View Full Version : 4 peltier setup
blackshadow
03-03-04, 10:46 PM
I have an idea of possibly using 4 tec's side by side to cool the processor. I know that i would need a pretty thick cold plate to spread the heat well as well as a powerful power source but i was wondering if anyone had any thoughts about it.
Prandtl
03-04-04, 09:00 AM
One Tec is already big in respect to an AMD core, so I really dont see the point in using 4 side by side.
L337 M33P
03-04-04, 11:30 AM
You would run into serious difficulties with expansion and contraction of the cold plates with respect to the peltiers. It's just about OK with one peltier, but 2x2? It's asking a lot of the little junctions which are the ones that bear the stress.
squeakygeek
03-04-04, 10:00 PM
One Tec is already big in respect to an AMD core, so I really dont see the point in using 4 side by side.
The point is better temps.
You would run into serious difficulties with expansion and contraction of the cold plates with respect to the peltiers. It's just about OK with one peltier, but 2x2? It's asking a lot of the little junctions which are the ones that bear the stress.
Can you back this up with numbers? How much will the coldplate contract? How would this be a problem? And what do you mean by little stress bearing junctions?
Are you planning on watercooling these? Are you planning on using 226 watters? Done properly, I think this could work well, although very inefficient compared to phase change. I'd like to see this.
Prandtl
03-04-04, 11:45 PM
I know you want better temp, but there is better way of doing it and it is call stacking! Having 4 pelts side by side will requires a hell of a cold plate to use all that surface. If you are to use multiple pelts to cool one cpu, then stack them, dont "spread" them.
But then again, you need two different pelt to do it. I remember you found a supplier for 360W pelts Squeaky, combine with a 170W pelt, that will work better than that 4 pelts thing, since the maximum achievable dT for stacking is the sum of the dT of each TEC (or close to).
Actually, 4 pelts side by side might work better, it really depends how well a very thick coldplate (i'm thinking 3/4") distribute the heat between the 4 pelts... But then again, you will need a very big wb to cool them and added with the very thick coldplate, it will be very heavy.
squeakygeek
03-05-04, 12:57 AM
Originally posted by Prandtl
I know you want better temp, but there is better way of doing it and it is call stacking! Having 4 pelts side by side will requires a hell of a cold plate to use all that surface. If you are to use multiple pelts to cool one cpu, then stack them, dont "spread" them.
But then again, you need two different pelt to do it. I remember you found a supplier for 360W pelts Squeaky, combine with a 170W pelt, that will work better than that 4 pelts thing, since the maximum achievable dT for stacking is the sum of the dT of each TEC (or close to).
Actually, 4 pelts side by side might work better, it really depends how well a very thick coldplate (i'm thinking 3/4") distribute the heat between the 4 pelts... But then again, you will need a very big wb to cool them and added with the very thick coldplate, it will be very heavy.
Stacking won't work because with each peltier you add, the heatload on that peltier will be huge. Stick with putting them side by side. Theres a formula used for predicting what the temp will be, but I don't have it in front of me so I can't do the math.
How about using a 40mm cube of copper as your coldplate, and use 4 120 watt, 40mm pelts? Four small waterblocks cooling the outside of the TEC assembly would be a pain, but doable.
Probably easier to insulate too, with the waterblocks surrounding the outside.
It'd take a lot of custom machining to do.
480 watts of Brrrrrr.
The problem with using a larger coldplate (by larger I mean to hold 4 TECs on one surface) is that most of the heat will go directly above the CPU and not to the outer areas of the coldplate. A thicker cold plate helps a little but not as much as you would think. I did some thermal modeling of this, best way is to make the area directly above the CPU the coldest spot. Which would imply stacking TECs, phase change or water cooling. I know many don't think much of stacking TECs, but it can be done, it's not as easy as other methods. It all depends on how cold you are calling cold.
A cube might work better than a coldplate, it would be a truck load more of metal and weight. A while back someone had suggested a pyramid shape this would reduce the weight but would require more machining. I should do some modeling on this and see if it has any value.
Prandtl
03-05-04, 11:44 AM
fast calculation for stacking.
lets assume a heat load from cpu of 110W
first, the 170W pelt : dT max is 60°C,
with (170-110)/170*60 = 21°C is actual dT of that pelt.
now the 360W :
heat load is 110W (cpu) + 220ish (the 170W pelt) =330
(360-330)/360 * 60 = 5°C so total dT = 26°C
while with only one 226W tec,
(226-110)/226 * 60°C (55 maybe?) = 31°C
you'd need a +400W TEC for stacking to be worth anything, of course all those calculation are made assuming linear performance of the TEC wich is probably close enough to reality to give good estimate.
