Sign in with
Sign up | Sign in
Your question

Building a Custom Waterblock-Updated[7/6/09]

Last response: in Overclocking
Share
a b K Overclocking
June 10, 2009 11:17:26 PM

I was wondering if any one would have any details/tips on building a custom waterblock or fluid/thermodynamics(, as in a new design for a waterblock. I have a $350 budget for this project. I have looked at the only block I know has been built by a individual modder/designer,Cathar(as a project/low volume block): The Cascade XS/XXX (and the Swiftech Storm is based off of that). I'm doing this for my project for the Siemens Competition (see: http://www.collegeboard.com/siemens/?CampaignID=4566 ). I have access to CAD and fluid/thermal simulation programs. I also have connections to get tops,etc Rapid Prototyped for low cost. And CNC milling for Copper, again for reduced price.

Note: my knowledge of Calculus is limited as I am only in 11th grade. So please list good references when it comes to formulas,etc. Most of the work will be done in CAD so I doubt there would be much hand calculations that I would need. I have looked at xtremesystems,etc but there weren't much info. I will post this at XS later. This will be my main pet project for the summer.

Thanks.
:bounce: 
a b K Overclocking
June 11, 2009 10:37:24 PM

^Thanks. I have seen that before. The problem is there are no info on design,theory etc to base my designs off of. Still searching...
Related resources
June 12, 2009 1:37:23 AM

i'll ask some of my engineer friends if they know anything
June 13, 2009 9:27:20 AM

The basic of it is getting maximum turbulence in the area of cooling (surface area) for minimum flow restriction.

The jet impingement design used by Cathar's Swiftech Storm in outdated. Too much flow restriction is it's main problem.
Popular/efficient design of today seems to come from two types; squeezing water through micro-channels as seen in EK Supreme and blasting onto micro-pins (w/o an acc. nozzle) then outting from the four corners as seen in Dtek FuZion (v1/v2).
June 13, 2009 8:14:17 PM

You're in luck, you don't need calculus to solve heat transfer problems like this - just algebra. If you're modeling a system like this you need to know the thermal interface resistance, the conductivity and thickness of your substrate, and finally the heat transfer coefficient (h) of your fluid flow. In simplest terms you find "h" of a flow (gas or liquid) in terms of something called the Reynolds number (Re).

The "heat transfer" article on Wikipedia has a good section on thermal circuits that you'll want to use to predict your performance.

All that said, my guess is that no matter what you decide in terms of your geometry none of that is going to matter...

As I said above your heat transfer is related to the Reynolds number of the fluid flow. One of the parameters in calculating this number is a characteristic length scale of the flow in question - like a pipe diameter or such. The flow channels inside of a water block are generally small enough that you can't get any help from that. The other important parameter in calculating Re is the velocity of the flow - or for a given cross sectional area the flow rate. Basically to have a good flow rate you need to have a screaming-fast velocity through your water channel. Generally this can only be achieved with a large pressure drop and most pumps that people use for water cooling are sissies.

You can't just take a large pipe though and weld it to a copper plate - you don't have enough surface area to transfer the heat across.

Now, if you're considering milling out a block that's got small passages in them that you're going to have to worry about cavitation in the water channel. Basically, liquids don't like turning corners. There's always a low-pressure region on the inside of a corner. If you turn too fast the pressure drops below the vapor pressure of the liquid. That's where the water will turn to water vapor even at low temperature. You don't want this to happen for 2 reasons. 1, the low-pressure gas regions such at heat transfer as compared to the liquid and 2, it blocks the flow trying to get past.

So, how are you going to fit a lot of surface area into your exchanger without turning sharp corners? How about a spiral path? Come in along the center and spiral your way to the outside. Find a really solid pump and figure out what sort of pressure drop vs. flow rate that pump can handle (the manufacturer's data sheet usually specifies this). Your fluid mechanics code can calculate the shear stress along the walls of your block. If you integrate this value (multiple the shear stress at each note by the area of that node and sum) you'll be able to calculate the pressure drop for any geometry and flow rate.

