The "Cascade" - mini-cup directed jet-impingement block design (56K warn)

[hara]

Should be interesting how such a block performs, I'd assume worse but maybe less difficult to manufacture.

[utabintarbo]

Originally posted by cristoff
what about this?

Again, no. Please refer to bigben2k's previous post. Same problems, same result.

Bob

[Balinju]

Originally posted by bigben2k
Hummm... no.

The jet would be forced around the pin, reducing the inpingement effect. It's the same problem as with my Radius.

It would also be quite difficult to manufacture.

[MadDogMe]

The pipies melted.... So that just tells me I will have to just drill the holes out instead.

You have to drill them out after?, I'd imagine if you tried to lazer them that thin (with the hole predrilled) the sides would melt as well?. What's it like to drill polycarb with walls that thin?, I thought it cracked/threw off lumps for a pastime?...

[jaydee]

Originally posted by MadDogMe
You have to drill them out after?, I'd imagine if you tried to lazer them that thin (with the hole predrilled) the sides would melt as well?. What's it like to drill polycarb with walls that thin?, I thought it cracked/threw off lumps for a pastime?...

Yeah make the pipes but without the hole and drill them out. Use a good sharp bit and it is all good. I drill acrylic all the time at work. We make all kinds of acrylic laser engraved stuff. Laser is to hot, it just melts the sides when they are that thin. Also doesn't look to good anyway. A good endmill will make much nicer milling job. But the laser rules all on making larger holes and cutting out the acrylic. Leaves a perfect see through edge.

[MadDogMe]

Ahhh!, acrylic machines much easier than polycarb does'nt it?. I've heard that polycarb tends to throw of chunks when drilling if not done uber slow with an uber sharp drill bit, it's very hard/brittle(nearest word I could think of), but acrylic is very soft yes?...

[Tuff]

I have been using Lucite (Acrylic) and yes its quite soft and does not chip very easy.

Tuff

[gone_fishin]

Polycarb machines much better than acrylic. Acrylic gums up and is much more prone to cracking.

[Al Kaseltzer]

Ok, here's a new question: Why are there two exit barbs ?

Since your flow is being split and managed by the jets and cups, isn't splitting it again to two exits just redundant ?

[Cathar]

Originally posted by Al Kaseltzer
Ok, here's a new question: Why are there two exit barbs ?

Since your flow is being split and managed by the jets and cups, isn't splitting it again to two exits just redundant ?

With the next prototype I'll be experimenting with a single outlet. The two outlets are there as a left-over from reusing the White Water's machining programs.

[utabintarbo]

Cathar, have you looked into a plenum for the inlet? It would seem that it would result in more coherent jet flow, resulting in higher impingement speeds.

Bob

[gone_fishin]

I have a question. Once the boundary layer is gone at X velocity, is there any point in going faster? It will expand the no boundary layer region but here it is being limited by a cup anyway.

[Cathar]

Originally posted by utabintarbo
Cathar, have you looked into a plenum for the inlet? It would seem that it would result in more coherent jet flow, resulting in higher impingement speeds.

Bob

It already has one. Under the central inlet there's a hollowed cavity that acts as a plenum chamber for the jet inlets.

[Cathar]

Originally posted by gone_fishin
I have a question. Once the boundary layer is gone at X velocity, is there any point in going faster? It will expand the no boundary layer region but here it is being limited by a cup anyway.

Yes, it does benefit to go faster as more of the water comes into contact with the metal directly as the increased downwards velocity will increase the chance the water molecules that would not have made it to the metal will now actually reach it.

There's a limit of course, which happens when every single water molecule comes into contact with the metal and each molecule absorbs the same amount of heat each. This would be "perfect" convection within the limits of the coolant itself, which is what you're talking about. We're not close to that limit though (but are getting closer).

Perfect convection will occur when the heat source rises by exactly the amount caused by the thermal paste layer, the metal conduction interface, and the coolant rise in temperature as a result of the coolant's flow rate.

If you graph that on a flow vs C/W chart you get what would be the absolute limits to waterblock efficiency. Given that we know the TIM resistance and can work out the metal conduction resistance, we can then plot a real data graph of our waterblock much like BillA has done in his tests.

Doing that for the White Water shows it to be still quite a long way enough, despite it being an impingement design, but it grows closer to the "perfect" line the more the flow rate is increased, showing us that increasing flow rates still helps, even though the boundary layer may have gone.

