Cooling idea

Has anybody ever tried using either denatured ethanol, methanol, or 99% isopropanol as the cooling medium instead of water?

What happens if were to cool the fluid in a reservoir with say...dry ice or something?

http://www.xtremesystems.org/forums/...1&d=1144980291

NOTE: FOR THE TIME BEING, THIS IS STRICTLY AN ANALYTICAL STUDY!!!

This is the 90 degree elbow that I plan on using at the top of my "waterblock" design to get around the RAM clearance issue.

http://www.xtremesystems.org/forums/...1&d=1144980291

Here are the flow results for the elbow attachment:

http://www.xtremesystems.org/forums/...1&d=1144980833

This was about 650,000 elements, and a mass flow rate (water) = 0.223 kg/s. (~ 805 l/hr)

On a dual Opteron 246, it took about 41 minutes to complete.

Lol I had this idea as well if you try it let me know, but the best thing I can say is instead of the things you listed I'd recommend using everclear ( hard liquor ). As that will be gentle on your plastics and will not freeze at dry ice temps.

Quote:

---

Originally Posted by friedchips

Lol I had this idea as well if you try it let me know, but the best thing I can say is instead of the things you listed I'd recommend using everclear ( hard liquor ). As that will be gentle on your plastics and will not freeze at dry ice temps.

---

Ethanol will be liquid at dry ice temps. Hey Alpha glad to see you posting your work here. Keep up the good job. Cant wait to see what else your working on.

Reminds me of waterchilling. Problem is massive condensation hitting anything you decide to cool, plus the res and all the tubing. Also going to be bulky.

Not trying to sound down though, let us know what you're thinkin.

Quote:

---

Originally Posted by Absolute_0

Reminds me of waterchilling. Problem is massive condensation hitting anything you decide to cool, plus the res and all the tubing. Also going to be bulky.

Not trying to sound down though, let us know what you're thinkin.

---

Actually, to be honest, the idea for this method came to me while working with Cpt. Planet on his dry ice cooling stuff.

The difference is that instead of putting a "dry ice container" over the CPU, it's and having dry ice be the primary "heat removal" medium, is to cool a fluid that STAYS liquid at those temperatures and using that as the heat transfer medium.

Everclear is "technically" illegal in MI.

Quote:

---

Originally Posted by absolute_0

Reminds me of waterchilling. Problem is massive condensation hitting anything you decide to cool, plus the res and all the tubing. Also going to be bulky.

Not trying to sound down though, let us know what you're thinkin.

---

Well, I haven't worked out all the details of this idea yet, admittedly, but the premise is that if the working fluid (such as the listed alcohols) are already at the target temperature; then I want to pump that stuff into the circuit, and cool the CPU over time. (i.e. rather than having the fluid SIT there to do the cooling, the fluid will be constantly moving so and gradually it will cool it.

At least that's the idea. And another premise for this to work is that the rate of heat transfer and removal is greater than the rate of heat generation.

However, I will stress again that I won't be able to experimentally test this out; only analytically for the time begin (unless someone is willing to offer to help with parts and stuff due to funding issues (of course)).

But see, the analysis side of stuff is "cheap". It only takes time and energy (from that standpoint).

And the difference about this idea vs. water chilling is that I can take it as low as -70 C on the working fluid, and it will still be liquid.

The only concern is right now is that I don't crack anything due to thermal stresses. (i.e. it's great that I have a LARGE difference in temperatures because that's a big yay for heat transfer (Q), but <-- also worried and thinking about thermal shock as well).

Interesting how what appears to be a "simple" problem, really isn't that simple. And I'm modelling it purely analytically.

And I'm also still working with physical real estate clearance issues.

Here's an updated picture of the CAD model with a roughed in position for the 90 degree elbow.

http://www.xtremesystems.org/forums/...1&d=1144985200

Overall it's a good idea but why do you need an elbow? Calculating flow dynamics on the elbow is interesting but there will be plenty of other direction changes in a working fluid system such as in the block itself. Once you have a good block the main challenges would be liquid medium, xtreme low temperature pump, insulation, and accurate temperature monitoring on the block.

http://www.xtremesystems.org/forums/...ad.php?t=85550

do you mean like this?
looks verry cool :P

@ Extera, I was thinking that as well :)

I`ve been thinking about that kinda chilling, apart from the insulation, shouldnt be too hard?