It sounds like you've got access to a good amount of simulation software. So use this software to optimize the geometries of the spiral channels (how big are they each, how thick a wall do you want in between them, how tall are they, etc.). If you can keep your Reynolds number high enough and have a strongly turbulent flow everywhere you'll probably want to have tall channels. Also consider using a ball-end mill to cut the channels as that will leave a rounded bottom where the heat can diffuse up to the top of the block.

If you do your design well you'll likely have lower thermal impedance going into the liquid than staying in the copper. This means you'll want the bottom layer of copper to be quite thin. It might be worth doing a stress/deformation analysis to make sure the bottom of your block isn't going to bow out under the pressure.

In terms of machining such a block you'll need to lap the surface then bring it to a near mirror finish if you want to get the best results possible.

In terms of your project, even if you pick a "simple" geometry like a spiral flow path there are several variables to be analyzed and optimized each of which will be dependent upon the others.

If you'd like more pointers to references, etc. just let me know in this thread. I've got a shelf full of thermal analysis textbooks but I'm not sure how good the online resources are. If there's a good university library near you it's a simple matter to visit the engineering library. Pull up their mechanical engineering course requirements. There's usually a heat and mass transfer course around sophomore/junior year. Whatever textbook they require will be in the library.

Good luck!

-Wick

p.s. though I haven't sat down to analyze any of the commercial designs using microfins or pins or arrays my gut tells me that none of them have a truly turbulent flow. There might be shedding vortices coming off the back of a step or something - but that's not truly turbulence. The length scale of the features they're making are very small to have truly turbulent flow. Instead the flow through the devices would be laminar - which is far easier to model using the software you mentioned.
a b K Overclocking
June 13, 2009 9:41:25 PM

Thank you both! Last year I read this book: Heat pipes by Peter D. Dunn for my project on the effect of fluid on heatpipe efficency. And used Re (and merit numbet
rs) for it. Most of the data,etc I pulled for my project was from that book. I wasn't aware Re can be applied to this kind of problems. Thanks Wick!

Quote:
Pull up their mechanical engineering course requirements. There's usually a heat and mass transfer course around sophomore/junior year.

I wish. I would have taken one hands down. Sadly those are only offered in College and I'm still in high school. ;) 

Wick, I would like to have any reference book names you can provide me with. I have already gotten my hands on this: http://books.google.com/books?id=VEZ1ljsT3IwC&printsec=...
Haven't read it yet (just got it today) as I am studying for the final exams for the year.
I have access to the George Mason library: http://magik.gmu.edu/cgi-bin/Pwebrecon.cgi?DB=local&PAG...


Quote:
In terms of machining such a block you'll need to lap the surface then bring it to a near mirror finish if you want to get the best results possible.

So basically, in the order of a few hundred microns?
June 14, 2009 12:03:53 AM

I saw you're still in high school. That doesn't prevent you from visiting GMU's engineering library. The heat/mass transfer class doesn't rely much on advanced mathematics (usually) but does assume a background in fluid mechanics to some extent.

The book you linked to is a thermodynamics book. That's a separate area of knowledge from heat transfer (often heat and mass transfer are lumped together because the diffusion of matter is often governed by nearly identical equations as the transfer of thermal energy).

I don't know what roughness spec. is called for in polishing the thermal interface surface but I'm sure Intel's datasheet will specify that. But flatness is more important than roughness imo. A bowlingball's surface can be smooth - but it wouldn't make good thermal contact!

-Wick
a b K Overclocking
June 14, 2009 12:30:17 AM

^I'll see what I can do about the access to GMU's classes.

Any links to tutorials,etc?

I'll pull up the Intel CPU/LGA775/i7 specs and see what I can find out.

Thanks again!
a c 324 K Overclocking
June 15, 2009 2:50:35 PM

Shadow, are you meaning lapping on the internal (water journals) or the external (mount surface)?

Not sure, but you could possibly see more heat dispersion if you have pins/impingements/increased surface area in the journals, like most commercially produced blocks. I would assume you want the mount surface to be lapped to, what, 1200 grit or so? ( I think you mentioned anything above that is irrelevant in cooling/conducting heat).

