down. as you can see, the big momma didn't lose much as you increased flow. the serck, with more air flow, increased dissipation as you increased the flow. i can't comment on his testing procedures, and if it would satisfy you guy's standards however. although i do remember him as being a pompous advocator of testing methods.

Sure is, its called block design. One block design could perform better given the right circumstances than a different block.

So in the example I have given, WaterBlock-A might have a fast flow rate (Maze-3), while WaterBlock-B has a slower flow rate (Innovatek3). So give the constant water temp coming into the block (Say Room Temp of 70 degrees), the higher flow block would have a temp closer to the water coming into the block. While the slower flowing block would have a higher initial heat (as the water isn't moving fast enough to keep the block as cool).

In a closed system (like the real world) with a radiator, the temp of the water will increase with the heatload generated by the CPU. Now because of the design of WaterBlock-A (fast flow, low water resistance - read less surface area), the water flows so fast that it doesn't have time to absorb all the heat from the CPU. While WaterBlock-B with it's slower flow rate (yet higher resistance - read more surface area) can release more of the CPU's heat into the water.

This is where a good radiator comes into play. While WaterBlock-A's fast flow actually hinders the effeciency of the radiator, the slower moving WaterBlock-B gives the radiator more time to remove the heat. This is also why you see people needing larger radiators (Big Momma, etc), to keep WaterBlock-A cooler.

Thus as the water heats up, WaterBlock-A actually performs worse than WaterBlock-B. But in your tests, you wouldn't be able to see this result as you would just see WaterBlock-A keeping a lower temperature. Again, this is why I stress realworld examples vs. Lab testing.

You are again talking about testing a complete system and not evaluating the performance of any one item. There is no reason (based upon physics) that a block would perform unrealistically by controlling an inlet temperature. You are instead talking about radiator performance affecting waterblockperformance, which is precisely the thing you want to eliminate when testing either blocks or radiators!

Isn't that like saying lets test a heatsink without any fans? They require each other to perform correctly. Although I suppose I could take the fan off my SK6 and watch my CPU melt.:D

Think of it this way: If you heat up a frying pan, and drip water onto it at a slow rate, it will cool less than if you have a higher volume of water flowing over it. So in a test, the higher flow rate would always win. Not much of a test if you ask me.

The more seasoned testers of heatsinks either include multiple fans or vary the voltage of the fan to get an idea of how it will perform with different CFM through the cooler. That is the same principle here. Blocks always perform better with increasing flow rate, but how much better is not constant. By looking at performance curves for the different blocks and comparing them to your choice of pump and radiator, one would hopefully be able to make choices in parts that would lead to an efficient system that is performing at close to optimum levels. You also have the possibilty of developing a more coherent "big picture" sense of water cooling loop design than if you just test a lot of different complete systems and try to draw no larger conclusions from them.

Just because you seek real world results doesn't mean you have to eschew the fundamentals of science. Isolate the variables of interest and then vary them one at a time while holding others constant. Build up a model that effectively includes all the observations. That is science at its most fundamental level. That has nothing to do with a laboratory either; it is useful everywhere. Any time you have multiple properties varied though you can extract nothing useful from the results.

I'll answer this one in reverse order.

1. Pure science doesn't always work in the real world. Here's one example: People with Diabetes need insulin so their body can use sugar. However, when doctors were testing insulin on lab animals, all of the animals died. So being a scientist your first thought would be that insulin would be unsuitable for people to use. Hopefully you know that insulin is one of the treatments people take for Diabetes, and although it kills the animals it was tested on, it works perfectly in humans.

2. You kinda said what I have been trying to say (just harder to understand it I guess).

"Blocks always perform better with increasing flow rate, but how much better is not constant. By looking at performance curves for the different blocks and comparing them to your choice of pump and radiator, one would hopefully be able to make choices in parts that would lead to an efficient system that is performing at close to optimum levels. You also have the possibilty of developing a more coherent "big picture" sense of water cooling loop design than if you just test a lot of different complete systems and try to draw no larger conclusions from them.'

Although you could look at performance curves of different blocks, you still need to match them with the correct choice of pump and radiator (which is what I have been saying). If you just test the blocks, you would lead people to believe that the top performing block in your test is the best. However, this wouldn't necessarily be true, it would greatly depend upon the pump and radiator used with the block. Which leads back to my arguement that all three components need to be tested to find the best mix for the "optimum levels.'

When it comes to a radiator, a little bit quickly adds up to the same as a lot slowly as far as flow goes. It's how efficient the heat transfer in the block is that should make a difference if the radiator is not overloaded by being too small or bad airflow through it for example. Also getting the heat out of the block as quickly as possible will keep excess heat from being stored in the metal of the block. I think faster flow with the right radiator and block designed for it should be better than a low flow settup.
my $0.02:shrug:

According to This thread, Birrman54 is going to test the Spir@l, Maze-2, Innovatek rev3, Innovatek Flatflow, Gemini High flow, and TC-4 in his first installment. He is going to test using both an Eheim 1250 and 1048 (not sure which radiator or tube size he is going to use).

