The velocity at which it strikes the baseplate stays the same. Only after it strikes the plate does velocity drop because of surface geometry, friction, surface tension(think the last is correct). In the case of submerged, surrounding water effects velocity also which I found out the hard way. As you get further out in diameter from the stagnation area and up to the ouside diameter of the impingement wall the greater heat transfer. So the base plate in it self can effect the results of the impingement.

This is the question I keep running through my mind. If there is micro-channels or any surface geometry that causes enough pressure drop, will it render the jet impingement useless? I'm leaning towards the answer of "yes".

I have seen those. Their should be another about ellipse. If you find that one I like to have it. I lost it some where between my links, papers and a kid.

I seen one about how the angle at which the ellipse jet was orientated to the plate performed well.

As a guy who somewhat surprisingly doesn't know the full theory behind it (the 90% rule here) my answer goes like this.

For a fine-channel (I say fine - because the term "micro" is open to subjective interpretation) setup given a particular channel width, I believe that there is a pretty tight range of values that specifies the "optimal" fin width and height. This implies a direct relationship between channel width and fin height.

Now given a particular range of operable pumping pressures, say 2-4PSI for the pumps that people typically use, this then (in my mind) places a lower bound on the usable channel width before the pressure drop becomes so high that the pump can't pump enough appreciable volume through the channels, and this will result in a performance loss as the water itself heats up too much because there is not enough heat mass in the liquid.

What I'm saying is that problem is self-balancing. The point at which the jet impingement would begin to suffer would also be the point at which the micro-channel implementation would begin to suffer as well, that is, it is possible to make the channel so fine and restrictive to "damage" the jet impingement action, but in doing so you will be losing performance in other places as well, so the problem as you describe is self-defeating given a correctly ratioed fin/channel width/height micro-channel system.

The question then becomes one of "what is then the smallest micro-channel dimensions that you can use effectively with a 2-4PSI pump". I already have a fair idea of the answer to that question, but it is also tempered with the need for secondary blocks in the system to not have their performance hampered in a significant fashion by a singular restrictive block in the system, so the answer suddenly takes on a new perspective.

I have an idea as well but no experience to back it up. I failed to realize the thought of extra blocks in the system. Thanks for the new sudden perspective.:eek:

I fully understand what you saying. It is a delicate balancing act and finding that balance is the key without the expense of effecting something else in the chain. Atleast, not to much. You ever try something with a 50 PSI and decent flow rate? To bad I can't find a pump like that.:D

I'm still going to try the total impingement design. If anything to learn even if it doesn't perform as expected or as I thought.

If I thought that just dissipating power in a waterblock would cool my CPU, I'd simply set up my watercooling system out in the garage and wouldn't bother mounting the waterblock on the CPU.

You think I'm suggesting that increasing the power dissipation in a water block is good, in and of itself? Not even close.

The fact of the matter is, for any given block, there is a relationship between the hydraulic power applied to the block, and the thermal resistance of the block. Within limits, the thermal resistance of the block decreases as more hydraulic power is applied. Because a waterblock is part of a system including (usually) a centrifigual pump, considering issues related to transferring power from the pump to the block is important. (It's not necessary to look at the system in terms of power transfer to determine how well it will cool, there may not even be any particular benefit to doing so, it's just a viewpoint that comes 'naturally' to me, and I find it interesting to consider watercooling systems in these terms.)

Because I'm an electrical engineer, I tend to think about a lot of this stuff in electrical analogies. One tool from electrical engineering that appears somewhat relevant is The Maximum Power Transfer Theorem. In short, it says that if I have a voltage source with a resistor permanently attached to one of its output leads, then to get the maximum power into the load, I need to connect a load with the same resistance as the permanently attached resistor.

This would be more directly applicable to watercooling systems if flow resistances behaved akin to Ohm's Law or:

dP = Rf * Q

Where:
dP is pressure drop
Rf is flow resistance
Q is flow rate

Instead, flow resistances generally behave as:

dP = Rf * Q^2

Because centrifugal pumps behave somewhat like ideal pressure sources in series with an inherent flow resistance, a modified version of the Maximum Power Transfer Theorem may be applicable enough to be useful. From the very limited number crunching I've done so far, it appears that the maximum power will be transferred to the load, when the load resistance is twice the inherent resistance of the (idealized) pump. I'll let someone who's had a math class much more recently than I, attempt to prove or disprove this.

The following graph shows the PQ curve for an Eheim 1048 pump as well as the following equation as some indication of how accurate a simple simulation of a pump might be.

dP = 1.5 - 0.015 * Q^2 (1.5 is the max head, and 0.015 is the inherent flow resistance)

http://uffish-thought.net/wc-gifs/eheim-1048.gif

It's getting too late at night to go on with this now. Some examples of looking at watercooling systems in terms of power transfer tomorrow night.

Only if the flowrate remains constant. You would find that as the gap became smaller, the flow would also drop. The generic answer is that "downstream" stuff affects "upstream" stuff if the flow velocity is less than the speed of sound within the medium. Our velocities are much, much less than the speed of sound in water. So the question becomes, "How big must the gap be given the jet size, speed, etc., before changing the gap has no appreciable effect?" I suspect the answer would be single-digit millimeters, but that's merely a hunch.

Maybe this is a boob thing to say, but here goes:

Isn't the benefit of the nozzle realized in the increased flow velocity over the point in the block where most of the heat is concentrated, thereby allowing more efficient heat transfer in the liquid? In other words, the coldest fluid is pounded against the hottest surface to get maximum heat transfer, while still allowing the exiting water to absorb some more heat from the rest of the block (where the microchannel portion comes into play, increasing the surface area thereby making heat absorbtion easier).

The jet allows you to replace the fluid on the "hot spot" at a faster clip, and turbulates the coolant directly on that spot. Without the nozzle, you would have more of the coolant slipping across the top of the channel, not cooling the hot spot as effectively, requiring it to spill more heat laterally across the block before it is absorbed rather than allowing a larger portion of it to be immediately absorbed into the coolant.

Microchannels are nice, especially at distributed heat transfer, and jet impingement is nice to reduce the necessary size of the block. I think the effective use of both is why Cathar's block works so well.

Yes...

In order to improve the heat transfer to the coolant, we have to try to reach a high Reynolds number (aka turbulent flow). Since 99.9% of us can't do that, because it requires a veri high pressure pump, we use jet inpingement to "throw" the coolant at the baseplate, which results in a turbulent area, similar to the above. Turbulent flow, without the high pressure pump.

Add fins, whose purpose is to spread the heat further, from the baseplate, and you'll reduce the resulting CPU's temp.

Combine both (above) and you've got an ideal waterblock. That's what Cathar did, yes.

But it can still be improved:D

This is true, only if flow rate is constant and concure that the smaller the gap the less flow. I done tried it. Your hunch is right though. With our low pressure pumps, anywhere between 2 - 6mm depending on the size of the jet being used.

Couple of interesting thoughts. How much of an effect the angle of the jet and the actual shape of the base plate play a role? It's quite surprising.

Another thing to consider here is the width of the jetted area.

The WW design doesn't really care how wide a heat source is that straddles the fins. It effectively "slices" that heat source up, and treats it as a series of elongated strips that run the length of the channels. This is a simplistic view of things, but given the thin base-plate, isn't too bad of an approximation. This basically means that a heat source under a channel can be effectively viewed as a 8-12mm long strip that needs to be cooled. This means that our useful jet impingement region needs to hopefully be around the 12mm wide mark within the channel.

Now myv65 was onto something when he talks about what I call diminishing returns with respect to making the nozzle smaller than a small number of millimeters. For a 3PSI pump like an Eheim 1250, the smaller the nozzle is made, the higher the water velocity, but to a point. Eventually you reach a stage where the water velocity goes up VERY slowly as the nozzle is decreased in size. A bit like plotting a hyperbola.

Now there are a few things going on here that need to be balanced. As the nozzle is reduced, so does volumetric flow. It's pretty easy to plot the raw C/W of water vs volumetric flow. Short of phase changing the water, it's impossible to get a better C/W than that line - you can only ever hope to be as close to it as possible. So as we reduce volumetric flow rate through a smaller nozzle, we can reach a point where the volumetric flow decrease overwhelms the gain from the marginally higher water velocity.

There's a second effect that happens here too. The width of the nozzle also dictates the width of the jet impingement region. The higher the volumetric flow rate, the wider the jet region. Also, the wider the nozzle, the wider the region - to an extent - if the nozzle is made too wide then no real jet region forms and the water just mashes out the side and never really hits the base hard where we want it.

So here it can be seen that there is again a fine balance point. I spent a lot of time, theory and practise behind the scenes away from any public discussions tracking down what I believed to be the optimal nozzle width for standard pump applications with the White Water.

It's easy to think that making the nozzle smaller will boost velocity and hence performance, but the reality is that it's not that simple. There are many things to consider here, and it's not all that obvious at first glance.

