I'm sure some of you may have stumbled across some of BillA's or others' work and thought "WTF is all this crap...where are the CPU temps?" Well, what they've done is given the reader the tools to make better informed choices when buying water cooling products than typical system testing can provide and a method which doesn't require comparison between waterblocks on the same CPU/motherboard and under similar conditions. Hopefully this post will help those who were overwhelmed by the technobabble or didn't know what to make of all the info. IÂ’ll try to keep it as simple and on-topic as possible since it's easy to get bogged down by terminology, especially when dealing with pumps. If your eyes start to glaze over after the first few paragraphs, just wait for the charts and itÂ’ll all make sense pretty quick.

Some terminology to get started:

Flow rate – We’ll be talking about volumetric flow rate expressed as the volume of liquid passing a certain point in a unit of time. Can also be defined as the velocity of the fluid multiplied by the cross-sectional area perpendicular to the direction of flow at the point of measurement for incompressible fluids. (Expressed in LPM or GPH in these graphs). An example: pretend we have a loop full of 1 gallon of water. If we pick a point anywhere on the loop and time how long it takes this amount of water to make one complete revolution, we'll get the flow rate. Lets just say that it's 5 minutes....1gallon / 5minutes=.2 gallons per minute or 12 gallons per hour.

Capacity – The flow rate at which a pump pushes a liquid to a specified point. Typically just used interchangeably with flow rate in everyday forum-speak.

Total dynamic head (or simply “head”) – Can be thought of as the height of a column of liquid caused by the push of the pump. This is often used in the place of pressure for pumps because the pressure from a pump changes with the density of the liquid, while head does not. (Expressed in units of length, in this case feet and meters.)

Head loss – The head required to overcome all the resistance in the loop caused by friction between the liquid and walls of the components, turbulence, changes in direction of fluid flow, change in cross-sectional area, etc… Sometimes used interchangeably with pressure drop in common forum speak sincec most of us just use water, but just keep the above in mind.

Maximum head or shut-off head – The point of zero flow. (The pump performance graph will intersect the y-axis)

Delta T (∆T) – Simply refers to a temperature difference or change in temperature between two points of reference, in our case the CPU and water at the inlet of the waterblock.

First some pump data sheets:



THE Y-AXIS LISTS HEAD (OR PRESSURE) AND THE X-AXIS LISTS FLOW RATE (OR CAPACITY)



IÂ’ll post the metric equivalents below as thumbnails. One of the reasons I started thisÂ….got tired of constantly converting between units.

These are referred to as pump performance, or PQ graphs (P=pressure vs. Q=capacity). Centrifugal pumps impart kinetic energy to liquids through their rotating shafts. Think back to the column of water. If the column of water is held completely vertical, max head would be the height of the column and there is 0 flow. If we did the opposite and held the pipe horizontal (at the same elevation as the pump) we would get the max flow, assuming the pipe was frictionless. If you can imagine a straight pipe with water pivoted at the origin of the graph and the tip of it moving along the curve we can visualize the relationship between head and flow rate. As the elevation increases the weight of the column against the pump increases until the kinetic energy bestowed on the liquid by the pump is balanced by the pressure energy of the water and max head is achieved.



Back in the real world, the pump actually operates at some point between these two along the curve, determined by the resistance in the loop. Specifically this point of operation is at the point where the pump performance curve and total system resistance curve intersect (see above). This point of intersection determines the flow rate of the entire loop and the flow rate is the same at any point in the loop. If it weren't it would mean that we'd be compressing the liquid.

YouÂ’ll notice from the total system resistance curve that pressure varies with flow rate in a parabolic relationship. This isnÂ’t a coincidence since the relation is defined by BernoulliÂ’s equation where we determine that pressure is related to the square of the flow. One immediate implication of this fact is that in order to double flow rate, we must quadruple pressure.



Backtracking just a tad, the total system curve is determined by the algebraic sum of the head loss curves of each individual component of the loop. This is demonstrated in the above example. If we have an MCW6000 water block with 7 feet of 3/8" tubing then the total head loss at each flow rate is simply the sum of the head loss of each component at each flow rate. An analogy to this is resistors in series in electronics. If it's easier for you to visualize, think of adding the distance between the x-axis and each curve (the head loss) and adding these values to get the total head loss. You're basically stacking the area under each component curve on top of each other, and that results in the total system resistance curve. Since we're only adding and subtracting, this makes manipulating these curves very easy and versatile (more on this later).



