notice how the inside wall of almost all channels has very little flow, this is obviously where a spiral is better.

The idea of running the block at an angle to cut cresents in it is really good, I'd like to see this done. Also, that picture shows a grey cylinder cutting into the block, what about having the inlet barb on an angle?

I'm just going to go with intuitive logic here, but you'd expect that with centrifugal force you will be getting faster movement around the outside rather than the inside.



You can run a sim like that for -any- .IGES design? Shit I need to bust out solidworks. That analysis is done in 3d right? So what happens, you produce cross-sections at different depths? What software is it that you use again? And does it take alot of training to use it? Or is it fairly intuitive? Because I might be able to gain access to it...

Furthermore, can you do something more than just the fluid flow patterns? can you factor in thermal characterstics as well? Or even just do thermal characteristics ignoring fluid flow?

www.digital-explosion.co.uk created a block that works on a whorle pool effect with pins over the cpu for increased surface area and turbulance. This would surely be a more efficient model as the only linear flow will be well away from the cpu.

MArco >The analysis is done in 3D. I prefer IGES format but I can take various formats from different CAD software.
When all the nodes are computed I can view where I want in any positions, the soft used is Cosmos Flow for the moment. I do other studies in order to use another soft called Fluent (the best in the world).
To use thes softwares knowledge of fluid mecanics is required, but Cosmos is easy to use (relative) but not Fluent :D

Thermal characterstics and others properties like radiation, etc can be used but it become very heavy and complicated. I realize some simple attempt to view how it works...
I do, in this moment, a simulation more precise for the maze2. have +7 hours of calculations and +300Mo in RAM so coupling Température should be very hard to do :( (Only Duron 700@1000)

For further informations take a look at : http://www.cosmosm.com

++

Ah yes, the "expense" (in computer time) of accurate CFD.

While you're looking at purely the flow side of things, you might find it interesting to do a little bit of look see at the affects of a channel's aspect ratio (height versus width) with respect to flow (but you need to use pressure for the boundary conditions on the channel, not volumetric flow rate).

To save yourself computer time, you can model a sub-portion of a block with a single rectangular channel that has only 1 bend in it's path (U-shaped section) and constant dimensions along it's length.

EMC2> I have finished the new simulation more precise than the others but with more computer time :D.

Pic about speed into a maze2 (red is high flow, blue is low flow)

http://membres.lycos.fr/roscool/maze2d.jpg

We can see well the "unsticking" (in blue) at each corner of the maze due to a high speed of the water flow. And as you say EMC2 I have modeled a sub-portion as the pic below to view if breaking the corner of my waterblock can improve speed at the wall near the corner and "down" the boundary layer.

http://membres.lycos.fr/roscool/corner.JPG

With this sub-portion I can put a big number of nodes (meshing) in order to have a good description of the phenomenom to view if a chamfer is necessary or not.

++

I got nothing but red X`s

None of Roscal's pics are showing up... just red x's. (even the first pics.) ....even with netscape.

Here's a thought, (looking at the last hi resolution maze2 pic) what if you re-design the channels, keeping everything pretty much the same, but make the blue areas (especially the parts after the sharp corners) solid metal, thicken the walls there so to speak, and see what the flow does then?

I thought this was the maze 2?

http://www.dorrellco.com/maze2.gif

Where are those images coming from?

If you guys are so worried about fluid velocity in your channels how come you use such large channels?

I personally made my WB by using the smallest milling bit the operator would let me. And I do not believe parallel channels are effective because surface area in a waterblock is not as critical as fluid velocity. (I did a quick heat transfer calc)

I believe the increased velocity by using a series maze style waterblock with many small channels in series overcomes the loss in flow rate due to the pressure drop over the waterblock.

I think velocity in the WB is important not flow rate.

To achieve this end I would suggest smaller fluid channels in series rather than different configurations.

jaydee, I was wondering that too ;)\

slipstream, thats all fine but you need to remember that mag drive pumps don't have much psi, so the water flow will slow right down

Isn't that the result that overclockers.com got with the gemini block with its small chanells, who out performed the larger chanelled maze 2 and single chamber swiftech.

Wonders how Roscal's latest number crunch is doing
 

JD - I just thought he was labelling his different test blocks consecutively, lol :shrug:

Slipstream - actually the main discussion here was affects of channel geometry to velocity and using CFD software ;) (at least that's been my take on it)

Brad - regarding the angled input... that would help lower the inlet restriction and decrease the side of the low flow area at the inlet, but an easier way to address the problem of the low flow in the back 1/3 of the inlet area might be to simply offset the inlet from the die's location so that area is outside where you are wanting maximum heat transfer rates (which is over the die). In the case of a spiral, that kind of fubars things though, so an alternative would be altering the inlet area's geometry to address the problem.

slipstream brings up a good point. Perhaps someone here can produce two blocks that we can test this out. The blocks would have to be of equal surface area. This would obviously be a constant.

The first block would be a high flow block with parallel channels. This could be done by using a design such as Xjinn's or a maze style block with multiple channels running parallel.

The second block would be a maze design that is one long maze using narrow channels.

The pump to do this should obviously be a fairly high powered pump. I have a 500 gph 7.5 psi pump that would be plenty strong so that it would not be a bottleneck when testing these. To put that into perspective, it can pump water straight up for almost 20 feet.

The high velocity of the narrow channeled block may indeed address the subject of this thread which is laminar flow and turbulence. I think it's worth a try and I would be more than happy to help make this work.

emc2, offset the centre, have a turbulant design directly over the core.

sounds good?

Nice to see some calculations/simulationss on this interesting matter.
One observation i've made is: All new blocks (maze,spiral.....) have a center inlet above the CPU core.
These simulations show that the turbulence on one side of the inlet is wery small. Wouldn't it be better if the inlet is slightly offcenter to have maximum turbulence over the core?
Just my 2 cents.

still dont see any of roscals pics damn...
still a very informative thread

If you want pics or anims go here (website was changed):

http://membres.lycos.fr/roscool/fotos/Simulation_num/

http://membres.lycos.fr/roscool/

I don't have time to continue studies and comparisons for the moment.. :confused: