xxx is x^3 not 3x though :)

I would have just put some silicone in the bottom of the cups, put the jets peice in, let it dry and fire it up but your way sounds more professional!

LOL. C'mon. Work with me here! :)

It's a reduction of the Cascade design in all 3-dimensions. It is, in essence, a direct cubic relationship between the Cascade and the XXX.

Every 2.8mm² of CPU real-estate will have its own jet/cup.

There. Is that better?

I'm outta here. Back a few hours...

You guys need to think outside of the box a bit more.

I like the tunable jets Cathar. Maybe add the ability to raise or lower them according to the flow.

Honestly though. I don't know why you would want the jet tubes any higher. The closer they are to the plate the better, until you restrict flow anyways. That will be anywhere around 1mm with those jets you use. Of course higher flow/pressure always allows for them being further away but from what I seen and the math Nusselt number will tell you the closer the better. Use Nusselt average over the area, the distance from the stagnation point.

Nu=hD/k

Re=VD/v

I'm missing a few variables as I don't know how to enter them in using a keyboard.

D=diameter
V=velocity
v=kinematic viscosity
k= thermal conductivity
h= average heat transfer

Of course the shape of the jet changes all that then it gets more complicated. Basically a different shape will allow a bigger cooling area.

I cant wait to see the new Cascade XXX. Announces like a movie "Come and see the new feature block Cascade XXX showing in a procooling forum near you, dont miss it!" Anyways...its bound to attract attention thats for sure. :cool:

syscrusher you can't simply post a basic equation and then back away. DO model the double impingement mathematically and produce optimum design as a function of water velocity and baseplate thickness. Especially with nothing more than a calc of those two numbers :)

Hint: it isn't a currently solvable problem mathematically

Syscrusher. In a nutshell as to why you want the jet to be stood off slightly can be explained in the following way.

When the jet is really close to the surface, the what happens is the water just squirts out the side of the jet tube, but in the middle of the tube it doesn't really move at all. i.e. the point of central stagnation is fairly large. Imagine filling a glass with water and sticking a flat piece of something on it and turning it upside down. Now lift the glass slightly away from the surface. The water that flows out mostly flows out the small gap, but the water in the middle is barely moving at all, i.e. the stagnation region.

Now do the same experiment but lift the glass away quickly and all the water pours out and strikes the whole area under the glass, rather than merely leaking out the sides.

By standing the jet off a certain distance we greatly reduce that central stagnation effect where the water is barely moving at all. The actual best distance to stand the jet off by is linked to the velocity of the jet stream, as one can well imagine. It gets a little more complicated in submerged jet scenarios where the jet loses power as it moves through the liquid around it, but also gains added turbulence as a bit of a bonus.

The actual math of all that is the subject of a large number of research papers into impingement and its effects. I won't pretend that I could construct a mathematical answer for any scenario, and in fact I don't think that many researchers could either, but the theory that's out there does provide guidelines for good starting points, and unless your jet velocity is extremely low, having the jet really close (<2d) often turns out to be worse.

Have a poke through that paper I linked to above for more information.

I've looked into it a bit. I found some useful info but it is mainly experimental data with submerged jets. The only conditions where nozzle plate spacing becomes less important seems to be when the reynolds number is below 800 which means very low water velocities. I'm basing this on some experiments carried out by Elison and webb mentioned in "Advances in Heat Transfer" volume 26. I would recommend it for anyone interested in jet impingement as it goes into lots of detail about nozzle plate spacing, modified impingement surface, wall roughness, jet splattering, jet pulsation, motion of the impingement surface, all from 180 different references and technical papers. Should keep anyone busy for a while :)

At 2LPM, the (per-tube) Reynold's number on the Cascade is about 1000, which basically means that, yes, it is very close to the level you're talking about. I stand by my statement that once you get down to the ~0.5GPM (<2LPM) flow rates that the WCP testbed tests at, then you're into the realm of where altering the jet standoff distance actually becomes important. At 4-10LPM flow-rates, the Cascade is seeing Re numbers of ~2000-5000, and the z/d ratio has been set for that sort of range.

sorry Cather, I just reread my message. I missed a bit out of my sentence, it was supposed to be in agreement with you. The nozzle plate spacing is very important to the heat transfer coefficient at reynolds numbers above 800. I have a graph here from a document called "Local characteristics of convective heat transfer from simulated microelectronic chips to impinging submerged round water jets". Unfortunately I don't have the whole document as I'm sure it would be very informative. It appears that the higher the reynolds number the more critical the nozzle plate spacing is. If I can get access to a scanner I'll scan the graph in, it might be useful for people who want to experiment

My apologies too as I didn't mean my response to come across that way. I was just backing up what you were reporting with an actual Re number for a certain flow rate with the Cascade.