The theory papers talk about the stagnation region convection efficiency in terms of the Prandtl and Nusselt numbers, where higher is better, and higher is achieved through increasing the flow rate.

[utabintarbo]

Originally posted by Cathar
It already has one. Under the central inlet there's a hollowed cavity that acts as a plenum chamber for the jet inlets.

Have you tried different plenum heights/volumes?

So many questions.....

Bob

[Cathar]

Originally posted by utabintarbo
Have you tried different plenum heights/volumes?

So many questions.....

Bob

With the one I'm using now the flow appeared pretty even when just pumping through the two top plates. Actually I want it to somewhat favor the center holes.

[gone_fishin]

Originally posted by Cathar
Yes, it does benefit to go faster as more of the water comes into contact with the metal directly as the increased downwards velocity will increase the chance the water molecules that would not have made it to the metal will now actually reach it.

There's a limit of course, which happens when every single water molecule comes into contact with the metal and each molecule absorbs the same amount of heat each. This would be "perfect" convection within the limits of the coolant itself, which is what you're talking about. We're not close to that limit though (but are getting closer).

Perfect convection will occur when the heat source rises by exactly the amount caused by the thermal paste layer, the metal conduction interface, and the coolant rise in temperature as a result of the coolant's flow rate.

If you graph that on a flow vs C/W chart you get what would be the absolute limits to waterblock efficiency. Given that we know the TIM resistance and can work out the metal conduction resistance, we can then plot a real data graph of our waterblock much like BillA has done in his tests.

Doing that for the White Water shows it to be still quite a long way enough, despite it being an impingement design, but it grows closer to the "perfect" line the more the flow rate is increased, showing us that increasing flow rates still helps, even though the boundary layer may have gone.

The theory papers talk about the stagnation region convection efficiency in terms of the Prandtl and Nusselt numbers, where higher is better, and higher is achieved through increasing the flow rate.

Yes, but my main question is the cup limiting the size that the stagnation region will spread out to. Maximising the area each of your jets has to work with inside the individual cups is what I had in mind. Which brings me back to my little sketch that suprisingly nobody has commented on. A cup alone, with ever increasing velocity, seems it would have diminishing returns because of the exit flow interfering ever more powerfully with the incoming jet, no? In other words I would not expect the benefits of increased flow to be linear, but perhaps the curve could be improved upon?

[Cathar]

Originally posted by gone_fishin
Yes, but my main question is the cup limiting the size that the stagnation region will spread out to. Maximising the area each of your jets has to work with inside the individual cups is what I had in mind. Which brings me back to my little sketch that suprisingly nobody has commented on. A cup alone, with ever increasing velocity, seems it would have diminishing returns because of the exit flow interfering ever more powerfully with the incoming jet, no? In other words I would not expect the benefits of increased flow to be linear, but perhaps the curve could be improved upon?

True, the cup will limit the size of the stagnation region, but this would only be important if there was only a single jet, which there isn't. There is no need to maximise the size of the impingement region for higher flows because even though the region is bound by the cup walls, there's a half-dozen other cups immediately surrounding it, each on benefitting from higher thermal convection as a result of the higher flows.

So better to think of it in terms of the whole, rather than the single jet.

Also, remember that thermal convection drops off fairly quickly outside of the main impingement region, even if the flow rate is increased, so by bounding each region and surrounding it with other regions, we actually get a more uniform heat dissipation spread.

This way, the size of the cups can be tuned to work well with lower flow rates, and still enjoy good performance improvements as the flow rate is increased.

[bigben2k]

Ok, but if you knew that you were going to have a good pump, like that Johnson pump (or two ), what would you do different, if anything?

[hara]

Cathar. I couldn't sleep last night, and I thought of an Idea. Since your design cools the cpu directly above the core, maybe 1mm to each side, a way to further degrease temps is to optimise it to specific processor core families. This has the disadvantage of making the block suitable for one cpu type. The Idea I had in mind is to make available a set of gaskets which one could fit between the top and middle plate to "seal off" holes that are outside the core area. In this way, users have the choice to cool what's needed and since some holes are sealed, the cross sectional area decreases and the water velocity increases where it is most useful. These guaskets could make the block more future proof. This is the idea (holes in red region covered):

The only disadvantages I thought of is that pressure drop of the block varies with each processor and you'd make people mess with the block.

