Project Block
 

Hello everyone. I'm somewhat new to these boards, but certainly no stranger to watercooling. I've been tooling around with some block designs for a little while now, and have set aside some cash to get a few prototypes done. I've settled on something of a 'jet' impingement' pin arrangement that's most like The Nexxos block. Here's the pin arrangement:

http://www.extremite.com/pins.png

There are 135 pins, each 1.4mm square, 2mm tall and spaced .2mm apart. These pins occupy an area 2.2cmx2.2cm. There are 29 circular cuts for .8mm 'jets' (on the middle plate). For clarification, the jets on the adaptor plate are actually just holes and not tubes like those that are used on the Cascade/Storm. This block is intended to be somewhat restrictive and designed for IHS processors. I'm looking for opinions on the following:

1) Pin height. Too short?

2) Gap between pins. Too wide?

I settled on the size, length and number of jets using data I've forgotten long ago. I'm certainly not looking for stellar performance - just looking for an enjoyable project that will end up being reasonably capable. If there any adjustments I could make to yield a more ideal, fairly-high restriction block, I'm all ears.

Re: Project Block
 

4 mm tall pins might be a better start.

I've always wanted to see that exact idea implemented.

How do you propose cutting .2mm wide channels? Last we checked, only an EDM process allows that.

Re: Project Block
 

I certainly wish there was a bit more technical data published about the various intricacies of water blocks. I originally anticiapted a 2-3mm pin height based on other blocks I've seen like this and the basics of thermal transfer (which I won't pretend to really understand). I figured that if the flow of coolant is turbulant around the central pins (spaced and sized according to the primary area of heat directly above the core), most of the residual heat would reside in the lower to middle portions of the pins. Overall surface area would increase with an increase of pin height, but would it really contribute to an overall lowering of CPU temperature?

I know little about machining and what I could achieve with quality CNC machines (plentiful around these parts) and really intended this block to be the means of better understanding machining/fabrication technology and its limitations. What sort of channel width would be reasonable using a slit saw? I could certainly offset wider channels by decreasing the pin size (slightly) and increasing density.

I appreciate your input. I've had my eye on your RADIUS project for a little while now.

Re: Project Block
 

We haven't discussed one important aspect; what baseplate thickness are you planning? You should be in the 1mm+ range, possibly up to 3 mm.

Determining pin height is difficult. I know only Cathar has been able to model it, hence the 4mm suggestion (searches will probably hint at 5mm being "essentially" useless). If I was building a block, I'd plan on the long side (both pins, and baseplate), so that I can shave it and see if the performance is better or worse. Of course that means that A) I have the testing equipment and B) the design lends itself to "shaving", which isn't entirely obvious in a pin grid design.

As for CNC, I believe you'll find that endmills are only usefull down to 1 mm (and take an insanely long time). A slitting saw would be more in line. You might want to focus on finding a machinist, and buzzing them on the limits of their capabilities. I had a source for slitting saws but the site seems to be owned by someone else now.

On a side note, the Swiftech Apogee has (apparently) been an exercise in determining the ideal pin density (from previous, albeit sketchy, info posted). Similar designs include the #rotor block, and Hoot's (Overclockers) pin-fin block, neither of which (I believe) included such refinement.

If you make it, and I have my testbench up, I'll be more than happy to assist you with this design.

Re: Project Block
 

Baseplate thickness is 2mm. I'm not entirely sure if I'll need to add a 'step' to alleviate any problems with mounting obstructions. Always something that can be done a little later.

I tried to find what information I could on the #rotor block, but his little site has been down for (what seems like) a while, and the only technical information I could find was included in an article on MadShrimps.de, and even then the data was a little sketchy. I also tried to track down the patent that Swiftech's Rouchon holds as the "Diamond Pin Matrix", but I guess it's still in pending.

From what I could tell from the images of the Apogee, the pins are slightly taller than the baseplate is thick, and Swiftech went with 3mm for the baseplate, so I'd imagine those pins are...what do you know...~4mm.

I'm 'shopping' for testbed components at the moment. Money's a little tight (bad time to start such an expensive hobby), but my to-be testbed should be suffecient to establish some fairly accurate, repeatable results. I'll definitely keep your offer in mind, however.

