Water Blocks roundup of my own

[morphling1]

Thanks, Fixittt . Sandblasting is doable but I don't think it would help a lot in my case, you see my friend is using one final machining step that I don't know english word for, but the surface isn't smoth at all. I'll try and make on very closeup photo so you can see. As for my custom made block I use dremel and roughen inside surface of the block.

Ultra PeePi ,like EMC2 wrote it's the cross section that matters most and that's larger than the cross sectional area of 1/2" fitting (9mm ID).
What now everybody likes my custom made block more

Prices hmm, I have to checked with friend but informational prices for cnc blocks, actualy we named it M&B blocks (morphling & Boeing)M&B Al 45$, M&B Cu 65$ it's not just material, that almost doesn't count, but 2h longer machining (3h for Cu, 1h for Al block), they both come with 1/2" OD brass fittings, teflon washer, and all the nuts and bolts for instalation to mb.
For my custom block I would say I could make it for 60$, but now everybody would want that, and my time is very limited I'm home only on weekends until end of April, so I could make few blocks, but for CNC it shouldn't be a problem once we optimize process.
And Btw production block would have 4mm thick base. And of course I'll retest them with higher power AXP and posted results here and if somebody would give me Joe's address, or where I could send M&B block for test, I would really apreciate. Any other info you can email me: turbo1@email.si

[decodeddiesel]

Yes I am also very very interesting in obtaining one of these beauties. I am in CO USA. If you could PM me, or E-mail with a price and some ideas for shipping that would be awesome.

[morphling1]

Ok, people, here is close up pic. the best I can do with my cam.


You can see the how the surface of the chanell looks.

decodediesel, send you pm.

[Fixittt]

Boy that takes an hour to machine? WOW, if U notice most machine shops charge $60 an hour, that is $1 a minute. So right there your block costs $60 without materials or setup time. OUCH!

[morphling1]

The cost of tools and machining is just covered, and we can make a little money.

[morphling1]

I find another mistake, but this time it's benefitial. I was checking max thermal output of Duron model 3 (amd tech. specs. pdf), the one I had on test, and it's 85W in my case and not 70W like I wrote, so the results are even better, and don't forget the Duron core is ~20% smaller than Tb's.

[EMC2]

Originally posted by morphling1

And Btw production block would have 4mm thick base.

You might want to do some testing with the base thickness, in the 3,5mm to 6,0mm range before deciding on the final thickness

Here are a few charts to give you the reason why I suggest that:

[EMC2]

(edit: The graph labels have an error. They are both turbulent flows at different rates. The laminar flow numbers, not included, are significantly higher)

And here is a magnified view of the moderately turbulent flow:

[Fixittt]

umm the images dont werk

[EMC2]

??? Don't know what to say, they work viewed from here and were done as attachments to the posts (i.e. they are actually hosted by the Pro/Forums server)

[Brad]

maybe the pictures are too large (in kb) and so the server won't show them.

and they are working from your computer because they are cached

[EMC2]

Strange...they're 29KB in size (6K under the 35K limit on attached images). I just pulled the page up IE instead of NS and they do indeed end up blank.

Hmmm....wonder if it's a problem with IE not handling PNG images with a color depth of 4bits...

Here is one with the color depth changed back to 8 bits (increases file size to 31KB) to see if it is a prob with IE....

<EDIT> That appears to be the problem... the image below shows up in both NS and IE. It appears that IE doesn't handle a color depth of 4 on PNG format images (or the forum code doesn't identify it to IE in a way it likes). I just looked at the properties of the original images in IE (right click on the images) and they show up as "unknown", while the one below shows up correctly as "PNG". I went back and flushed the cache on NS and reopened this page and checked the original posts properties. NS shows them as being PNG as they should be and they were not cached before being displayed.

(repost of 2nd image with color depth of 8)

[Brad]

so 2mm is the best for you....

[EMC2]

The best plate thickness for the calculations done is in the 150mil to 250mil range, with the best value showing at 200 mil calcuation, which is 5,08mm btw (note: the numbers were only run at 50 mil increments of plate thickness)

The calculations aren't for a 'specific' waterblock. They were done for a wb design of the generic nature being discussed, where there are water channels, there is an approximate 2"x2" active cooling cavity (where water flow and thus heat transfer actually occur), and an Athlon XP die was the heat source.

To do the calculations, one of the parameters needed is the film coefficient, which in turn requires a channel length. In this case I used a channel length in the range typical for the channel designs.

Below is another curve created with the exact same parameters except that instead of turbulent water flow through the channels, the flow is completely laminar. If you compare it to the other graphs, it shows why laminar flow in a waterblock is not a good thing

[Brad]

ok, 5mm, not 2mm. WTF is this mil measurement then? hundreths of an inch or something?

do you think it would be possible to do a heatload of 2x 172mm pelts, over a 40mm x 80mm area?