As for the 4 TECs, if you can find a good way of actually spreading the heat equally between the four. With 4 170W TEC and assuming that they individually "cool" 1/4 of the heat load:
110/4 = 27.5W each
(170-27.5)/170 * 60°C = 50°C max dT
theorically better than the other 2 options...
L337 M33P
03-05-04, 12:24 PM
OK so we do some modelling about thermal expansion. Using the assumption that the cold plate is thick enough to allow an approximately homogeneous distribution and removal of heat through the peltiers, the delta T would be 50C with 170W pelts, as calculated by Prandtl.
Asssuming the assembly was one waterblock and one coldplate with negligible distances between the peltiers screwed together at 20C at the edges, and the waterblock temperature was 35C, this would give a change in temperature of +15C for the copper in the WB and -35C for the cold plate.
The coefficient of thermal expansion of copper is 17x10^-6 /C. Substituting the length and width of the assembly as 80mm x 80mm and the temperature changes, we get 0.020mm extension in the waterblock and 0.048mm contraction in the coldplate.
Now unless your peltiers can move slightly, they will be subjected to a fair bit of stress along both the ceramic/solder joint and at the edges of the little doped semiconductor elements. I don't know how elastic ceramic is, but I'm willing to bet it isn't much :) - just be careful when mounting all the peltiers at once is what I'm saying.
If you used 4 separate (normal pelt sized) waterblocks instead of one huge one then you would be fine.
Originally posted by doc
A cube might work better than a coldplate, it would be a truck load more of metal and weight.
Actually, it would be less weight. Think of the cube as 4 10mm thick coldplates stacked. Minus the extra centimeter at the outside edges that would be needed to bolt the single cold plate assembly together.
If the single large coldplate were a 1/2" thick (ie. ~12.5mm), you'd be talking about quite a bit more weight than a 40mm cube, 1.95 times the size of a cube.
I haven't looked, but there might be more cost involved with ordering a piece of copper that thick, but maybe not. It would depend on their pricing scale, wether by it's by weight or uncommon size.
Hey Doc, are you the one with the tapered coldplate from a while back? Maybe it was muddoctor...Old age/memory sucks most days. :D
JFettig
03-06-04, 11:06 PM
Here is a good read:
http://www.ocforums.com/showthread.php?s=&threadid=108848
He did dual pelts on a vertical cold plate and got pretty good results.
I have stacked pelts and without a load I got about 10c out of it, but with lets say a 40watt load it would have been crap.
Theres a point when you have gone too far, that point I beleve is going past 2 peltiers. Seriously, how much more overclock are you gonna get? Probably none, I understand the 'cool factor' and all but sometimes its just overboard.
Jon
Diggrr, I think it was muddoctor. I have used various thickness's of flat coldplates. Stacking 4 10mm plates would require some thermal grease between layers this would cause some losses, don't know how bad it would be. When I get the time, I should add this to my list on thermal models.
"OK so we do some modelling about thermal expansion. Using the assumption that the cold plate is thick enough to allow an approximately homogeneous distribution and removal of heat through the peltiers, the delta T would be 50C with 170W pelts, as calculated by Prandtl." Posted by L337 M33P
Thermal expansion is not much of a problem here, the real problem for thermal modeling is to find out how thick for a flat coldplate, or what other geometries can be used without big time losses or maybe the gains of having more than 1 TEC. The trick is to think outside the box, but first you must know where the box is.
I didn't mean to actually stack 4 little colplates to make the cube, just to use that as a model to picture how much copper is in it. The fewer thermal junctions, the better.
A 40mm cube would be 64,000 cubic mm, and a 100mm square plate (allowing room for bolting together) at 10 mm thick would be 100,000 cubic mm, and require much more lateral heat transfer through the plate (ie, huge wastes of energy).
With a cube, the waterblocks could make allowances for bolting together at 2 edges. Plumbing the blocks would be the great pain though.
The tapered coldplate I alluded to was full size at the TEC, but tapered down to ~1" square at the cpu contact side. This made insulating pretty easy between the cpu's edges (which doesn't need cooling) and the plate.