*Famous last words* :)

K

Well i't should be a nice solution for a SLI/CF setup I think :D

Quote:

---

Originally Posted by G H Z

Overall it's a good idea but why do you need an elbow? Calculating flow dynamics on the elbow is interesting but there will be plenty of other direction changes in a working fluid system such as in the block itself. Once you have a good block the main challenges would be liquid medium, xtreme low temperature pump, insulation, and accurate temperature monitoring on the block.

---

The elbow is needed because in the idea the "coolant" delivery system is going to have major interference (issues) with the DIMM slots currently. (That is based on the DFI LP UT nF4 Ultra-D geometry; a board that I DO NOT own, but since that I figured that enough people use DFI boards for OCing that I might as well start with that and go from there.)

Presuming that it's for a tower (mid, or otherwise), with the DFI board, and having the board mounted in place, and the tower standing; the coolant is going to come from the TOP of the waterblock (relative to it's "as built" position), or side of the waterblock relative to itself, and come out the other side.

http://www.xtremesystems.org/forums/...1&d=1145010898

The idea is also to use gravity as an aid in flow such that the coolant stream internal to the block (heat source, herein referred to as HX) will be between 4-10 ft/s.

Like I said, unfortunately, due to RAM clearance issues (I've blocked off an area (volume, technically) denoted as the RAM keep-out (KO) zone. That means that under no circumstances can ANY part or geometry interfere or intersect that region. That should provide enough space clearance for mounts and installation (HOPEFULLY).

However, I do know right now that chances are, it will also mean that if you have RAM in DIMM slots 3 and 4 (on that board), after you install the cooling system for the CPU, you will not be able to get access to it without removing part of it. That's the only foreseeable downside at this point.

However, I can't honestly picture people running with this type of setup on a 24/7/365 basis, where they would need to constantly be changing the RAM.

Quote:

---

Originally Posted by Extera

http://www.xtremesystems.org/forums/...ad.php?t=85550

do you mean like this?
looks verry cool :P

---

No, not quite. You see how he has the acetone in a bottle and the dry ice somewhere else? My idea would have been to put the dry ice IN the bottom, and a lot smaller amounts too.

Using the amount of dry ice, and also controlling the voltage of the pump (i.e. output flow rate), that combined will control the temperature and also the RATE of temperature change. (i.e. it is possible that at some points, I will actually remove the dry ice and also stop the pump if it got too cold and try and see if the system will warm itself back up to temp.)

See what the other guy has is that he's trying to control temperature just by the volume that is delivered. Except that he doesn't have any other way to control the delivered temperature while maintaining a fairly constant flow rate throughout the system.

Another thing that I also find interesting is that most people do not necessarily start with ambient temps, put the system under full load, and THEN work their way down gradually.

With the system that I am currently doing the analysis on (and proposing) you start with a few pieces of dry ice to cool it about 10 C below ambient and go from there. (as opposed to nearly flash-freezing the CPU). The other problem that I see with his setup: it's a GPU. Most GPUs don't put out that much heat.

Therefore; that pot of dry ice that he's got is OVERKILL!!! For continuous run, I can probably go with less than 1/10th of that and still get results.

Quote:

---

Originally Posted by G H Z

Overall it's a good idea but why do you need an elbow? Calculating flow dynamics on the elbow is interesting but there will be plenty of other direction changes in a working fluid system such as in the block itself. Once you have a good block the main challenges would be liquid medium, xtreme low temperature pump, insulation, and accurate temperature monitoring on the block.

---

I do know that there is a chemical that you can use to coat all of the electronic components so that if condensation forms on the system, it won't short it. I forget the name of the compound, I will have to look that up.

As for insulation - none. I don't plan on insulating anything because I don't really care about heat transfer losses inherent in the system.

low temperature pump would be rather interesting. I honestly have no idea what I'm going to do for that (haven't gotten that far yet).

I would like to screw in a temperature either transducer or monitoring on the inlet and the outlet pipes of the block.

The reason for that is because by that point, I would have done SO many flow analysis on that thing that I wouldn't need to know the internal core temperature. If the outlet temperature is lower than what's coming in, we got issues.

*edit*

Correction:
read it wrong. He was using it for his CPU, NOT GPU. My bad.