EDIT: Alright, someone really ought to off me for my gross miscalculation of my channel width. The correct distance between pins is actually .35mm - not .2mm. Seems more feasible.

Re: Project Block
 

You'd only consider a step if you plan to have the block sit over the cam box.

Feel to query me, if you need assistance.
[begin shameless plug]
...and/or join the WBTA! (wbta.us)
[end shameless plug]

Re: Project Block
 

You can cut out everything that is more than 3-5mm away from the die. So don't make the pin-area bigger than that. you cant count + 2mm heatspread horizontally for the IHS (an A64 IHS is at least 2mm thick) remember always that the thermal dissipation of the die is more on the edges than in the corners, so the pin-area could be somewhat roundshaped.
4mm Pins are too tall, that won't bring any improvements especially on IHS CPUs, because the 3-5mm precept also apllies to the height. everything beyond that 4mm will have a relatively high thermal resistance and higher your milling costs, reduce the tubulence in the lowersector, ...
I would better make the channels a little wider ore more channels. If you wan't more flow don't change the length of the pins but the size of the channels, that won't remove the tubulence on the baseplate.
Pindesigns have a very high turbulence because of their restrictiveness. If you calculate the reynolds-formula the variable of the channel diameter is very important. the smaller the diameter of the channel the better the tubulences.
Turbulences affect directly the heat transfer efficiency.
Kepp that in mind when you try to make it somewhat bigger or less restrictive than the original principle (the nexxos XP). But it would be a great idea to remove all the restrictive channels on the outside...
To maximize channel surface and minimize thermal resistance the rest bp has to be as thin as it could be, but below 1.5mm the chance of bending is too high.

From my experiences 4mm is a good copper thickness, with 1.5mm bp and 2.5mm channels 0.35 sounds like a good pintopin distance

Re: Project Block
 

Thanks for the reply.

I've been tweaking the pin arrangement over the weekend, and I've arrived at (what I believe) will be a better pin arrangement. Channel width has increased slightly - to .4mm. The pins themselves have been somewhat reduced to size (after reading RoboTech's analysis of the Swiftech Apogee) to 1.1mm x 1.1mm (down from 1.4mm). The pins now encompass an area 1.9mm x 1.9mm. It seems I went somewhat "hog wild" when making some corrections to the original design, which is why the pins occupied a seemingly large area. I've also tweaked the size and number of jets to ensure there's water always forcibly striking the area directly above the die, as Cathar said in his findings during the White Water project.

The jets are still something of a gray area for me in terms of what's good and what's not so good. What kind of velocity should I be aiming for, and what pressure drop through the jets would be too high? I've calculated that a single chamfering pass will help alleviate a fair amount of pressure drop, but I'm still somewhat unsure what would be a good range of velocity, though I recall LIG saying something in the 200-300 meters/sec range long ago.

I wish the money flow was good enough to allow me to do a number of prototypes, but at the moment, I only plan on making 2 to 3 complete prototypes, and I need to start defining what variables will change with each 'version'. I'll likely alter pin height for two of them - most likely 2.5mm and 4mm.

Re: Project Block
 

A little update here is warranted, I suppose.

I bought the student version of SolidWorks (for a pittance!) and have spent a few hours fooling around with it. I've decided to optimize the arrangements for a lower restriction pin block and to also design a block that performs on entirely different principles (for a pin grid block, anyway).

Following Ben's advice, I buzzed a couple local machinists about their capabilities, and was told, essentially, that if I could supply a slitting saw, they could cut the copper. After finding .0080" and .010" (.254mm) solid carbide saws (with reasonable +/- .015 variances), I'd figure I'd try and design for the latter. The result is 612 diamond pins .6mm x .4mm with a height of 3.7mm (effective cooling patch is 1.9cm x 1.2cm - a little on the small side with respect to width).

http://www.extremite.com/render_5_low.jpg

Debating my choice of material and baseplate thickness for this one - I'll likely reduce BP thickness to somewhere around 1.7mm to 1.8mm in the hopes that edge chamfering will give it enough structural rigidity. May also implement G7-style braces on the midplate to help keep things as evenly flat under loads as possible.