Oh.
I thought you were disagreeing:
WAJ UK "The only conditions where nozzle plate spacing becomes less important seems to be when the reynolds number is below 800 which means very low water velocities"
and "The nozzle plate spacing is very important to the heat transfer coefficient at reynolds numbers above 800"
Cathar " once you get down to the ~0.5GPM (<2LPM) flow rates that the WCP testbed tests at, then you're into the realm of where altering the jet standoff distance actually becomes important"

No matter

I was just working out the reynolds number for cascade couldn't figure out why I was getting such a huge number. I forgot to change jet diameter to metres. I made it 1020.22598 :)

yeah I know Les, I've confused myself now. All I have is bits of paper with other people's results on. Cather has done all the hard work of experimenting with different possibilities so his comments are probably more valid

Sorry - should've made myself clearer.

From a mix of emperical testing and research. Very roughly:

Re = 2000 => z/d optimally about 4
Re = 20000 => z/d optimally about 5
Re = 200000 => z/d optimally about 6

At Re < 1000, we're pretty close to the region where that pattern starts to break down, and we may as well stick z/d at anywhere between 1 and 2 and be happy.

At Re=1000, z/d should be somewhere between 2.5-3.5, hence somewhat away from the >4 value that the Cascade uses.

Poorly worded. My apologies.

I'd be interested to see the graph WAJ_UK, especially to see if it correlates to the above from my understanding of it all.

Yeah. It's a little gnarly. There were some minor differences between different revisions of the Cascade. The one tested at WCP, being from the first batch at 0.5GPM (1.88LPM) would have had an Re down around 850 or so. For the second batch, Re would be around 900. For the 3rd/4th batches, up around 950. For the SS, up over 1000. This reflects the fine tuning of the production process over time to allow for tighter tolerances.

I must have been using the dimensions for the later batches at 2 lpm.
I'm not sure when I'll be able to get to a scanner so I took a photo of it with my phone. Sorry for the poor quality but it is just about readable

Thats what I meant exactly :)
You chaps obviously have to much time on your hands to post so much or are hopless addicts as I am ;)


Discussion about jet sizing and distancing pontificates over 'How to obtain trully turbulent flow' age old cooling dilema.
Reynolds number tells us if liquid flowing over surface (pipes in our cases) is turbulent or laminar. We want truly turbulent. Reynolds number is proprtional to Velocity and specific lenght L (pipe diameter here). We obviously cannot increase pipe diameter as much as coolant's velocity so we go for the latter.
To obtain truly turbulent flow we need Reynolds >4000 (
reference here ).
The whole reason for turbulent flow is to make as many water molcules get in contact with 'sticky' layer as possible and to reduce boundary layer thickness to minimum (Fourier's Law of conduction Q= k*A*dT*time/d, where d here is thickness of boudary layer and Q energy transferred).
Jet impigement system substitutes for larger heat transferr area limited by dimensional constrains here.

Does it make any snense? :)

Today I have too much time on my hands and I'm a hopeless addict :) I've been sitting around all day taking far too many temp measurements. I have 9 different combinations of blocks to test and I'm almost halfway, after pretty much a solid week of testing. I don't have any lectures at uni on a friday so I've spent all day sitting at the laptop updating my spreadsheets as my test continues.

There is a bit more to water jet impingement than whether or not the flow is turbulent. It is highly likely that the flow is turbulant anyway at lower reynolds numbers. You need quite a long, extremely smooth pipe for the flow to become turbulant as high as 4000. As soon as the jet exits into the surrounding water it will become increasingly turbulant the further it travels through the surrounding water. However the centerline velocity of the jet reduces the further the jet travels through the surrounding water.
These two elements (turbulence and centerline velocity) are the key ingredients to impingement cooling. So basically, the further the nozzle exit from the base plate, the greater the turbulence and the closer it is the greater the jet velocity. So it is a case of balancing the two components to give the optimum heat transfer.

But now I'm off to the pub :)

Typical student :rolleyes:

I understand all that and your correct in this. To close and performance drops because of the stagnation area but that's not what I was getting at.

This is why I made my last post before this. I might of misread your post but I took it that you were saying further away then what you have now was better. I don't know the cascade in it's real form but from my understanding your jet tubes are just about level with the rims of the cups which I would consider optimal as this is what I was doing with my cone shaped design - though inferior to the cup or spherical shape. There is a point were further becomes worse and there is a fine line inbetween good and bad with our flow rates/pressure. Diameter also influences the optimal distance. Being submerged complicates the situation more as the surrounding fluid acts as a restriction. In a low flow situation your about to lose the game as the surrounding warmer liquid could actually heat the incoming liquid or restricts it to the point it never reaches the base. It's a balancing act just to get some where in the middle of the two situations - low/high flow. But then, we're assumming there is no restriction after that where the restriction is greater then the block or equal to it. Though I doubt we would see that.

The math I'v seen has been a bit different from one study to another. But as you say, they do provide a good reference point to start with. But as you know, that's with a flat plate in mind. Throw in furniture, base plate thickness and a different heat source, all that changes. I missed the paper you linked but going to take a look after this.

