The use of turbulence

[Mr Evil]

I have conducted a short experiment into the effects of turbulence on waterblock performance. You can see it here.

I hope you find this helpful.

[Dix Dogfight]

Nice going.
I'll try it next time i build a new WB.

[Brad]

10% is a big gain, do you think it would be possible to try differant patterns?

[Mr Evil]

Yes. I started testing two more waterblocks today: One with the same grooves as the last one, and one with dimples. I need to do more testing, but it looks like dimples don't work. I would also like to try grooves perpendicular to the flow, and changing the scale of the pattern. If you have any suggestions for other patterns I will try them too.

[Brad]

you know the sort of double cresent shape you get when two circles slightly cross over, you could do thta

[ondaedg]

Mr Evil, do you have any other pics that show a closer view of your waterblocks?

[Mr Evil]

Yes I do, but you can't see them until I finish the film in my camera.

[Cyco-Dude]

what about little bumps instead of dimples? thats kinda what koolance does to the base of their blocks; they have small raised triangular bumps.
or what about a wavy or uneven base? or a sandblasted base, or an otherwise really rough base. just some ideas...

[Mr Evil]

I could try using very rough sandpaper. I couldn't make a bumps though, at least not with the equipment I have.

[Cyco-Dude]

perhaps you could take a hand drill and make small holes (lots of holes, make it like a fine grid), so there are little 'islands' or bumps? *shrug*

[gogo]

I'll be the first to admit I have no exp. w/block design. but an idea struck me. (wait, that was the wife) as a way to check turbulence patterns, run some dyed alcohol in with water. (not in the system! well if you want too) maybe use one of those printer refil kits to inject the dye in a slow stream using low flow water,maybe hook it up to a tap to test flowor something. kind of like a smoke test for airflow. anyway just a thought.

[mkosem]

wow, for such a simple design that is quite a gain!

--Matt

[Mr Evil]

gogo, that's an excellent idea. I just need to make some waterblocks with transparent tops.

[Marco]

Very interesting. Would you provide information on testing procedure, your raw data collected, and more detail on how the block is built?

[gogo]

This looks to be turning into a lesson on the cost of R & D

need a way to accurately test each design w/o throwing every block on the proc. and then if it passes, throw the block on the proc, and see how it'll clock (heh, sorry that sounds funny). Otherwise there would be a lot of time hooking up systems, bleeding, etc. I could be wrong though. I'd be interested in helping, but I'm 180deg on the other side of the globe, and I lack a CNC machine. (yet, I want new toy)

[Mr Evil]

I have finished a second set of tests, this time comparing grooves and dimples. Results on the same page as before (here).

gogo: If you want to help, try making a test rig as similar to mine as you can, and try to verify my results. The same goes for anyone else who's interested: Good science relies on reproducibility.

marco: I think everything you may want to know about the testing procedure I use should be on this page. Do you really want the raw data? It's not very interesting. I will describe the processes I use to make the waterblocks more clearly when I get the photos developed.

[EMC2]

Gogo - tell your wife she thinks like a scientist (that's a compliment in this case) That method is how Osborn Reynolds (whom the Reynolds Number which is a measure of turbulence in fluid flows is named after) did his original experiments in the latter 1800s. He used a glass pipe and inject dye into the center of the flow while varying the velocity to see the affects of velocity on turbulence. This same general method is still in use today (a staple of college labs for introduction to fluid dynamics). Higher end research now uses special dyes and lasers for investigation into turbulence affects.

Attached is a rough drawing of what you would see when injecting ink into the center of a channel with fluid flowing with laminar and turbulent flow. Notice in the case of laminar flow, the ink will remain as a well defined stream as it progresses down the channel. In the turbulent flow, you can see how it begin breaking up into eddies and at some point down the length of the channel would be mixed into the water uniformly - just a nice reddish tinted water.

Remember this will occur even without the need any turbulence inducing surfaces. (which leads us into part two of this)

[EMC2]

As could be seen in part 1, given sufficient velocity, you will have turbulence without any added surface features. Until you get into very high Reynolds numbers, you will also have a boundary layer against the walls with low to zero velocity.

Thus, a way to use the ink injection method to investigate the affects of surface features on wall proximity turbulence, we need to do a few things.