On the subject of the lower restriction prototype I have in the pipeline, I'll be rethinking some of the priciciples in hopes that I'll be able to reduce crossflow and stagnant flow regions. Something of a microcup+pin configuration. More on that later.

I'm still collecting items for the testbed. I intend to pop the IHS on my old Intel 2.4C, probe it and power the loop with an RD30, though searches for the RD30 have proven to be feeble. If anyone has any leads on these things, I'm all ears.

Re: Project Block
 

Great, now take your design to the shop, and ask them if they can do it. Next is how much...

A tight fin density like this can return impressive results.

Re: Project Block
 

looks very good. I'm actually reading a paper right now on some prof who did exactly the same thing (only with compressed air) so the design obviously has some merit.

I'm unsure what you mean by crossflow, but crossflow + jet impingement is what they do in industry. In fact they bare use cups at all, just a horzontal flow and some jets. Can get about 3 times the performance over jets on a plate.

A general rule for jet impingement is more but smaller jets is better. decide what jet pressure drop your willing to accept and then see how small you can make your jets and that should tell you all you need to know. 200-300m/s is way way way off what you'll be getting. Take a look at a storm style design and work of that (flowrate per jet/area). should be some nice data in the storm g7 thread from awhile back.

Re: Project Block
 

that should perform VERY well on a bare die...if you can get the velocity up enough that thing i would bet would perform EVERY redily availble bloc on the market...maybe even match dare i say cathers...g-7....but that is only if you can get the jets righ and the bp thickness and so on...

Duke

Re: Project Block
 

While I'm optimistic that this block will perform adequately, there are a number of factors keeping this from performing as well as Storm G-series blocks, for a few reasons:

• The G5 and G7 both have silver baseplates, which probably leads to a .5-.6 decrease in die temperature. This block has a copper baseplate, and I don't really intend to try a silver variant. Silver would be feasible, but I have serious concerns about baseplate rigidity with such a soft metal. There's very little I can do to increase rigidity enough to make silver a feasible material. I've done a little testing with my G5 and have found that the baseplate can 'bow out' slightly when applying a modest amount of pressure. Where it counts, the G5 baseplate is twice as thick as the one I've drawn up here. You do the math :eek:
• All Storm blocks are not highly restrictive. The G7 is less restrictive than the G5, the G5 less restrictive than the G4. High density jet impingement pin grid blocks inherently have a fairly significant pressure drop. My designs are more effecient with respect to the jets than other blocks (mainly the MP-05 blocks), but it should nonetheless be a restrictive monster.
• On a small, simulated die testbed, this block should be surpass the G4 slightly at most flowrates. In the real world, on real configurations, where flowrates vary from block to block, I don't expect wonderful results. Certainly good enough for me, though. And that's the idea.

I'd like to eventually duplicate the G7 some day, but by that time, we'll have multilayer procs with tiny nanopumps doing all the work, and $600+ worth of a fancy waterblock won't be able to compete :)

The G7 is no doubt a marvel (though not based on particularly new concepts), but it doesn't seem as if I'll ever have the chance to own one.

Thanks for your comments.

Re: Project Block
 

I'm worried it is too restrictive though and that the pins may end up being to thin

Silver vs copper should mean next to nowt.

Confined jet impingement effectively has 2 jets, one hit the surface and one at the corners so i should mostly beat a hybrid, jet fin design

Re: Project Block
 

make the pins shorter so that the water and turbulences can get to the baseplate where the thermal resistance is the lowest and turbulences to lower the boundary layer would be most effective. If the heat has to cross a long tiny pin before it reaches the cold water, the thermal resistance is up...

Re: Project Block
 

Uh um

Turbulence and boundary layers dont work like that what so ever and in fact the entire sentence makes no sense.

You are right about the pins being shorter because heat can't effectively reach the top.

Re: Project Block
 

Shouldn't turbulance at/near the baseplate be essentially the same regardless of pin height? (within reason, of course). I'm attempting to also take into account the cross-sectional area of the main plenum; 3.7mm tall pins seem about right. As for boundary layer formation, I don't expect much at any point in the pin grid. I'd also like to note that there are many more jets than you'd think, so there are no gaping gaps where some pins will get 'hit' with much less turbulant coolant.