Here's the latest revision of the concept/idea.

http://www.xtremesystems.org/forums/...1&d=1145024141

I don't have the flow analysis yet, but that's the first "rev" of the part in assembly.

(Hopefully that should help to clarify some questions).

Ummmm very well idea , ya CAD pics look very nice .
but can you try to made it ?

Its not a bad idea, but I would think the most important parts would be the internal design of the water block and getting a pump that will work at low temperatures.

Quote:

---

Originally Posted by mrhelix

Ummmm very well idea , ya CAD pics look very nice .
but can you try to made it ?

---

Can't make it due to cost considerations. (though I AM willing to accept donations).

Basically, right now, I have nothing that would be suitable for experimental tests/validation.

(See above posts.)

I don't have the board, the CPU, or anything else that would be required to run the system.

Although, if someone either can get access to materials and machining (CNC or otherwise), it is being designed for manufacturability as well and I don't see why it can't be done.

Pump specs are going to be based on an actual pump (although, theorectically, it shouldn't matter so long as it can function at low temperatures and also that it can provide the required flow rate).

Quote:

---

Originally Posted by Yeti2

Its not a bad idea, but I would think the most important parts would be the internal design of the water block and getting a pump that will work at low temperatures.

---

I am working on the "guts" of the waterblock right now. Just tried to generate the FE mesh before I left for lunch. Still working on the specific details of it so that I can get an initial "feel" for how this is going to perform.

Quote:

---

Originally Posted by alpha754293

Can't make it due to cost considerations. (though I AM willing to accept donations).

Basically, right now, I have nothing that would be suitable for experimental tests/validation.

(See above posts.)

I don't have the board, the CPU, or anything else that would be required to run the system.

Although, if someone either can get access to materials and machining (CNC or otherwise), it is being designed for manufacturability as well and I don't see why it can't be done.

Pump specs are going to be based on an actual pump (although, theorectically, it shouldn't matter so long as it can function at low temperatures and also that it can provide the required flow rate).



I am working on the "guts" of the waterblock right now. Just tried to generate the FE mesh before I left for lunch. Still working on the specific details of it so that I can get an initial "feel" for how this is going to perform.

---

Models are looking good. Mind sending me the latest version of it so I can check it out. I have the board and cpu. I think I might be able to get the block made . The internal design needs some work though.

Quote:

---

Originally Posted by Cpt.Planet

Models are looking good. Mind sending me the latest version of it so I can check it out. I have the board and cpu. I think I might be able to get the block made . The internal design needs some work though.

---

Running one of the preliminary flow analysis with just water as the working fluid.

NOTE: NOT actually looking for empirical results at the moment.

Because I had issues meshing the entire thing as a unit on my craptop, I had to mesh the components individually and then reassemble them just prior to setting up the case.

So, for the time being, I'm most concerned with making sure that it has a clear path of flow from the inlet all the way to the outlet and making sure that the domain interfaces are set up properly to reflect that, as well as the heat source interface.

The current run is sitting at nearly 2.0 M elements, and my dual Opteron has 335 MB RAM free (out of 2 GB), and also has a 1.55 GB swap file.

And that's the SIMPLIFIED model as well.

ETA to completion for the first run: ~50 minutes.

I have another one that I've got waiting on deck where I took out the heat transfer stuff, and looking specifically at flow path ONLY, so; more details about that once I get that going.

Here are the initial results from the run that just completed.

http://www.xtremesystems.org/forums/...1&d=1145057885

That is a vector plot of the flow velocity from 1.22-3 m/s. (~ 4-10 ft/s) which is used for ideal heat transfer.

The analysis took about 2 hours to complete.

That's still with the .223 kg/s mass flow rate for water.

I don't know why the heat transfer didn't really come into play. *shrug* will have to look at that more closely in a little bit and do a bunch more runs.

Unfortunately, based on THAT design (which is still preliminary BTW), it requires more flow in order to get higher velocities because the majority of the flow is between 0-1.5 m/s, which is too slow by my account.

The other problem with it right now is because the inlet is so high up that it has to wait until the fluid rebounds and then exits.

http://www.xtremesystems.org/forums/...1&d=1145058466

Well, the good news is that it appears as though the domain interfaces for the fluid flow is set up properly so that I don't have to worry about it all getting "stuck" in the inlet (for example).