LOL ok pHaestus. My humble apologies for the little girl remark. I'll admit, that I was a bit rude on my part. But I still stand in my opinions and everyone has one. ;)

I have modeled the impingement mathematically and have been over it and over it. I won't sit here and go through it with you just for your proof as I assume your just as well versed in it as I am. Neither will I play the "Who's smarter or better" game. Base plate thickness is just another variable in the "whole picture of the block" that one can tailor to benefit the impingement you have. Design the impingement based on the flow/velocity/pressure and use baseplate/furniture to benefit that. With all the different systems and flow rates, the problem becomes unsolvable mathematically - I agree. What it can do is get you somewhere close with the constants you have as a reference point to work from.

Ultimate Fluid Physics & Mechanics
 

I've just found this
I think it presents all basic theory in quite simple way.

I am spawnig some ideas of how to improve jet design but atm most o them are too expensive to produce (to complicated to machine.

I am off to produce some 3D models to show you

Looks like we're somewhat on the same page. I'v seen this one and used it to base my block on also. Though quite different from your cascade design, performs just as good.

What would be interesting is to try a capillary tube design in an array that would benefit pelts but be interchangable plates to use in a IHS design. Machining that would be a nightmare though. I'v got a decent design in the works using that idea but you can forget using any pump less than a L30 or equivilent as it restricts flow to much. Not to the mention the amount of drill bits I killed. Now if i had access to a laser. :eek:

Intel has a good white paper you might be interested in also. If I can dig it up again from them I'll post a link to it. It's an interesting read.

Here's something that migh catch your eye Cathar. Check out the chapter called "MICROPROCESSOR POWER AND HEAT FLUX TRENDS"

http://www.intel.com/technology/itj/...cles/art_4.htm

Ignore this

He He He I hope you managed to get yourself properly pi$$ed m8 :)
I used to have my best ideas after two,three tumblers of 18 year old highland single malt ;)

Anyways, what I meant in my post was that turbulence and turbulent flow causes boundary layer to thin down. Jetted water stream is exerting additional localized high pressure causing further thinning of boundary layer over small area (cup). The overall goal is to make boundary layer as thin as possible to achieve otherwise unimaginable conductive heat transfer coeficient (sp?). Optimal distance between cup and nozzle is dependant on nozzle diameter and jet stream velocity. Water exisiting nozzle is shapad like paraboloid function graph due to simple fact that middle of the stream is the highest velocity and the closer to nozzle wals the slower water travels beacuse of friction (viscous) forces (again boundary layer effect). Adjusting distance tries to aim 'peak' of the jet before it starts to turbulate (jet is and must be as laminar in it's flow as possible to obtain the highest speed). The biggest problem with jet design is that it's peak working paremeters windows is quite narrow and that's why Cathar is designing his blocks for certain flow range adjusting z/d accordingly. Imho that's all what could be possibly done in this area.
Looking at the equation I wrote earlier one can clearly see that the next step is to fight for bigger surface area and this is the biggest challange here. Production costs (there are many designs which are MUCH more effective but are exorbitantly more expensive) versus performance ratio is the killer here. Who in his right mind would pay £500 for water block???
I only spent 15 mins on 3D model and it's not yet presentable but I just thought about a clever way to make it feasible to mass produce.
Unfortunately it is going to be yet more expensive than Cather's baby... :(
T'nite I hope to post my model for you guys to check out and beat the living daylights out of my design :)

Yeah, it's like with HP's idea of direct die open air evsporstive CPU die cooling using their piko droplet technology and ultra clear coolants leaving no residues over the hottest parts of silicone. Strangely enough nobody ever heard of it again ;)

The coldet spots are cache areas (memory as we all know do not heat up much). The rest is a mixture of controllers and logic arrays.
Well, the best idea would be to have silicone sandwiching layer through which w could pass veru low energy atoms on much smaller than copper size creating effectively in-die nano cooling solution (quite similiar to Peltier Effect;) )

Would possibly suggest*
h=151.44Re^0.623
for calculating a convection coefficient for 1mm nozzles in a dashpot(bowl or channel urinal).
This generic suggestion is most probably incorrect.However finding an alternative may be difficult and without a reliable estimate of h all CFD(or any other) modeling is completely useless.

* From here

http://www.jr001b4751.pwp.blueyonder...ieder-Tate.jpg


Edit(very late): Mis-read my own graph.
Corrected h=263.66Re^0.643 to h=151.44Re^0.623

Les, when stating h, and the area associated with it, what are you defining as the area? The area of the heat source? The convectional area? The "net effective" convectional area?

Would describe as the average h within a distance of d/5(ish) of a vectorial(terminology?) change in velocity.
Edit. In the WW case the considered area is within d/5(ish, ish) of the die area.
Would tentatively suggest that Impingement is only a special case of a vectorial change in coolant velocity.