[Volenti]

hrmm interesting, so are these charts based off theoretical or measured data? and if off measured data, how did you define and create the turbulent and laminar flow?

[EMC2]

Brad - It's one of those darn fangled US units
1/1000" (i.e. 1 mil = 0.001") 1 mm ~= 39.37 mils

Yes, it would be possible given more info... but I don't know that I'm so inclined at the moment

Volenti - a combination of both, although the model portions used are validated. Regarding the flow regimen : in this case a validated model for rectangular cross section channels. Used the physical parameters for pure water at the given temp (25C). The laminar/turbulent determination is based on the Reynold's number and done without any turbulence generating features added to the channels (i.e. purely flow rate dependent based on a relatively smooth channel wall with no turbulence inducing extras).

There's a fairly large body of research data in the area of altering flow regimen. ONE way is altering the surface of the channels properly. (properly here defined as in a way that introduces turbulence rather than one which induces laminar flow and does so without sacrificing effective surface area).

One further note: turbulence is a double edged sword, it is very good for heat transfer in this case but comes at the cost of an increased pressure drop (the amount of increase depends on the method used and the amount of turbulence generated <i.e. small scale or large scale>)

[Brad]

so if this is all theory, can you do some differant channel designs, and other similar alterations we've seen in waterblock design, and to see how well it will do?

[morphling1]

Again EMC2, great data, and to think I got you from hidding...

[EMC2]

Brad - not all. And yes, it's possible. There are a few "issues". Good data is needed on the blocks, a fair amount of time is needed to create some of the models, and depending on the design it wouldn't be a validated model so time and $ would have to be spent validating it. There are certain aspects that are very tough to model correctly.

Other things can be done with a fairly high degree of confidence in the results and with lower amounts of time/effort. By putting certain constraints on the desired results and limitting the scope of things I was able to give the gang here doing wbs some hopefully helpful info on one of the aspects of their designs (more or less point them in the right direction so to say) with only a few nights of work. From time to time I may do some other things in that regard. But I have neither the time, $, or inclination to go too far overboard


Morphling - no problem, hope it helps. Just didn't want to see you set that plate thickness and get less out of your block than might be possible. Oh, and when you do the testing to find the best thickness, do it for each material Copper and aluminum won't necessarily have the same sweet spot.


Which reminds me of another tid bit to ponder for block designs that have channels, fins, etc. embedded in the flow channel. While the lexan/acrylic tops may look cool, a nice top plate of copper/aluminum (depending on the base's material) which makes contact with the tops of the channels/fins is much better for cooling. It does two things, helps to distribute the heat energy better across the block and provides more cooling surface area

[Brad]

hmm, ok, I was hoping you could do things like take a maze2, and make the whole block higher, so the channels are deeper, then do a differance in base thickness's, etc

You might be able to sell this data to a manufacturer as well, take their block design, simulate it, and find out what is best

[morphling1]

I'll probably make one block with let say 6mm base thickness, and then test block get data mill down to 5mm, test. So on to 2mm this way I gotta find the optimal thickness. And yep Al and Cu results will be different that's obvious.
But I must disagree on importance of the top it doesnt help at all, I had plexy top on one of my prototype block but it leaked a little (no o-ring) so I soldered Cu top and the results were the same. You must consider that I use o-ring desing and that the contact between the top and the bottom isn't perfect at all, so heat transfer -> not good.

[EMC2]

Are the channel tops below the plane of the outside mating surfaces then?

(like this?)

||..................||
||...||...||...||...||
||...||...||...||...||
||...||...||...||...||

(ignore the ... in the pic, had to put them in for alignment of the rest)

As far as the testing, you might want to do 0,5mm steps. (or after you find the best using 1,0mm steps, use a 2nd block to explore the "sweet spot")


Brad, time & $ issues (more the former than latter). With current volumes and suspected margins in the wb business, I don't suspect selling the info is likely (just being realistic).

<edit: darn whitespace removal makes ascii pics a b>

[morphling1]

The tops of the inside fins are aligned with top plane of the block.
One more thing there is a difference in optimal thickness of the base if you're cooling smaller cpu (like duron or in the future thoroughbred) or larger (Tb,AXP), for smaller ones the thickness should be bigger, so the heat can spread through the active part of the block. So I should find optimal thickness for both kind of cpu's and then some compromise between them.

[EMC2]

Yes, the size of the heat source (CPU die) does indeed have an affect on the best base thickness. The data I provided was for an XP die.

You could always keep the thickest baseplate for 80% of your stock, and the thinnest for the other 20%. That way you can always mill down the thick base blocks as the need arises