Though, I am currently running the full blown version of it, and I'm doing the analysis with acetone at -70 C as the working fluid (because the properties for isopropanol aren't defined in the program and I don't really feel like taking the time right now to make sure that all of those parameters and the material definition is set up properly for this (or any other) analysis.)

Here are the preliminary results.

http://www.xtremesystems.org/forums/...1&d=1145150362

http://www.xtremesystems.org/forums/...1&d=1145150362

http://www.xtremesystems.org/forums/...1&d=1145150678

Latest batch of results is telling me that it's TOO cold.

For the majority of the analysis cases, the working fluid (tried with both isopropanol and acetone) doesn't get higher than a few degrees than what it went in as.

The highest temp I've been able to get right now is -12 C.

I've remodelling it some partially to clean up the case, and partially to give it a SLIGHTLY more representative version of the run itself.

The ONE interesting thing that DID come out of the analysis so far though is the design of the elbow isn't very great.

The sharp turn on the inside corner causes substantial losses in momentum to the point that it stalls. So, about upwards of 20% of the exit area results in a pressure drop that wants to suck fluid BACK in.

As a result of that, the following change has been issued (so that it would direct the flow over the center of the fins better).

http://www.xtremesystems.org/forums/...1&d=1145302621

After spending quite a susbstantial amount of time trying to get the heat transfer means, methods, and mechanisms to work properly (based on thermal energy dissipation, both as heat flux and as a point source); and the lack of decent results, I ended up switching it to an arbituary temperature of the source.

While the point source gave me reasonable temperatures (from that perspective at 260 K), the area that was affected by it is much too small.

I also switched the analytical working fluid from isopropanol to acetone because I know that the fluid properties for acetone are set up correctly within the program so that way I wouldn't have to worry or question whether the fault was in my definition of the fluid properties (to the best of my ability) for isopropanol or whether there was something wrong with the set up.

I also did not attempt to try to set the analysis up as a transient case (though I am fairly certain that it would have helped with the results). I looked at it briefly and then decided that I wasn't going to spend any more time trying to figure out what to do with it; and also to figure out the correlation between time and temperature for the different regions. (I was also worried that the case may not run because of hardware limitations (2 GB of RAM)).

Here are the global parameters of the analysis cases:

Inlet/Outlet: 90 degree elbow barbed tube fitting, 1/2" ID, 3/8" pipe size, from McMaster-Carr (P/N: 5372K324)

Block material: Aluminum, with initial temperature set at 293.15 K (20 C)

Working fluid: Acetone, initial temperature at 203.15 K (-70 C); and presumes to be externally cooled by dry ice (not modelled).

Heat source surface temperature: 293.15 K (20 C)

Note: This is done as a steady-state analysis, which also means that the result I get is a steady-solution.

In plain English: That means that there is a constant in flow of 203.15 K acetone into the system, and that the heat source is constant and I am looking for the flow velocity within the block (across the fins) and also the exit temperature and velocity.

For a heat flux of 200 W, I got 205 K (-68 C). Conclusion: Too cold.

For a point heat source of 200 W, I got 260 C (-12 C). While the magnitude of the temperature is correct, the area of effect is too small, and doesn't really represent what it would be like.

Tried setting it as a boundary source; that didn't work.

Ended up making the entire bottom surface 293.15 K (constant), which is different from having the initial temperature of the rest of the block at 293.15 K because the former means that it can cool down with the fluid (and it does, quite dramatically too), whereas the latter isn't "allowed" to cool down.

While that isn't technically truly representative of the real world either, it does get me a lot closer to where I need to be in terms of what works and what doesn't.

Another note is that in it originally model, I had it in a different orientation and that also affected the results slightly because of gravity.

So, in the 'C' revision of the prototype, the orientation of the entire assembly was changed/remodelled so that it would be in the "as installed" position. (Also part of that idea is to try and use gravity to aide in flow.)

Another note is that I also simplified the model construction to 4 parts (two of which are identical in geometry, was duplicated, and put into position).

(Original model had upwards of 15 parts/components to it).

Picture descriptions are as follows (in order):
Fluid velocity vector plot.
Block external temperature contour plot.
Fin internal temperature contour plot. Note the deeper blue which indicates that it is colder.
Fluid temperature vector plot.

