# Design of a heat sink, combined with a pc cooling system

I currently have the task to cool a system of four small devices $\left(1 \times 5\; \mathrm{cm}^2 \; \text{each} \right)$ with a total heat dissipation of $150$ W.

My idea was to use a heatsink to transport the heat away (they are also connected to an optical system which should be separated from the cooling system) and afterwards either connect the heatsink to a GPU-cooler designed for $300$ W dissipation, or use a fan-/fin-system for dissipating the heat.

The main problem is to keep the devices at $30\;$°C at max. Which of my approaches is more suited for that, or should I completely rethink the approach?

Edit: I forgot to mention that three devices dissipate $10$ W, and the fourth device dissipate up to $120$ W.

• Really more of an applied physics question (but physics.SE dislikes applied physics/engineering). Tmax isn't the main parameter, dT is. What's the ambient temperature? If that's 31°C, no amount of passive cooling will work. Jul 28 '15 at 10:21
• @MSalters: Ambient temperature will be set at 25°C Jul 28 '15 at 10:30

With an average power of 40 watts per device (EDIT- See below), and a max temperature of 30C, and assuming the ambient temperature is 25 C, that's a thermal conductivity of .125 deg/W, which is extremely demanding. I would guess that you'll need water cooling for each unit. Either that or a set of heat pipes.

A fin/fan arrangement is unlikely to do well. Using CPU coolers as a baseline, you should check http://www.dansdata.com/coolercomp.htm which tested a large number of such coolers and presents results. About the best you get is 0.5 deg/W, which is only 1/4 of what you need. Another reason to avoid this approach is the potential for fan vibration to disturb your optical functions.

TEC coolers won't work well, either - the power levels are too high.

EDIT - With the more detailed information in the question, the answers change a bit. The 3 10-watt devices can reasonably be serviced by fin/fan CPU-type coolers, since a good one (with 0.5 deg/W) will nominally produce a 5 degree rise. The most likely problems I see are vibration control and the need for physical spacing of your devices, since these coolers are usually about 3 inches square, so you can't pack your devices very tightly.

A word of caution, though. Your coupling area (1 x 5 cm) is rather small compared to a standard CPU, and it may be too small to allow adequate heat transfer to the heat sink. Maybe, maybe not. You need to look into it.

The bad boy, at 120 watts, will definitely take something stonger, and a heat pipe/cooler arrangement certainly seems reasonable. If you go this route, be sure to select a heat pipe with a sufficiently high Qmax - that is, make sure its thermal resistance is rated for the power level you need. Generally, the bigger the pipe the more power it can handle.

Of course, if you go to heat pipe/cooler for the biggie, it might make more sense to run heat pipes to the other 3, but this will depend on packaging constraints, available airflow, device spacing, etc. Just a thought.

• How do you set the average power to 40 Watts? And which formulas did you use? Maybe we can set the maximum temperature to 40°C which could ease the cooling. Jul 28 '15 at 6:17
• Addendum: Using heatpipes like that de.farnell.com/webapp/wcs/stores/servlet/… and a CPU cooler designed for +10°C @200W, is that a possible cooling combination? The heatpipes reduce the vibrations on the optical part. Jul 28 '15 at 8:26
• @arc_lupus - For average power I took your 150W and divided by 4 units to get 37.5, then rounded up to 40 watts per device, not total. I've edited to reflect that. Yes, those heat pipes seem reasonable. However, remember that they are fairly rigid, and vibration will still be coupled. They need to be bonded/clamped to the devices. A 10C/200W cooler seems a fair choice, but this gives you a nominal 35 C device temp. You've indicated 40 C may be OK, but this suggests you're not certain of your requirements, and this need to change. Jul 28 '15 at 9:33
• I forgot to mention that the heat dissipation is not equally distributed, I'm sorry, question is updated. Our problem is that in our current application we are working with this dimension of heat, but we do not know if we have to scale it up later (devices are operating until a temperature of 50°, thus still a buffer). Therefore I want to overscale everything a bit to have a certain buffer in the background. Jul 28 '15 at 9:43
• @arc_lupus - See edit. And good on you for adding margin to your requirements. Jul 28 '15 at 10:01