Trying to alleviate restriction mostly by tweaking the jets. Since ~60% of the pressure drop in a G5/G7 is due to the jets, I'm mostly focusing on tweaking this particular area.

We'll see what happens with the pins. My machinist may advise that they be shortened, but until that time, I'm going to go ahead with 3.7mm.

Re: Project Block
 

Before I say anything.

TURBULENCE IS NOT A PRIMARY GOAL OF WATER BLOCK DESIGN. HAVING LOADS OF TURBULENCE MEANS DIDDLY SQUAT.

The main thing loads of turbulence means to me is that you are having loads of unnecessary pressure drop. It can decrease block performance by creating recirculation zones.

Secondly boundary layer has no concept in this kind of block. It is just a science word in this case.

Some due to height (maybe), velocities are definitely a function of turbulence. So you can say turbulent varies with height but it is more complicated than that and beyond what is needed here.

~60% of the pressure loss comes from the jets but 100% of the performance comes from them.

Re: Project Block
 

I think we're in total agreeance across the board.

I've spent a great deal of time working on the number of jets, their diameter, height, spacing, and the number of times their "inlets" will be chamfered (or is fluting the proper term?). I don't intend to sacrifice anything by tweaking them for the least amount of associated pressure drop - they'll produce as much velocity as possible with as great of an associated pressure drop as I can stand (and is reasonable).

I want to work with a set jet velocity, pin size and grid density and optimize without sacrificing the set variables. Trying to discover exactly how far is too far.

It would be nice if, at the outcome, I could easily say, "Don't do this, because this is excessive and stupid." I could then begin to understand what factors make a high density jet impingement pin grid block ideal or not so ideal.

Re: Project Block
 

No, do a little research on the reynolds formula and you will realize that besides the velocity, the channel diameter is the most important factor of Turbulence. When two channels have water with the same middle velocity flowing through them, the channel with the smaller width will have the more turbulence.



No, see above.

That is right. The Goal is to lower the boundary layer through turbulence at the heating surface and not inside the jets. In a tiny Pin jet Block, the jets are only for water distribution, not for velocity as in the Storm. A different approach...

The primary goal of water block Design besides the old verities of blockdesign surface and distribution is to enhance the contact between the heatplate (copper/silver) and the transport medium (water). The only way to do this whithout boosting the surface is to lower the boundary layer with flowspeed. But as laminar flow goes, even extreme high flowspeed of the pump will have nearly any effect on the boundary layer. That is why you need turbulences to do the job. See here a very good Illustration from cooling-masters:
http://www.cooling-masters.com/image...uencedebit.gif

Avoiding recirculation is not so important as the heat-capacity of water is enormous, but keeping an eye on it while deisgning the flow through the block sure would pay off a little.

Oh really? I think the whole block is based on that concept. Maze-style blocks may not be constructed around the laws of the boundary layer, but pin style blocks like The NexxosXP or this one here sure are.

I don't think so. This is not a blind hole style block like the G4 or G5. Why should only the jets bring performance, they are not even connected to the baseplate?!
In The Storm waterblocks the Jets create the needed Turbulence for the Cups, that is why they are so long. In a pin grid Block the channels should be so narrow that turbulence will result anyways, the jets are only for that the water from the laminar flow in the tubing gets spread over the surface of the pins.
In a Nexxos XP the biggest pressure drop comes from the pinGrid array. But there is still more pressure drop than needed at the inlet and with the outlet channels. But the narrow channels in the middle are close to perfect.

Re: Project Block
 

Ah where to being.

I have a degree in mechanical engineer. As part of this degree I have lectures on turbulence. I did Reynolds number in first year. In my locker at the moment I have 3 books on turbulence, ones 500 pages long and looks mean. I also gotta see a man about some turbulence tomorrow as well.

Reynolds number. It’s a non-dimensional number not a formula. You change the formula if you so wish. It can all get a bit random.

So I will ask you this question what does the Reynolds number actually mean and where does it come from?
Hint it is only partially to do with turbulence. You will from that answer that the Reynolds number only tells a bit of the story about turbulence. A flow with a Re=50000 is not necessarily turbulent.
It might take you a few thousand words to answer that. Probably by the non-dimenisonal navier-stokes equation (hint hint).

