Custom CPU water block

[Diamon]

Right. It's about time I start machining my custom water block.
The block will be of solid copper and will have the dimensions 32x34,5x11mm. The paths will be 8mm deep and 2mm wide. It will be machined in an "EMCO" CNC mill running the Siemens programming language. Now for some blueprints!

http://img820.imageshack.us/img820/5080/vblockfinal.jpg



The idea is that water will enter from the top in the middle and follow the blue paths towards the bottom of the block where it will go to the rad via a tube.

Hopefully I'll be machining the prototype tomorrow. Pics of it if I do.

And if you want the blueprints or the program, just send me a PM :D

-Diamon

[Konrad]

How are you attaching the top plate? Copper brazing? And how are you able to ensure the water channels remain unobstructed in the process?

I've been considering a custom waterblock ... evaluating CNC milling vs lost-wax casting methods, somebody even suggested sintering though I doubt that's workable/desirable for this application.

I'm wondering if a solid block (which might also serve as heatsink if large enough) with a series of (maybe 3-6 or more) bored parallel pipes running through it might be better ... it could be used to move a larger volume of coolant per second (it could even be used to alternate or criss-cross isolated cooling loops through the same block), though the greater number of fittings would be more of a chore to solder or leakproof.

[Diamon]

The idea of having a large copper block with pipes in it is interesting. It would indeed be able to move a lot of water. But you'd loose turbulence and surface area it seems, both of which increase the block's effectiveness.

I'll solder the top plate on with silver. Same with the fittings. It becomes very tight. It's possible that some water will take a shortcut between two channels but I doubt it. And if they do it's not a big problem unless it's a lot of water.

[Konrad]

lol, turbulence being desirable didn't even occur to me. I suppose in-block turbulence increases efficiency without making the rad/pump/res requirements excessive. This puts an entirely different emphasis/complexity on optimum block design.

I've even half considered direct-die liquid cooling ... though I'm not certain which parts of the die (if any) can actually get wet under power when the IHS is removed, especially if the liquid is at all conductive. In fact, I'm a little cowardly about the manly methods needed to remove the IHS in the first place.

Your approach is the best I've seen yet, Diamon, and well within my means to duplicate. Are you milling square- or U-shaped channels, and if U-shaped then is the U-curve on the "top" or "bottom" plate?

[Kayin]

Honestly, at this point, insead of all that Gcode, you could do a design on a manual mill that's just cross-sawn, and probably pick up a few C. There's a good reason we've abandoned maze-style designs for impingement and slitsawn fins.

Once you do the block, you can shotblast or acid etchit for more surface area, and pick up some more, smaller gains. Every little bit counts.

You guys are making me redownload sketchup just to mess with this.

[Konrad]

I kind of like the idea of a single block (of whatever design) accomodating multiple independent loops. For redundancy, in case one pump goes down or whatever. If you're gonna go custom then why stick with one-in, one-out?

Or is a single larger cooling system generally better than several smaller ones? There's only so much space inside the chassis, only such much radiator surface available, and a practical limit to how much power/noise you want on the fans.

I'm still a little terrified of the idea that a block can jam up inside.

[Kayin]

It's why I advocate clear tops, but the ones most capable of succumbing are the ones that are impingement based. the smaller the insides, the easier it happens.

Fedora and I are fighting, I'll elaborate shortly.

[Diamon]

Finally got to milling the aluminium prototype (only got so much copper) :D

Here's some pics:
http://img535.imageshack.us/img535/7147/dsc00813o.jpg
http://img545.imageshack.us/img545/4425/dsc00815j.jpg
http://img814.imageshack.us/img814/6366/dsc00814w.jpg
http://img510.imageshack.us/img510/6058/dsc00816g.jpg

The path is only 2mm deep now though. In the real version it will be 8mm deep.

Will be taking a break from it over christmas. Just milling the block, etching it and soldering it together left now though :D

[x88x]

Like Kayin said, you want as much surface area as possible. What about adding crosshatch pins on the top and bottom of each channel? How small a cutting bit do you have on your CNC?

[Diamon]

The cutting bit is 2mm in diameter. Think it's the smallest I got.
What do you mean with crosshatch pins though?