Now we expand our example of the point of operation to multiple pumps with a given system. The ****ty red ovals are the points of operation with each pump. Taking a closer look at the intersecting of our system with the MCP650 in particular, the vertical and horizontal dashed lines drawn as a guide indicate the flow rate and head loss of our system respectively. Higher flow rates are more desirable for any waterblock (wonÂ’t cover heat transfer here), but something which also must be taken into consideration when choosing a pump in the performancec arena, and something I wonÂ’t mention after this sentence, is the heat dumped into the water by the pump. That aside we can see that the "best" pump for this setup is the AquaXtreme50Z, since it will give us the point of operation furthest to the right, corresponding to the greatest flow rate. Now lets change things up a bitÂ….



If our system now had a resistance curve like the one above, the greatest flow rate now occurs with the MCP650 (red circle once again represents this "point of operation").

Some curve analsis: When comparing two pumps, the “critical point”, so named because I thought it made it sound dire, is the point of intersection of 2 pump performance curves. Typically they will only intersect once unless, of course, you have a funky one like the ViaAqua or Mag3. Depending on which side of this point our system curve lies on, one pump will be favored over the other. Compare the results from the graph above and the one before that and you’ll notice the system curve fell on different sides of the “critical point” of the MCP650 and AquaXtreme50Z curves. If two pump curves never intersect, the one of greater radius will be superior at all flow rates. Also, since the system curves are parabolic, if we have a rat’s nest of pump curves, the one giving the greatest flow rate will simply be the one highest on the system curve.

A quick qualitative observation: The greater a systemÂ’s resistance is, the more likely it is that a pump with a steep curve, such as the MCP350, which has high head and low flow compared with the others, will be favored. Similarly, if the system resistance curve is relatively shallow (little resistance) a high-flow pump may be more desirable.

Now we know what pump choice to make if we already have a water block in mind, and given different restrictions in the loop, but what if we want to compare the performance of different water blocks? In that case we must introduce a third graph of ∆T (CPU temp - water temp at the water block inlet) vs. flow rate:



Here we have the whole shebang. The ∆T intervals looks slapped on there because they were. First, I think it looks a little too busy with all 3 types on there and because intersection between ∆T curves and total system curves and pump curves may mislead the novice. Just think of the ∆T curves seperately.

On with interpretation. We know how to determine flow rates of systems given their resistance curves and pump performance curves, now we need to use this information to predict the performance of our water block. In order to do this we simply note the flow rate of our total system as determined by the intersection with the desired pump curve and check the ∆T at the same flow rate. The lower the ∆T, the better. LetÂ’s take the NexXxoS with a CSP750 as an example. The point of intersection is at a flow rate of ~46 GPH. Now we check this value against the ∆T graph (the black one) and find a ∆T of ~9.7°C (IÂ’ve drawn vertical black lines coinciding with this flow rate to help visualize this relationship). Now letÂ’s compare this value to that of the MCW6000 with a MCP650 pump, which gives us a system flow rate of ~115 GPM. Checking the **** yellow curve at this flow rate yields a value of about 9.5°C. A MCW6000 + MCP650 combo will outperform a NexXxoS + CSP750 on this test bench, even though the ∆T vs. flow rate curve of the NexXxos is below that of the MCW6000 at all flow rates (remember lower ∆T is better). The NexXxoS will be superior with any other pump on that chart, however.

Some quick analysis of the ∆T vs. flow rate graphs: Here we can see the truth to the earlier statement about higher flow rates and water block performance. The ∆T graph is always decreasing, which means, the greater the flow rate, the smaller the ∆T. Furthermore, by looking at how much an incremental increase in flow rate impacts the ∆T we can make predictions as to how much a larger pump will affect waterblock performance (those who've taken calculus will notice this rate of change is defined as the derivative of the ∆T curve). If the ∆T decreases significantly proportional to flow rate (steep slope) a larger pump may be a worthwhile investment. Conversely, if the ∆T vs. flow rate graph is shallow, a larger pump will have little benefit and may actually have the opposite affect, since larger pumps dump more heat into the loop (Doh!). The ∆T vs. flow rate graphs for every waterblock become more and more shallow as one progresses along the curve....the greatest impact on water block performance is therefore seen at low flow rates. Both of these situations are embodied in the NexXxoS and MCW6000 ∆T vs. flow rate graphs shown above. From 60GPH to 135 GPH the MCW6000 only sees an improvement of about 1°C, the NexXxos, on the other hand, sees an improvement of 2°C from 30GPH to 75GPH. These facts are what typically cause people to classify blocks as either "high flow" or "low flow". Finally, the length of the curve on the Procooling site can give you some idea of how restrictive a particular block is. The NexXxoS is very restrictive.