Other conclusions that can be drawn from the analysis case:
- Using the streamline plot (i.e. much more indicative of the path of motion of the fluid), the majority of the fluid would just wrap around the two outside fins and not much would actually flow through the ones in between the center and the outside. That's bad, because that means that it's totally useless in having them; and doens't really do much for me. Tried to figure out a way to get the flow in there, but without adding another part to it (such as a diffuser or partially blocking the acetone getting it (think check valve)), it called for a design change/revision.

- The fluid typically would come in, remove quite a bit of heat when it comes in off the top of the fin, and then loop around (both directions), drop lower to the bottom of it, pick up a bit more heat, and then exit. Not efficient at all. I wasn't happy with the design, because a) it takes a lot of time for it to come in and leave. b) it does not remove heat from the hottest place, and as the fluid warms up, it becomes less effective for heat transfer; despite coming around to pick up more. Reason #2 for me not liking the design.

- The sharp corner in the elbow causes a sudden and rapid pressure drop ("crashing" of the kinetic energy/momentum). If it were air and it flowed like that, that would have been cause for concern because of shock wave formation. In any case - on the inlet, it SLIGHTLY alters the direction of the flow (about 2.5 degrees up towards the sharp corner). So, in that case, it has a tendency to want to go up and slightly to the right. On the outlet, it actually causes even more problems because the pressure drop is high enough that the fluid has a tendency to want to backflow INWARDS. BAD. (The analysis program actually warns me about that and automatically puts a wall boundary on a portion of it to prevent backflow.)

Other analysis details:

1.3 M elements
936 MB RAM requested/allocated
2 h 24 m for the run to complete.

Although I don't have a log of it, but I think that the swap file size was like some 2.45 GB.

Here's another design rev that I tried (prototype 'D').

http://www.xtremesystems.org/forums/...1&d=1145895547

It too, was also plagued with problems. Where the inlet stops and where the first wall is; there's less than 1.5 mm in there.

Ever taken a chemistry class in high school and they had the facets at each of the lab benches? (or anywhere really...)

That would be akin to putting your finger closing off the outlet and watching the water spray everywhere (and everyone).

Bad.

Major flow problems with that design as well.

Tried to do another design change/revision, but there are issues with it.

First off; I honestly really DON'T want to have the fluid running up and down with and against gravity.

However, due to real estate concerns, I do want to try and confine it to as close to the AMD spec'd out space allowed for a heatsink. Part of the reason why the block is so large is because it has to sit on above the RAM (DIMM slots 1 and 2) and that the outlet (on a DFI LP UT nF4 Ultra-D) would also render the x8 PCIe connector useless (given that I am not aware of too many cards where it would be short enough to fit both the outlet connector and the card).

Another note is that I also honestly haven't checked the clearance to the x16 PCIe connector either because I don't have that board (or any other board similar to it.)

Amazing how such a simple proposition can turn into such a complex design case.

And on top of that, I also don't want to make the geometry of the internals of the block to be too difficult (once again, the battle/debate of design for operation/functionality/efficiency vs. design for manufacturability kicks in).

Hopefully that I would be able to get another couple of runs in tonight to see where I can go next with it.

I already have prototype Rev. E on standby. Just waiting to upload the run/case to my system at home, and hit the 'Go' button. (est. 615 k elements)

Another couple of interesting notes:

The CAD importer is having issues so I am using the older version to import it and do the initial meshing. Then I save it and open up the file in the newer version and get it to re-mesh it. Typically results in 5-20% fewer elements, which means that it runs faster.

Also, I've noticed that with the newer version, it seems to be a lot more tolerant of errors that may occur within the geometry. (i.e. small angles, sharp corners, etc.)

This "little" project has also added 4.06 GB of data onto my systems. And this doesn't include any of the other components yet. (The JOYS THAT will be when I add the rest of the components and try to do the fluid flow analysis on that)!!!!

Here are the preliminary results.

460 k elements
294 MB RAM
41 m 26 s total time

http://www.xtremesystems.org/forums/...1&d=1145914276

http://www.xtremesystems.org/forums/...1&d=1145914643

Here are the results from a more refined case (same as the one above).

1.48 M elements
934 MB RAM
2 h 12 m 57 s

http://www.xtremesystems.org/forums/...1&d=1145976787

http://www.xtremesystems.org/forums/...1&d=1145976787

http://www.xtremesystems.org/forums/...1&d=1145976787