[x88x]

Like, you do a sort of

_|_|_
_|_|_
| |
pattern, with the lines as cuts. Better yet, make them round and the full height of the channel, like these:
http://www.dangerden.com/store/image.php?type=D&id=623?1292448355001

[Lothair]

Having pins in the channels would certainly be interesting. You might have the benefit of both maximum surface area, but also higher pressure because of the fixed channel design, causing the coolant to move through the block more rapidly? Hmm.

[Kayin]

Pins in the channel will kill flow. It's a fine line there...

Let me see, I have a simple 3-d diagram here...

http://img.photobucket.com/albums/v310/KayinStorm/1.png

http://img.photobucket.com/albums/v310/KayinStorm/2-1.png

http://img.photobucket.com/albums/v310/KayinStorm/3-1.png

a simple pin grid can be created by slitsawing at 90 degree angles. Much more area for less overall effort.

[Konrad]

Rounded pins would increase flow. But how substantially?

[Kayin]

A fair bit, but simply doing one at a 45 would make diamond-shaped pins. And also be flow direction dependent, but that's neither here nor there

[Diamon]

That sketch looks nice. Maybe I should make a double layered block :D
First it goes through the maze, then the water continues down to the next level in the block which is a chamber full of pins. Would look pretty extreme with such a huge chunk off copper on the cpu. Hell, it would probably be enough to cool the cpu passively if I'm not doing anything demanding.

[Lothair]

Now that would be interesting.

[Kayin]

Konrad, I took the liberty of turning your idea into a working design.

http://img.photobucket.com/albums/v310/KayinStorm/konrad5.png

http://img.photobucket.com/albums/v310/KayinStorm/konrad1.png

http://img.photobucket.com/albums/v310/KayinStorm/konrad3.png

http://img.photobucket.com/albums/v310/KayinStorm/konrad4.png

http://img.photobucket.com/albums/v310/KayinStorm/konrad2.png

The three holes in the midplate are for heatpipes. Currently 5mm, but could be made larger. 5mm is plenty though, for ammonia or Midel pipes.

Currently, those heatpipes would be great for RAM on video cards, but a revised top and mount and you could get a CPU block that could do chipset or MOSFETS as well.

All images copyleft KayinStorm Customs, Konrad, if you want have fun with the design.

[Konrad]

lol, thanx for the copyleft, Kayin. Your design is more complex than what I had in mind; I was thinking more along the lines of simple rectilinear blocks with pipes drilled through them, either parallel (as in your 3-pipe center block) or even staggered like "Olympic rings" ... a number of smaller circular holes would provide greater inner surface area than a large groove.

What is the purpose of your extra geometry and multi-piece assembly? To increase interior surface area?

My first idea would be to concentrate not on the surface area as much as on stimulating turbulence. Smaller diameter holes (say 1/8") tapped out as a screw (wide-diameter low-pitch V-threads) might "swirl" the liquid through it's passage. Alternately, some kind of machined texture (http://www.google.ca/images?q=machined+texture) could be used on interior planes. Screw machining/boring isn't really my thing, but the boys in the shop can figure it out easily. It can't be too hard to tap solid copper, it's fairly soft and ductile.

Something like this (http://www.frostytech.com/articleview.cfm?articleid=2281&page=2). Though serving as a solid copper waterblock/heatsink instead of as a heatpipe/fansink.

[Kayin]

Turbulence only gets you so far, and with heatpipes you have extra places to remove heat from, hence the extra geometry. Also, the plate serves to force the water through the pin grid instead of simply over it, increasing efficiency at the expense of some flow rate.

After machining, ideally the midplate's bottom chamber would be beadblasted for maximum surface area.

I have further work showing a CPU top and midplate, though I really need to refine my work. Let me hack at it another day or so and we could have a real idea.

[Diamon]

I've taken a look at other designs but since I had already made the program for the maze-style block I decided to mill it . And today I finally got an opportunity to do so (the total cycle time was 80 mins 0_o). Here's some pics of the block, just etching the insides, soldering it together and lapping it left now:

The maze
http://img211.imageshack.us/img211/2725/dsc00834s.jpg

http://img27.imageshack.us/img27/7752/dsc00836in.jpg

The lid
http://img3.imageshack.us/img3/2238/dsc00837wh.jpg

Note that the fittings will be at different heights when assembled.
http://img402.imageshack.us/img402/6203/dsc00839tt.jpg

I will test this block and then consider making another block similar to Kayin's design for comparison :D


-Diamon

[Kayin]

Did anything come of this?