That was ages ago, so I'd bet the pics he'd posted are long gone from the database.
Very slick setup.
*Silver's dual TEC vert plate was cool too, and a whole lot easier to build than my idea IMHO...:D
Meow Cat
03-10-04, 04:08 PM
stacking pelts, bad idea.
running 2x 226w pelts in parallel is preferred, not more than two pelts.
Reasons why you shouldn't be using so many pelts:
- temperature dissipation will be monstrous. how are you going to manage it?
- you will need more electric power, meaning more PSUs, meaning more electric bill.
Your best bet is:
- Use one or two 226w pelt.
- High clamping pressure between the modified cold plate and the base of the waterblock.
- Use one block, not separate blocks.
- Apply sub zero liquid temps to cool the pelts.
I get frost build up like in an ice cream truck all over the backside of the socket by applying a single 226w pelt unit & -5C waterchilled cooling. Even beats MachI like a easy pie. Very comparable to the performance of MachII.
JFettig
03-10-04, 05:50 PM
like easy as pie? I just had to point that out, how difficult is it to get a setup to use 2 226w peltiers and -5c liquid, what kind of money would you think this kind of setup would cost, how big, how much power would it suck etc. I think i would rather have a MachII
Jon
Meow Cat
03-10-04, 09:25 PM
i didn't say 2x 226w pelt with my wattachilla.
i said "single" 226w pelt + wattachilla....
i dun need 2x 226w to beat MachI... just 1 and my wattachilla method, give a good squash.
he.. he
It is possible to have multiple TECs and have low power consumption. Most people that use TECs, run them maxed out. This is OK, but if you are concerned about your electric bill as I am, then this is not the way to run them. On my chiller I run my TECs at less than 1/2 Vmax most of the time, the COP varies by how I have them running, but is about 0.8 COP most of the time. This means for every watt I put in to the TECs, I pump 0.8 watts heat out out of my coolant.
With multiple TECs, this allows me to use many different setups on the TECs with the use of a couple of switches. These switches change the TECs to operate in parallel or series and I have either 5 or 12 volts. If I really kick it down, I can get COPs greater than 1.
Diggrr, you are thinking of a hollow cube. I did the model for a solid cube, looks good. Let me get back to ya on the hollow cube, my first guess is that it would not be as good. For some reason I am unable to attach the image that my model has generated, if anyone has any advice on how to do this, THANKS!
Prandtl
03-11-04, 11:00 AM
something like this?
http://www.theforumisdown.com/uploadfiles/users/rheault/files/avatar2.jpg
you either use an external web uploader, in my case theforumisdown.com or you could use the "Attach file" option that you have when you do a "reply" (not quick reply) to a post (just under option). Also, you could send the file to me and i'll upload it with my account over @ theforumisdown. pm me, i'll give you my e-mail address if you want it.
OK here goes, I'm going to try to attach an image. If it is a success I will inclued some explanation.
It worked this time. The image is a cross section of a cube with a temp profile, the scale is on right. The full cube would be 50mm X 50mm X 48mm. The bottom or the side that mounts on the CPU is to the left side and is warmer in the middle. You can see the heat goes up the center and out to the sides as well as the top. Just for the heck of it I put TECs on 5 sides of the cube and set the temp on the cold side of the TECs to 10C, with about 90 watts of heat centered in the last face of the cube. In this case I used Al as opposed to Cu.
Prandtl
03-11-04, 04:26 PM
do you have something similar but for the temperature gradient or "q" if you can have it at one of the TEC face? What I am left wondering is if it wouldn't be better to cool the edge of a rectangle 20x40x40mm with 8 20x20mm TEC and one 40x40mm TEC on top. I mean, it is obvious by the simulation that you made that the temperature gradient in the cube is alot more important at the bottom half than the upper half (for the side face), so a big 40x40mm TEC won't be use very well that way.
In any case, good work with the simulation ;)
doc
Interesting
Do you have any values for the Contact Conductances(h) which gives the illustrated profile?
Here (http://www.ocforums.com/showthread.php?s=&threadid=150430) have played with crude models for Coldplate/Peltier, using Waterloo to calculate spreading resistance.
Am still very much unsure whether should use "h" of the heatsink.Any thoughts?