Second question what is a boundary layer?
I’ll answer this one. Its when a flow is 99% (okay so it doesn’t need to be 99% could be 90% or whatever, a large percentage) of the mean flow velocity. Actually there is the fluid boundary layer (velocity boundary) and the thermal boundary layer (temperature). You haven’t defined which one you are talking about. Common engineering language would say that boundary layer is fluid boundary layer. Now I’m going to bet that your actually talking about thermal boundary layer here instead. Shrinking the thermal boundary layer could be a good thing it would imply lower thermal resistance but its undoubtedly complicated. Arguments saying that the thermal gradient is equal to conductivity etc maybe (k * dT/dx). I’m not convinced entirely. Could also argue that jets have their fluid boundary layer and I would accept that too. But in between the pins possibly not.

Pin block jets only for water distribution?
Hmm, a lot of fluid mechanics for people are trying to do other stuff and use them for cooling. Jet impingement cooling is mainly used in these situations. Get a normal heatsink that is not doing its job and add jets to make it better.

You can do an awful lot with maze blocks and a fluid mech block. Using some boundary layer stuff you could work out heat transfer rate. You may even find out that for you average maze block the thermal boundary layer would cover 10% max of the flow.

I was talking about blind hole (as you call it) blocks. They most definitely are connected to the baseplate. Jet hit fin, fin touches base plate.

Recirculation zones are not good ever (well actually there are some cases, storm blocks might get 50% of their heat transfer from). They are generally not advantageous.

I think you should go and read a good book on heat transfer then you might see that turbulence is only a marginally good enhancer of performance. Velocity has everything to do with it. I was doing some CFD today on fins the correlations I came up with were heat transfer was Nu = Cl x Ra^(1/3) for lamina and Nu = Ct x Ra^(3/8) for turbulent. Not a lot of difference there. Ra = Rayliegh number its like Reynolds number for convection.

There’s much more I could post but I’m going to bed. I might add some more in the morning when this post isn't brought to you by beer :).

Re: Project Block
 

So we have a variance in the Reynolds occuring at various points in the pin grid (height). I don't think the impact in effective heat transfer performance is really a need for concern, nor would I really be able to come up with a reasonable solution to eliminate these variances. One fix may be to reduce pin height, which would decrease the height of the plenum. This would have the nasty side of effect of reducing the cross-sectional area of the plenum to something I wouldn't be terribly pleased with. Pin height, as it's designed, is something of a priority since it also impacts how other areas are to be designed (to be most effective). As long as the pins hold up to the slitting saw, I'll be satisfied with them.

May be experimenting with different inlet configurations. I previously considered a simple White Water style slit, and may end up giving this a shot. The only real snag with this is that the middle plate is a fairly elaborate component to machine. Having a number of different mid plates, however, may prove to be a pretty good exploratory value.

Re: Project Block
 

Just give up trying to estimate stuff as it is impossbile without CFD.

There will be an optimum pin height. To heigh and you lose flow velocity. Could always try and make it then starting shaving off to find a good level, then make a new one.

Re: Project Block
 

I actually enjoy discussing with you because you have some knowledge, i would be delighted to continue this and see what fruits come from that. Sorry for making your thread a messy discussion, i hope you don't care so much. I apologize for my bad english, i never wasn't good in any language (tried japanese and french too a little bit but that was far worse), but i estimate that you wouldn't be able to write in german too, so considering that i think i mustn't be too ashamed of my bad language skills.

The Reynolds number describes a critical state where turbulence starts in a circular tube. The values to calculate the number are velocity, density and viscosity of the fluid and the tube diameter. Of course can't you calculate a watercooler with the formula as it is only meant for round tubes and doesn't say anything about heat transfer between the mediums. But you can use the knowledge that channel diameter and velocity are important for turbulences.
The navier-stokes equations are the right formula to calculate the flow in more complex geometry, but i would hell not calculate something with it by hand. that is what CFD is made for. But befor CFD you need some hints to create a Design. And when you basically know what the factors for heat transfer between a less conductive but high capacity Newtonian fluid and a very conductive but low capacity solid body, you can actually plan a design and adjust the fine proportions with after CFDing that. Plus the navier-stokes equation is OK for laminar flow situations but you don't know how accuarte it is for turbulent flow, you can only guess, even CFD is guessing. For turbulence you can't simplify the equation as much and DNS eats a lotta CPUpower. The Solid works plugin can't do that, you need Fluent+Gambit for that (i used Fluent once and it is a thousand times more powerful than the solidworks plugin simulations i have seen so far).