The more heat you can soak out of the CPU the better as long as you can also remove it from the block.
a b K Overclocking
June 15, 2009 4:18:51 PM

^I believe Wick meant both the internal and the external. And yes, any lapping above 1000 grit has no use because the TIM (esp. things like AS5) won't be able to penetrate the holes.
June 15, 2009 4:28:47 PM

My concern about lapping the surface of the heat sink was only for the exterior flat that interfaces to the cpu. On internal passages (where liquid is flowing) the only downside of having a rough surface is a bit of extra pressure drop. However, roughened surfaces will keep/transition a flow turbulent at a lower Reynolds number. So you may gain a bit of benefit from leaving any internal passages with tool marks from the milling process.

-Wick
a b K Overclocking
June 15, 2009 5:07:43 PM

Here is a sample run of a square pin matrix (6x6) block with 4mm between each pin. Front view:

Image at 95sec mark. As you can see, the turbulence is quite low for most of the area. Flow rate was set to 3 GPM (water) and the base was Copper.
a c 324 K Overclocking
June 15, 2009 6:24:33 PM

Regardless, you are leaps ahead of most users by making your own block. I give you major kudos for going that route and engineering your own stuff. I like that DIY approach!
a b K Overclocking
June 15, 2009 6:35:41 PM

^I was more or less inspired by Cather and this thread:
http://www.tomshardware.com/forum/forum2.php?config=tom...

And since it's Summer and I don't have much things to do, along with the Intel Science Fair and the Simens competition I thought why not? I'll be able to learn some thing (that I can use in the future) and may even be able to make some money (win regionals,etc).
June 15, 2009 7:58:10 PM

Shadow703793 said:
^I believe Wick meant both the internal and the external. And yes, any lapping above 1000 grit has no use because the TIM (esp. things like AS5) won't be able to penetrate the holes.


For external surfaces i.e. IHS-waterblock try Ceramique or TIM Consultants T-C Grease 0099 or even generic white silicone paste. They do a much better job of spreading thinly when crushed as most will be metal-to-metal contacts rather than TIM filled in between. With mine, both IHS and waterblock are lapped to 1200grit and using just 1/2 grain size of Ceramique reveals almost all of the paste has been squeezed out, so I could've used even less paste! The ultimate perfect surface requires no TIM at all, used by expensive lab equipments apparently.
Thicker pastes like AS5 (or the crazy thick Diamond7) is really meant for unlapped surfaces used by typical users and are very hard to spread thinly without using extreme pressure mounting.

For your square pin matrix, milled or forged?

To increase turbulence, the design of its surrounding area/enclosures of the cooling surface (pin matrix) contributes too.
A good example is Dtek FuZion v1(what I have) compared to v2. Apparently Dtek reduced the height of the pin matrix by almost 2mm plus small changes to the plastic chamber area above it, the result was higher turbulence (2C better on avg. with an OC'd quad) and slightly higher flow resistance.
a b K Overclocking
June 15, 2009 8:50:15 PM

^Eventual product will be milled (Copper C110), not forged, so yeah I have to keep in mind the limitation of CNC milling in the design. Thanks for the info on the TIM wuzy. Also the 6x6 square pin matrix was only simulated just so that I can have a base line for comparison. The sim ran only 95 sec (~45 minutes real time). I will run an extended test (10-30minutes, which would be a few hours of run time) over the week end. I have access to a few PCs so I may be able to set up the fluid sim for running off of those with out installing the Inventor (student license, no limitations) + Thermal Desktop (trial) on each PC.

I'll be building a few proto blocks like this:


Mainly to get a better understanding of jets,etc.

Since the top is going to be rapid proto typed what I could do with the top (design wise) is pretty unlimited (comapired to CNC milling, forging,etc). I could have some crazy stuff (ie multiple flow channels with in the top itself) done in terms of routing water from inlet to the Copper base and out the outlet.
a b K Overclocking
June 16, 2009 5:07:50 PM

Results for a 9x6 square matrix. Simulated time: After 10 minutes.


Pink line shows the water flow. The color gradient shows the intensity of the turbulence at 10minutes.
a c 324 K Overclocking
June 16, 2009 8:28:29 PM

That looks nice...are your inlets and outlets part of the simulation? Center and corner like a CPU, or same-side like most GPU blocks? Your prototype (above the graphic) looks like a diagonal crossflow design...corner to corner?