Prandtl, a cube was easy to do and it has perfect symmerty, I used 5 of the 6 sides with imaginary TECs holding the surfaces at 10C. So what I did here was used the same cube but took off 2 of the TECs on the sides. I'm not sure if I'm answering your question but this was easy to do ( I didn't have to make a totally new model). But anyway the view is the same as before, a cross section of the cube so the temp gradient can be seen inside as well as outside the cube where the TECs have been remove. If this is not what you were getting to let me know and I will make a new model when I get the time.
Les56, I'm not sure what you want. h is heat transfer coefficient used for gasses, fluids and sometimes solid to solid contact. They are not writen in stone, so they have a ranges like this:
1 to 30 W/m2 K for free air convection
10 to 100 W/m2 K for forced air convection
If the conditions are different in any cases h will vary. Say for example you polish your heat sink a little more, h might change.
What I do most of the time is ignore h for the first go around unless it is convection of gasses or liquids, then it can be found in a book or online. But for solids there is thermal grease between them aleast in most cases. So I guess I saying I don't have any good numbers for you. My models are simple and are only to show trends at this time. However they can be made more complex, but this requires more starting info as well.
Prandtl
03-12-04, 09:38 AM
yes, that's about it, but can you actually show the heat flux out of a surface? Or how much of the 90W of the heat load goes through each faces and top?
Heat flux would be alot harder to do with the software I am using. Instead of telling you what the flux is, you set up the conditions for a dT, and see if the temp is what you need it to be. However the heat flux will be proportional to dT at any point on the surface. But as you drawing heat from that area the temp will drop proportionally to the heat that is removed from that area. Not an easy problem to solve. So my answer is no, but if you have a specific model you would like to try, let me know and I will see what I can do.
squeakygeek
03-13-04, 11:49 PM
Hey doc, why don't you model it in copper, since that's what we're talking about using?
I will do a Cu based model, but I can guess at the outcome already. The temp at the heat source will be a little cooler. The only change will be the themal conductivity, the trend or look of the model will be the same, just different temps. What modeling is good for is to see what happens when you change geometries or things like that.
squeakygeek
03-14-04, 09:46 AM
Originally posted by doc
I will do a Cu based model, but I can guess at the outcome already. The temp at the heat source will be a little cooler. The only change will be the themal conductivity, the trend or look of the model will be the same, just different temps. What modeling is good for is to see what happens when you change geometries or things like that.
I see. This modeling thing is interesting. What is the software called that you use?
The software that I use is called TAS, which is short for Thermal Analysis Software. This was developed by Harvard Thermal Inc., they can be found at harvardthermal.com. I use TAS in my job frequently and do a few jobs on the side with it. I am using version 4, I think they are up to 6 now. Version 4 cost about $2500, I dont know what 5 or 6 cost, have not looked at the cost lately. So if you have something of interest let know.
AudiMan
03-14-04, 11:21 PM
Originally posted by blackshadow
I have an idea of possibly using 4 tec's side by side to cool the processor. I know that i would need a pretty thick cold plate to spread the heat well as well as a powerful power source but i was wondering if anyone had any thoughts about it.
Maybe you'd be better off with this. Jump on it while you can :)
http://www.ocforums.com/showthread.php?s=&threadid=280173
katanaman
03-18-04, 03:42 AM
I have thought about using four peltiers also.
The only thing that really stops me is the power consumption.
The power consumption for the two 172 watt peltiers I currently use is high enough!
My cold plate is around 42mm x 90mm and has two peltiers mounted on it. Each peltier is 40mm x 40mm in size.
My website: http://www.e-magic.co.nz/starlet/
To mount four peltiers I thought an "inverted T" shaped, thick, copper coldplate would be best. That is with two peltiers on each side of the vertical part of the T cold plate and a water block on each pair of peltiers.
Serious power consumption and bulk though. It would also need a huge radiator/fan setup too. I think the next step will be phase change for me. A Prometia Mach II looks too good to resist when the funds allow. :D
I modeled your inverted T cold plate for 2 40mm X 40mm TECs to show the trend and I had done this before. I asumed that all surfaces not used by the TECs or CPU would be insulated, therefore no heat flux. I made the TECs mount 24mm X 40mm X 40mm and the base plate 50mm X 50mm X 4mm, then cross sectioned it to see the heat flow and temps. This is what I have.
I redid the model but with 2 20mm X 20mm TECs on top. In both images (above and below) the TECs are holding the surface temp at 10C. This is what I got.
vBulletin® v3.8.7, Copyright ©2000-2012, vBulletin Solutions, Inc.