∆T vs. flow rate graphs can be found here:
http://www.procooling.com/html/pro_testing.php

A quick cautionary note about these valuesÂ…they were obtained using a ~71W T-Bred discussed here: http://www.procooling.com/articles/h...ethods_-_p.php
CPU cores of different sizes and heat densities may yield different results.

**QUICK SUMMARY:
1.) Pick what components you want to include in your system.
2.) Find the flow rate at which the system restriction curve intersects the curve of the pump you want.
3.) Evaluate your water block's performance at that flow rate on it's ∆T vs. flow rate curve.
That's the jist of what I've spent about 2500 words saying. If you can do that....you're golden.

You may now be thinking, “Great, now where do I find all the head loss vs. flow rate data for all my components so that all my newly acquired knowledge can be put to use?” Catch-22. There is no one source where you can find all these figures. The reason I have them is because I’m a nerd who gets off on water cooling science so I have a habit of saving or bookmarking any graphs/webpages of interest. My original intent was simply to explain the method above, so I compiled the pump curves from manufacturer data sheets, but then decided I’d like to check the pressure drops of the water blocks I had data for…then why not include all the components in a single loop? Anyway, I went overboard as usual and I wrote a program to calculate the PQ curves of a combination of up to 5 components in series for 3 different loops and plot them against the pump performance curves to get the total system flow rates. If anyone would like the file, they have but to ask….but you’ll need Matlab to run it. Regrettably I don’t know Excel or HTML. Either post here or PM me and I'll be happy to develop the graphs you need and offer help with interpretation (as will others I'm sure)

Here is what I can model with relative accuracy at the moment:

Any size and length of tubing
Any size fittings
NexXxoS XP
G4
Cascade
LRWhiteWater
MCW6000
ThermoChill 80.1
ThermoChill 120.1
ThermoChill 120.2
ThermoChill 120.3
Â’86 Chevette HC
Fedco# 2-234 Single-pass HC
Any heater core of 2” depth (sim)
Dtek ProCore (sim)
DanDen 120x1 (sim)
DanDen 120x2 (sim)
Most popular chipset blocks (sim)

Anything that I didnÂ’t find hard data on I put (sim) next to, however, IÂ’m quite confident in the heater core data. The chipset blocks (the current popular ones) produce very little head loss, so even a simple linear graph has a miniscule impact on the final system curve's margin of error. This covers a surprisingly large area of the field, however, there are a few popular items I feel reluctant to put on the above list:
RBX
TDX
Any Black Ice rad
GPU blocks

The BI rads are similar to their TC counterparts, but their dimensions vary. If you need a ballpark I’d use the TCs. For the TDX and GPU blocks I only had 2 data points (0,0 and JoeC's), which doesn't make for anything better than a line. I’ve actually developed a pretty decent sim of the RBX (#1 nozzle) posted below. The TDX…..not so much. It has the lowest pressure drop of all the blocks on the list by far…so there is nothing close enough to model after. This is one of the most popular recommendations on these forums so I spent quite some time scouring for data and trying to make a workable sim, but when it’s all said and done, it’s nothing more than a “good guess” and the margin of error is significant. I haven’t worked on the GPU blocks yet but think they’re doable for the same reason as the chipset blocks. Still don’t feel comfortable putting them on the “approved list” though.

Hope this proves enlightening for some of you. Questions / comments / corrections all welcome. Also any requests for modeling dataÂ…it only takes me about 20 seconds to throw it together.....so ask away.

P.S.> Almost forgotÂ….just wanted to thank all the guys who spend their time testing and posting their findings for the good of the WCing community. Much appreciated.

Attached Thumbnails

     

     

     

Dang Anon, finally some one went ahead and explained this. I have explained this to so many people because there was no guide for it. You get major Rep from, and everyone that reads this needs to give you bling too.