The guaskets needn't be made out of rubber since it's not like we're making something leak proof

[Cathar]

Originally posted by bigben2k
Ok, but if you knew that you were going to have a good pump, like that Johnson pump (or two ), what would you do different, if anything?

I have an Iwaki MD-30RZ here that I test with, along with an Eheim 1048, and Eheim 1250.

I can achieve 3lpm, 4lpm, 5lpm, 7lpm, 11lpm and 12lpm flow rates by hooking the pumps up in series or alone, and/or by turning one of a pump in series off (which artificially adds restriction).

For the Rev 1 block, the performance difference between 3lpm and 12lpm, after factoring out the difference in water temperatures, is close to 2.0C on a ~80W heat load.

I expect the difference to be slightly less with Rev 2 because of what I talked about above.

I wouldn't do anything different. In my mind, density of the jets is more important that maximising each single jet for the strongest possible pump. Get the density up high enough and the block will still benefit from a high pressure pump while losing less on the lower pressure pumps.

Basically the goal here is to get the flow vs C/W curve as flat as possible, but be starting off at a much lower point at low flow rates than other blocks.

Look at the flow vs C/W of the WW vs the Atlantis (an impingement block) for an example of what I mean. Above 2lpm the Atlantis benefits from an ever increasing flow rate, moreso comparatively speaking than the White Water. From 2lpm to 10lpm the Atlantis picks up 0.05C/W, while the White Water picks up 0.03C/W, but the White Water starts off so much lower. In the end the WW is miles in front, despite gaining less from higher flow rates.

I think that best highlights my thinking on this. Yes, the gains may be somewhat limited when contained as opposed to a more open impingement approach, but we're starting off with the curve so much lower that by the time the upper-end gains come into play, we're still ahead.

Tune for the low-end, and the high-end performance follows. It's important to keep one eye on the absolute performance and not let relative performance differences of different approaches overshadow the absolute number.

[Cathar]

Originally posted by hara
The Idea I had in mind is to make available a set of gaskets which one could fit between the top and middle plate to "seal off" holes that are outside the core area.

http://forums.procooling.com/vbb/sho...5147#post65147

[gone_fishin]

Originally posted by Cathar

Also, remember that thermal convection drops off fairly quickly outside of the main impingement region, even if the flow rate is increased, so by bounding each region and surrounding it with other regions, we actually get a more uniform heat dissipation spread.

That is tied in with the size of the jet and all other variables but what I am getting at, are the cup walls limiting the size of the stagnation zone? Higher flow rate against the copper always is better (let's call this the generic benefit of velocity) but in this stagnation area the benefit is huge in comparison. How sure are you that there is no room for improvement? The idea of horizontal parallel tubes cutting into the cups still has me intrigued. The copper in that region is basically wasted realestate in a honeycombed array. Relieving the pressure in each cup may show further improvement by allowing a larger stagnation region to form if all else is fine tuned properly.
I realize you are at rev 2 but come on don't stop now.

[Cathar]

Uh, gotta keep in mind structural integrity here. Widening the cups too far won't leave much copper to hold the block together.

I already experimented at various flow rates, and the results basically showed a drop off in cooling performance beyond a certain cup width.

I understand what you're getting at, but I don't believe that it's an issue here, again because the width of any one stagnation zone is largely unimportant given the surrounding cups and their zones.

Basically we're swapping the efficiency of any one region for a mass of regions.

It's a fact, the further you get away from the center of the stagnation region, the lower the cooling effect. Better to have a mass of properly formed stagnation zones, and allow no wider than these properly formed zones, to avoid the performance drop off as you move away from the center.

Making the cups wider than I've established causes a performance loss across a broad range of commonly accessible flow rates, and would result in a block tuned for excessively high flow rates that not even I can achieve. What's the point?

The optimal width of the cups seems to vary in very small amounts between 4lpm and 12lpm, basically being near indistinguishable. There actually is a fairly happy tight range of cup widths (in proportion to the jet widths) that works for a large range of flow rates that people commonly use.

If we were pushing 50lpm, then yeah, maybe enough to require a different approach.

[Cathar]

Has a few issues with the holes not being drilled cleanly. I've talked with the machinists on this and will be getting a new plate made up using a different material and revised machining procedure that should fix the issues. Have plugged it in though and on the first go is achieving ~1.5C better than the WW, even with the clogged nozzles. Gotta head off to grab a drill bit and clear up the swarf binding issues with the holes on this one and will retry.

Anyways, here's the happy snaps of Rev 2.