Depends on what your block design is and it's surface geometry. Throw in micro channels and too small of a nozzle size will kill the flow rate to the point your channels no longer function as intended. Your right. Can only make the nozzle diameter so small till the point that velocity levels off and then drops along with flow rate.

I found that out also. Making it smaller will just kill the flow rate to the point the impingement is useless. Shape of the nozzle effects the impingement region also which I find interesting. The circlular impingement is actually the worst compared to an ellipse, triangle, the square in that order. Have to consider the shape to work with in your design and take advantage of it. There is alot of variables to keep in mind. One will effect the other and there is a balance in it all. Impingement by it's self I believe can be very promising.

Totally agreed - the question I guess is how promising it is given 3PSI pressures that people use?

Given 50PSI pressures, it's definitely the way to go.

The old theory vs practicality thing. Then again, I guess that if we all had 50PSI pumps, then we'd be looking for more still.

Ever considered using one of those high-pressure jet-sprayers that are used for cleaning mud and crap off cars and engines? I've always been curious if the rapid pulsing effect of the flow would create favorable turbulence by rapidly increasing and decreasing the jet impingement radius, and the pumps can typically push up to 600PSI or so. It'd be a "cheapish", if rather noisy, way to conduct the experiment, and when you're done with it, you get to clean your car too.

I think you can see an example of this in BillA's MCW462-H vs MCW462-UH data.

One can compare the thermal resistance of the two blocks at operating points with equal inlet nozzle velocity. Even though the MCW462-UH is has a greater volumetric (and mass) flowrate at the same inlet velocity, the MCW462-H still has lower thermal resistance.

and there is also data on different internal surfaces in the 462-B modding article

Or you could use one of them sewage line cleaners:D Just don't get your hand in the way or it's a quick amputation.

I have gotten comparable results with impingement as micro-channels would with atleast 3 PSI. Atleast with 1mm channels anyways. I have no idea how true micro-channels(as the name applies) would do. But I have to stop and think if I'm effecting another block in the cooling loop that might be there.

Do I think impingement can out do a micro-channel block with just 3PSI? No I don't. I believe it would be an even call. 50 PSI? Sure but it isn't practical.

Never considered that. Some hefty hose clamps and high pressure hose would be in order. LOL I did hook up my garden hose to my block without impingement and gave it try. The ground water always stays roughly 23C here. Just 3C under what my water temp is now. I was amazed how well it did. Now you got me wanting to try it again with impingement. I enjoy your insight Cathar! You got me to realize a few points I didn't think about.

More on power dissipation.
 

Continuing where I left off...

Having an equation to approximate an Eheim 1048 PQ curve makes it easy to generate some more graphs in Excel.

The following graph shows aspects of a simulated watercooling system composed of a DTEK TC-4, a Big Momma Heatercore, and an Eheim 1048. (Tubing, fittings, etc. neglected blah, blah, blah)

http://uffish-thought.net/wc-gifs/tc-4-pow.gif

Drawing a line down from the intersection of the yellow line and dark blue line, (intersection of loop PQ and pump PQ) it can be seen that the flowrate through this system will be about 5lpm, and the power transferred from the pump to the loop will be about 0.92 Watts. (Out of about 0.94 Watts maximum pump output.)

Because the system pressure drop is split about 50/50 between the TC-4 and the Big Momma, the power dissipation in the block itself will be about 0.46 Watts. From my first chart it can be seen that at 0.46 Watts dissipation the thermal resistance of the TC-4 will be about 0.226 C/W.

The following chart is the same, except with a MCW462-UH simulated in place of the the TC-4.

http://uffish-thought.net/wc-gifs/462-pow.gif

In this case it can be seen that the flowrate is a bit higher. (6.4 lpm instead of 5 lpm) The power dissipated in the loop is about the same. (0.93 Watts instead of 0.92 Watts)

The power dissipated in the block is a much smaller fraction of the total power delivered by the pump. (About 5.5% or 0.05 Watts) Looking at my first graph, the thermal resistance of the MCW462-UH in this system would be about 0.246 C/W.

So is there any particular value to doing watercooling system performance calculations in terms of power consumption? None that I can think of. The data Bill provides yields all these results without any need to calculate a power consumption.

It's been interesting to me to look at this stuff from this perspective though. I've got a lot better feel for how the parts of the system interact.

Hmm... Starting to think that Dodge Viper's putting those extra barbs on the wrong end of the heatercore.