Nice work, Anon I'm going to bookmark this thread & read it properly later ... it's too early in the morning for anything more scientific than newsbots posts about cheesy poofs

heh ya, nice post there Anon! now i just have to wake up a little bit and read it over again, besides looking at the pretty pictures heh i'll throw ya out some Rep! =D

Alright, made alot of changes / revisions and addendums. I tried to explain things a little easier and included the quick summary towards the end. If you can keep those three steps in mind, the rest is mostly expansion on those points.

And as promised.....some simulated RBX and TDX data with WW data for reference:



To get the RBX curves:
1.) Manipulated the quadratic equation fit to the Whitewater data until I felt I had a nice curve. Basically all I used was intuition and pattern recognition.....also known as guessing.
2.) Took the derivatives of the quadratic equations for all the blocks and found a slope that I thought was suitable...integrated. AKA wingin' it.
3.) Multiplied the Whitewater data by a constant to shift the entire curve to the data point @ 60GPH obtained from OCers. (Whitewater since it's the closest curve and has a similar internal design.)
4.) Used a few actual data points.
5.) Took the average of all 4 (RBX on the graph).
To my pleasant surprise they were all very close. I'm curious how close I came to reality.

To get the TDX curves:
For this one I didn't even bother with #'s1 and 2....simply not close to guesstimate. Instead, I shifted the WW data as in (4) above, and did the same with the 1/2" tubing data by multiplying it by a constant to get the actual value at 60GPH. Why the tubing data? I first did the same with all the other water blocks to see how those curves would compare and the adjusted tubing curve was lower than the actual water block curve in every situation, quite substantially towards the end of the curve for some blocks. It's only logical to assume the same would be true for the TDX curve. For this reason it makes an approximate lower bound for the possible TDX curves and a "best case" scenario for the least head loss of the TDX. I used the adjusted WW data as the upper bound and the "worst case scenario". Took the average of the 2 as you can see on the graph.

Keep in mind this method is shoddy at best, but the best thing I can come up with given only 1 data point. If any of you TDX users are interested in doing a crude "bucket test" the next time you dismantle your system to give me some more data points, it would be much appreciated. Not the most controlled measure, I know, but since we're guessing anyway.....

I'll also post some full system analysis in a minute.

Attached Thumbnails

 

System1 = MCW6000, 7ft 3/8" tubing, Thermochill 120.2 with 8.46mm ID fittings (~.33")....assuming you stretch it over them.
System2 = LRWhitewater, 7ft 1/2" tubing, Thermochill 120.2 with 12.83mm ID fittings (~.5)....assuming you stretch the tubing over.
System3 = TDX sim (the avg. from above), 7ft 1/2" tubing, Thermochill 120.2 with 12.83mm ID fittings (~.5)....assuming you stretch the tubing over.
System4 = TDX sim (the "lower bound" from above), 7ft 1/2" tubing, Thermochill 120.2 with 12.83mm ID fittings (~.5)....assuming you stretch the tubing over.

Here is the graph:


∆T of TDX loop with MCP650 (DD D4) = 10.2C
∆T of "lower bound" TDX loop with MCP650 = 10.1C
∆T of Whitewater loop with MCP650 = 9.5C (9.7C for Dtek WW assuming the head losses are = for both WW's, although they aren't in reality and vary a little)
∆T MCW6000 with MCP650 = 9.9
∆T MCW6000 with CSP750 = 10.2C

What does this mean? According to the numbers the performance of the TDX with the highest flow rate pump on the list is matched by the MCW6000 with the weakest...in a more restrictive loop. Some food for thought. Increasing the restrictions in both loops only seems to favor the MCW6000 even more.

Note on fittings: I still need to go through some testing methodology for info on the fittings used on each of these water blocks (or if their restriction was already subtracted) so that may alter results a little. May also need to make some adjustments to the radiator fittings as posted above.

EDIT: Included "best case scenario" TDX curve and data.

Attached Thumbnails

 

An excellent summary for anybody who hasn't taken a collegiate-level fluids class on how to judge water cooling performance.

Another thing would be worth posting is a comparison of the thermal exchange surface area between different waterblocks and the subsequent effect upon performance. In the absence of varying flow restriction, it's a very important characteristic without having to get involved in examining the ability of the exchange surface to disrupt the laminar flow layer in order to produce sufficient turbulence for effective heat exchange to the fluid.

.......................
............
sounds good

I can't believe this is not a sticky.

The author of this thread deserves great praise for his work.

Impressive is to weak a term for this.......

Thanks .this will take more than one read.Youve given us yet more priceless knowledge.Your hard work deserves praise and rep.