I'm talking about Fluids, so i am talking about the Fluid laminar boundary layer, which is in that case a thermal isolation layer.
The boundary layer is most important to consider for watercooling and this is why:
- Water has a conductivity of 0,58 W/mk while copper has 402 W/mk. So why do we use water for heat trasportation? because copper isn't a fluid, we can circulate the water and use its unbeaten heat capacity to transport as much heat as possible.
But what is when the water doesn't circulate, when there is a dead spot of water? Isn't it as if you would isolate the copper with some low conductivity plastics when water is not flowing? That is exactly why some waterblocks with an equal surface but different flow can perform much different. When planning channels always avoid dead spots, solid copper is much better than water which is not flowing. because of the moving molecules pursuit of biggest Entrophy (Fluctuation), this Isolating effect isn't as bad as the raw numbers tell, 0,58 to 402 W/mk, but it still has the biggest impact on conductivity.
- In Conclusion: Because The Boundary layer consists of water molecules latent due to friction on the sides of the channel and some very slow moving molecules, and the bad conductivity of water, we have thin Isolation layers on our copper cooling structure which more or less effectively hinder the heat transfer between copper and water. Turbulant Flow has multiple decimal powers the Fluctuation and crossdiffusion compared to laminar flow (even if we only have a relative small pressure drop) which enhances the heat transfer significantly.

First one i agree, there is a boundary layer in Jets, second one is that there of course is also a boundary layer between pins. Even if its only some molecules thick between two grinding materials there are always some molecules of a boundary layer especially when one of the material is a fluid.

Dunno what you mean with fluid mech block.
The 10% figure depends on how you measure the boundary layer and your middle pump velocity. In laminar flow situations the fastest velocity is in the middle and it lowers linear to the sides. That is why you won't lower the boundary layer much when you increase the overall velocity because the biggest increase will be in the middle.

That is different in turbulant flow scenarios.

There is no thermal connection of the jet itself to the baseplate (at least no good). Due to high velocity inside the jets and most often quite good turbulences the boundary layer there is really thin. We should use that surface of the jet entry plus the inner jet tube surface for heat dissipation.

I fully agree with the second sentence. Velocity is the key to heat transfer between a solid body and fluids. But you didn't specify which velocity. In general when we talk about the velocity, whe mean the middle velocity, which we can measure easily by the amount of fluid which flows in a certain time. But this velocity is only important for the heat transfer time and not for the heat transfer between cooler and coolant itself. The Important velocity is the velocity directly near to the dissipating surface. Due to the laminar boundary layer we know that the velocity decreases rapidly the nearer to the dissipating surface we measure.
When we take turbulent flow instead, the velocity in the crossection is far more even to the sides (that is where the pressure drop comes from).
With turbulences the middle velocity may decrease but the velocity which is effectively making contact with the dissipating surface can be much bigger.

The Rayliegh number is only important for convection in a fluid, not between a solid body and a fluid. Of course it is not bad when Ra is high or rather that there is much Fluctuation (which basically is the same as Convection) and Turbulences can help that but it is not the most important figure in watercooling.
BTW: what software did you use for that, it doesn't seem too accurate ;)

hehe, needed a beer for the post you quoted too :D

Re: Project Block
 

So we know that there is pressure drop inside of jets. That means that there is high velocity and probably are turbulences too. So why not use the jets aditionally to the cups/blind holes?

This is an example of how one could do a jetblock with jets and cups from one material, it has nothing to do with pin-fin blocks, so its OffTopic, but it has to do with Stormlike Blindhole blocks. It is an attempt to remove any restrictivity (and going with that thin boundary layer) on surface which is not the dissipation area. I would call that pressure drop efficiency. This is not a block design, only an illustration, it is nearly impossible to machine (needs 5-axis cnc for the wholes to drill) and there are still problems to solve with the restrictivity of the in and outlet channels to the V-jetholes but just look and try to understand what i mean:
http://forums.procooling.com/vbb/att...1&d=1143912487