First, you want slower flows than you would normally have so that you are either in the laminar flow stage or close enough that you have a thick boundary layer on the walls. This is so that you can more clearly see the affects of the added surface features.

Second, you want to inject the dye/ink at a point downstream from the inlet so that the local affects associated with the inlet don't interfere. You also want the injection method to be one that causes the smallest amount of alteration to the flow.

Third, you want an area of no surface features between the ink injection point and the surface feature test area to help as a control.


Attached is a picture of one way to do this, showing the various things discussed. Note you don't have to make entire water blocks to do this, you can make simple straight channels to save time and $, and only test in a water block those surface features that look the most promising from the initial experiments. You could even take a piece of metal and put multiple seperate channels in it, with different features in each channel. (the ink inlet could be done with a hyperdermic needle, inserted through a hole drilled into the back wall of the channel)

[EMC2]

* After having looked at Mr. Evil's latest test page *

Very nice method to do the test simply I would suggest you do a run with flat plate as a control.

Other suggestions offered for your current test method (since you aren't using a processor for the heat source) -

1) use PVC pipe instead of copper to eliminate any possible influence of the connection between the plate and the pipe (variance in the thermal properties of the weld) as well as any flow characteristic variances do to the pipe itself (dents, inner surface roughness, etc.)

2) use a non-stick pliable thermal pad for the interface between the test metal and "heat plate/source" to help eliminate the possible effects of variance in the thermal interface between the two, both due to mating surface features (flatness, smoothness, etc.) and possible variances in the thickness or coverage of the thermal epoxy.

[ondaedg]

you ever wonder if Intel or AMD is "using" us to do their R&D for future water cooled systems......

[Brad]

emc2, I love the last pic personally, it is possible for us to do. use an acrylic block to see it all.


as far as expensive r&d goes, dden should sell all their test blocks, as long as they can be mounted some way and don't link. I'm sure they'd make quite a bit of money this way

[Mr Evil]

EMC2: Thanks for all the information. It looks as though this project could turn out to be much larger than I had originally anticipated.

1) Good idea. I will see if I have any PVC pipe lying around (I'm sure I saw some somewhere...).

2) I don't think variations in thermal interface are a problem. I remount the waterblocks between each test, and I see very little change in the results. Also, a thermal pad has a relatively high thermal resistance (about 1.5C/W for one that size?) which would need to be measured and subtracted from what I measure.

[gogo]

Originally posted by Mr Evil
2) I don't think variations in thermal interface are a problem. I remount the waterblocks between each test, and I see very little change in the results. Also, a thermal pad has a relatively high thermal resistance (about 1.5C/W for one that size?) which would need to be measured and subtracted from what I measure.

I was thinking the only reason he suggested the pads was because they are more consistant than paste. After all, (i could have mis-read, or be confusing threads) but aren't you trying to see how different cannels/patterns effect performance? What difference does it make if the bock is warmer? The concern I had was not so much their thermal resistance, but that, in my exp. they aren't as touchy. i.e. adding a bigger fan to a HS with a pad may drop temps a degree C, but adding the same bigger fan to the same HS using paste may drop it 5C. Then again, what fun is doing this if your working with warmer temps.

[EMC2]

Correct Gogo The reason I suggested a thermal pad is for consistency reasons in your testing. The newer thermal foams give extremely repeatable and consistant results when used properly (controlled by contact pressure). They also conform to surface contours very well, helping eliminate inconsistencies caused by surface features.

From a recent work project, some of the best currently available materials are available from Thermagon. In your case, either their T-Pli or T-Flex series would work well (depends on the how much pressure your test devices can handle. The T-Flex line is VERY good if you need low pressures and/or something that conforms extremely well to surface undulations and would be my suggested type.) (Note also the thermal impedance goes down with increased pressure).

Some numbers -

T-flex 620 @ 10PSI - 0.46 C-in^2/W
T-Flex 620 @ 50PSI - 0.30 C-in^2/W

T-Pli 210 @ 10PSI - 0.18 C-in^2/W
T-Pli 210 @ 10PSI - 0.13 C-in^2/W

You can get complete spec sheets on the stuff from their website, with complete curves of Thermal Impedance versus mounting pressures for the various thicknesses (go to their Data Table link for the curves). Oh - and their #%^_+@! site only works properly with IE.

[BillA]

another method (the best I've found):



a description here

makes the 462 hard to beat