What about a central inlet with an offset outlet?

Another idea for your impingements:

keep similar impingement design over the central area 2/4 core region, create larger impingements on the outlying areas or even reduce their height to allow greater flow once the inlet passes from the center (cores) to the outer areas of the block.

a b K Overclocking
June 16, 2009 8:43:36 PM

Quote:
Your prototype (above the graphic) looks like a diagonal crossflow design...corner to corner?

Yes, it's corner to corner.

Quote:
keep similar impingement design over the central area 2/4 core region, create larger impingements on the outlying areas or even reduce their height to allow greater flow once the inlet passes from the center (cores) to the outer areas of the block.

I'll try that later this week. (W00t! 2 more days left before SUMMER BREAK!!!)
a c 324 K Overclocking
June 16, 2009 8:48:37 PM

I just thought it might be worth while to look at. I figured that most *good/great* block designs have a central/offset flow design so they might be on to something with the water coming straight in over the cores. As for the dual-layer approach to your design...I think it might work well. Most of the ones I have seen only employ a single design.
June 16, 2009 9:51:05 PM

Any design that allows the flow to go over top of the fins isn't going to do a good job of pulling heat out of the fin walls.

You mentioned that you can do some fun stuff on the inlets. What if the water comes in on the top from 16 tapered holes, impinges on the bottom plate (maybe you machine out cups for the jets to impinge into) then exhausts into the sides.

-Wick
a b K Overclocking
June 16, 2009 10:30:17 PM

^Kind of like the Swiftech Storm? Yeah, I could do that.
June 17, 2009 2:25:48 AM

I noticed all modern blocks have top part of the chamber touching the pins, so no free-flowing water over the top of the pins.

What's the dimension of each pin anyway?
a c 324 K Overclocking
June 17, 2009 2:38:21 PM

I only suggested the dual level pin approach to increase flow. You would still get the flow around the pins, but depending on the design you might have more or less surface area outside the center where the cores reside.

Impingement cups are a good idea but only if you have the nozzles to direct the flow into jets to spray into them, directly.
a b K Overclocking
June 17, 2009 4:05:32 PM

Quote:
What's the dimension of each pin anyway?

5x5x10 (lxwxh) in mm.
a b K Overclocking
June 18, 2009 4:11:46 PM

rubix_1011 said:
That looks nice...are your inlets and outlets part of the simulation? Center and corner like a CPU, or same-side like most GPU blocks? Your prototype (above the graphic) looks like a diagonal crossflow design...corner to corner?

What about a central inlet with an offset outlet?

Another idea for your impingements:

keep similar impingement design over the central area 2/4 core region, create larger impingements on the outlying areas or even reduce their height to allow greater flow once the inlet passes from the center (cores) to the outer areas of the block.

http://i286.photobucket.com/albums/ll98/gcarver2006/block_design.jpg

Running the simulation for your suggested design right now, initial results (~20 seconds) look promising.
a b K Overclocking
June 18, 2009 4:33:47 PM


^At ~20 seconds

The block:

a b K Overclocking
June 18, 2009 4:57:19 PM

I just thought of something, would having the inlet at a different angle (instead of 90) make any difference?
a c 324 K Overclocking
June 18, 2009 7:51:06 PM

So how'd they turn out vs. your initial tests?
a b K Overclocking
June 18, 2009 8:17:26 PM


Final results, at 40 minutes.

As you can see, there is more red-green areas meaning more turbulence in those areas. Overall, it's at least 5x better than the sample matrix sim. Also like said before, I moved the inlet a bit closer to the middle so that would have helped too.
a c 324 K Overclocking
June 18, 2009 8:55:28 PM

So...I assume lots of bright, pretty colors = better in this scenario?
June 18, 2009 10:00:31 PM

Shadow703793 said:
I just thought of something, would having the inlet at a different angle (instead of 90) make any difference?