The results will be posted at www.subzerotech.com.

Here is the list of most of the blocks he will be testing (there are others he is going to add once he gets ahold of them)...

1. Spir@l (www.dtekcustoms.com
2. Maze 2 (www.dangerden.com)
3. Maze 3 (www.dangerden.com)
4. Swiftech MCW462-U (www.cooltechnica.com)
5. InnovaCool Rev 3 (www.highspeedpc.com)
6. InnovaFlat (www.highspeedpc.com)
7. Space 2000SE (www.case-mod.com)
8. TC-4 (www.dtekcustoms.com)
9. ?? (Steve from www.ocforums.com chancell@ihug.co.nz)
10. ?? (Steve from www.ocforums.com chancell@ihug.co.nz)
11. Gemini Hi-Flow (www.geminicooling.com

From the few tests I have read (one at Overclockers , the other at PCReports (Need to use a translater on this one, its in German), the best blocks all had 10mm OD (3/8 OD) tubing, with only the Eheim 1046 or 1048 pumps. Oh, and the blocks had large amounts of surface area (unlike the Maze-3).

It makes sense that the best performing blocks with smaller pumps would have smaller fittings. This increases water velocity as it goes into the block and improves heat transfer from lowering the size of the boundary layer. That doesnt mean that these blocks are universally best, however, as other blocks may be designed for bigger pumps and more pressure.

That was exactly my point with Birrman. I see no mention of temperature measuring, number of replicates, testing parameters.

I have a few blocks here I will be testing too:

Center Inlet "maze type"
Maze2
Spir@l
sysfx dual channel

Die Area Impingement
Swiftech MCW462U
Swiftech MCW462UH

Others
Innovatek rev3
Heatkiller 1.5

That really doesn't say much about the quality of the final results though does it?

Yea science and proper analysis procedures break down ALL the time. Perhaps we REALLY need to be asking our ministers which blocks to use?

Your example is actually consistent with my statements as well. Inproper testing methods don't always yield useful results. Please find me an example of where thermodynamics breaks down. An example (from the life sciences) that would be more consistent with what I am talking about would be the observation that pharmaceutical development times and efficiency have been dramatically reduced since the 80s when biochemistry really became a popular science. Chemists and biologists began to elucidate enzymatic pathways and fundamentally understood how they work and where they are going wrong in sick people. This let them target particular pathways with drug therapy, and greatly accelerate the whole R&D process. The same thing is needed here. You will never understand heat transfer and fluid movement and the important design variables with your approach. Without that understanding the improvement of overall systems will be slow and fraught with misinformation (lets just inject this into rats and see what happens?).

A few weeks of thought in designing an experiment can save you a few months of testing; figure out what your questions really are and then focus your studies on getting to that. To me, the goal would be to educate readers in heat transfer and to provide test results that are useful in mixing and matching a good performing complete setup or in building a good custom block. You can't get there (easily) by testing just a whole bunch of complete kits; there are too many variables. Consider the variable pressure production of different pumps of approximately the same GPH rating. This will influence the performance of the setup and in particular the need to move to larger fittings to try and lower the resistance and get higher flow to a block. If you have pressure drop information on the blocks as well as temp, then you can try to match a block to the pump you are using. Many people (myself included) like to cool in smaller cases and so pump choices are limited. Case size also limits radiator choice somewhat and that will also affect block choice. I choose fans based upon manufacturer supplied curves all the time without needing to put them in my case first to "test" them. I don't understand why water cooling equipment is any different, other than the fact that the manufacturers don't (can't? won't?) provide that data so testers need to generate it themselves.

I can see that my comments have kept a few people up late...

Powerhouse, I think you've got a couple of misconceptions.
(so I'll list what I see, and you tell me if I'm wrong!)
1-Although water temperature is a factor in the performance of a waterblock, controlling that temperature is not something that is a part of a common water cooling setup (i.e. there is no chiller/heater). As a result, water temperature is irrelevant, for the purpose of testing a waterblock. If you don't agree, then look at it this way: if a block's performance was (theoretically) optimal with a coolant temp of 15C, what could you really do with that information?

Also, in your reply to pHaestus, you (essentially) stated that temperature should be tested as a function of flow rate, when really, it's the other way around. Water temp is not relevant, it's just a consequence.

In the same reply, I don't think you quite understood that the test of a WB would be performed at different flow rate with a fixed (read controlled) water temp (wether or not there is a rad is irrelevant), for the purpose of finding its optimal cooling setup. Selecting a rad is a separate issue, which makes my case for matching components.

Your frying pan analogy is funny, but doesn't quite match the number of factors involved in water-cooling. But just for fun: what if your frying pan had a thickness of 10mm? What if it was cast iron, instead of Aluminium? What if the stove was turned on 7 instead of HIGH? What if you matched a heatsink with a fan providing it with its optimal air flow rate? I think it would be of interest...