The Gigabyte waterblock utilises such design, but I don't it benefiting pin matrix which works best with inlet blasting straight down the top on the pins on the area to be cooled.
With cross-flow design like Apogee (GT & GTX) it should benefit, but then GTZ steered away from that, clearly indicating pin matrix is better with above design rather than cross-flow as can be seen by the performance increase.

Theoretically angled barbs should also benefit blocks like XSPC X2O Delta which uses a cross-flow pin matrix design also. Reducing any tight angle within a loop reduces flow resistance.
June 18, 2009 10:14:49 PM

Your latest result looks promising, now think you've just got to narrow down the design of the upper chamber area.

Just a quick idea of what you should be aiming for:


... M$ Paint, rofl!

[EDITED]You may even want to do it with two chamber like Dtek FuZion. The upper chamber collects water squeezed out from the sides in the bottom chamber before outlet through the barb.
June 18, 2009 10:41:25 PM

when I get a chance later tonight I will post a design I came up with.

-ouch1
a b K Overclocking
June 18, 2009 11:14:46 PM

Quote:
So...I assume lots of bright, pretty colors = better in this scenario?

More orange/yellow/green = Better. More blue = bad.

Keep the ideas coming.
a c 324 K Overclocking
June 19, 2009 3:02:38 AM

I like wuzy's concept of the upper piece to the block. This would definately force all incoming water (given the intake is directly over the center) right through the tall, middle pins over the cores and then moving outward.

(Think of the movie "Independence Day" when the aliens park over the highrise tower and blast all the hippies. The blast then radiates outwards through the smaller buildings in waves.)

You may even want to incorporate smaller 'tops' to the larger/wider/shorter cubes around the perimeter...perhaps something to the scale of the narrow pins in the middle. This might give you some additional turbulence and surface area at the same time.
a c 324 K Overclocking
June 19, 2009 3:24:22 AM

...I just now realized who you switched avatars with...randomizer.

:/ 

Wow...
June 19, 2009 4:56:12 PM

Ok here is the block I put together. It is based off of the DD MC-TDX and a few others.
3d view:

Wireframe view:

I think it can use a little tweaking but I designed it using eMachineShop's tool. If you want I can provide you with the file in Autocad.

-ouch1
a b K Overclocking
June 19, 2009 7:04:04 PM

^Nice. send, CAD file to shadow703793@gmail.com
June 19, 2009 8:14:20 PM

Ok Shadow I sent the Autocad files to you.

-ouch1
a b K Overclocking
June 21, 2009 11:30:52 PM

^Still converting from 2D to 3D. Can you send me a STEP,etc file of that design? Measuring each point,erc is a bit tedious.
a b K Overclocking
July 6, 2009 5:18:28 PM

Hello all, another update.
This is a design VERY smiler to a GTZ. Infact, the matrix is the same height/#of pins,gap,etc as the GTZ and the inlet/outlet is about the same.

As you can see, solid temperature is quite even. There is less than .0001C difference from max to min.

On the other hand:

Turbulence is very low, meaning it's laminar flow.

Still working on ouch1's design.
a c 324 K Overclocking
July 6, 2009 5:59:40 PM

You might be seeing this due to the much smaller pins and lesser surface area per pin. Adding more pins with lesser surface area might be the same cooling result as bigger/fewer pins. Maybe modify your central pin structure, the ones over the cores?
a b K Overclocking
July 6, 2009 11:28:32 PM

^Adding more pins to the center is virtually impossible. There is .42mm gap between each pin.

edit: WTF? I'm not a Forum MASTER any more?????
a c 324 K Overclocking
July 7, 2009 12:32:34 AM

No joke...and I'm a newbie with 1300+ posts?
July 7, 2009 4:02:48 PM

Shadow,

I am not sure if you can go that small with the gap between pins. Especially with copper for the material. According to a machinist friend of mine (who works in a shop that makes heatsinks) you will risk the pins getting smashed during the machining of the block. I will check with him to see what the smallest you can go is according to his info.

-ouch1
July 7, 2009 4:38:10 PM

hmm. I did not see anything other than they are able to get precision of up to .0002 inches. But nothing about machining pins that small and that close together. I still think there is too much of a risk of the machine bending finished pins while cutting new ones.

-ouch1
!