2-The heat dissipation properties of a waterblock do NOT change with the temperature of the coolant. The only situation where this statement would not be true, is if the temperatures would change the physical state/properties of the materials involved (i.e. tygon melting at 80C, water boiling at 100C, copper turning into liquid at 2500C, etc... I know that it's hard to believe.

Now I'll answer some questions/post comments...

PH: you suggested "comparing the difference with the new CPU and adjust all you numbers accordingly". Now you're talking about a lot of expensive gear!!! Keep in mind that this gear CANNOT change from one test to the other, so if something fails, it would have to be replaced with EXACTLY the same components, which may or may not be available anymore. Also, some gear fit for an old Duron, may not even be appropriate for an Athlon Xp, and vice versa.

PH: about load testing low/medium/high
Unnecessary, and too time consuming. it would be far simpler and more accurate to measure the performance of each component individually under ONE typical environment.

About reviews/tests: if all you're going to do is ignore the numbers (which is fine) and take several people's opinions on which is best, then that'll work well, for air cooling. For water cooling, there are several components available together or seperately, but the best of each will not yield the best results.

About the time it takes to do the tests:
I would trust a test if the testing methods were consistant, and gave me usefull numbers, REGARDLESS of how long it took. The method/results speak for themselves. How long it took to get those results is IRRELEVANT.

About pumps:
This is one point that you do have, in that tests don't match real life. e.g. if an optimal flow rate has been established at Xgpm for a particular configuration, there may or may not be a pump available to reproduce the exact flow rate. Pumps do perform differently based on the flow resistance of a rig, but this can be equated with the "head" rating of the pump, and so, you already have all the data you need. (actually you'd still need the net flow resistance of a particular setup, see below)

About rads:
as the OC review suggested, there may be a "sweet spot" to most rads. Wouldn't it be nice to know what they are for each rad? Or at least how to find it?

About different rigs:
My point is that you may be able to reach the same results, with a completely different rig.

About results:
There are a number of conditions that would affect an actual rig, and those include: tubing length (not size), actual volume of coolant, coolant additives, number of elbows, ambiant temperature AND humidity, placement of radiator, effective rad fan flow rate, etc... but those are simply beyond anyone's control. Knowing that though, basic recommendations can be made, but otherwise, all those factors add up to establish THE ACCEPTABLE MARGIN OF ERROR of any test. Sure 0.1C temp testing accuracy would be nice, but is it necessary?

About the Maze-3:
It may not be a significant factor, but the interior is sandblasted, giving it a higher surface area.

pHaestus:
Your testing rig sounds very nice. The importance will be with your testing methods, as they have to remain consistant. It may be atypical, but a typical person would not get a flowmeter, pressure gauge, and many temp sensors, they'd just use the same gear, but straight out.

pHaestus: I like your graph... It'd be nice to see the data on a wider range though (take those results, determine a range that is of interest, then exceed it at both ends by 25%). so 3 GPM (180 GPH) would be pretty good for a MCW 462U, huh?

myv65:
You've nailed it right on the head. There is a temperature point where the wb will not cool under, regardless of flow rate. What I'm wondering is, if there's a sweet spot for rads, could there be one for WB?

About this discussion: there is one very important factor that has barely shown up:
MONEY!!!
In any company, when a project comes up, there is a cost study made, for the "best bang for the buck". Here, the consumer is the company...
My point is this: you could spend an outrageous amount of money on a water cooling rig, but you may never be able to sell it, simply because it's not cost efficient.

Would anyone really buy a solid silver block, with 3/4 inch barbs?

Wow I saw the references to PH in that post and was quite confused: I don't agree with that? Did I say that? Then I realized PH = Powerhouse not pHaestus. I think that you and I are basically in agreement that you should try to learn something useful from the testing or else it doesn't have very much point.

For many pieces of information, 0.1C accuracy really is needed. The error propagation otherwise gets really out of hand. You see differences, but if they fall within the overall error they aren't necessarily real. With a water temp and CPU temp probe that are both +/-3C, you need some BIG differences to spot. That isn't even accounting for the compression of most people's CPU temperature due to how they measure it.

The MCW462U data is Bill's not mine. I purchased a much larger (industrial) pump to counteract the restriction of the flowmeter and allow me to generate flow curves at much higher GPM. The curve I have there is the most the Eheim 1250 could accomplish :)

The issue becomes the amount of heat the pump throws into the water when throttled back, and properly accounting for that (with good water inlet temperature measurements). That is why I want to move to a controlled water bath instead of a radiator.

The issue for testing is also very much one of money. Especially the accurate temp monitoring and the production of a good die simulator. Look to spend LOTS for accurate results, and then contend with the problem that the average end user just wanted you to give them a "clear winner" and a lot of pictures of the blocks and really isn't interested in the real information that is in the tests.