# Water Cooling a Personal Computer

I intend to water cool my next PC. There are three areas of heat production in a PC - the CPU, the GPU and the motherboard itself. Would it be more effective to treat these three areas separately (with three cooling loops) or as one system? And by 'effective' I mean best cooling. My thermodynamics is woefully inadequate, but my hydraulics is pretty fair. (I used to design high pressure sprinkler systems for acft hangers many years ago)

If as a single system would it be better to run the radiators as a common plenum cooling the three heat areas via their own common plenum? Or, in true series - Heat unit, radiator, heat unit radiator, heat unit radiator, pump...

I use 'three' radiators because in the case I have chosen I can fit three radiators...

--How will you control fluid flow thruogh the three if in parallel? – Solar Mike--

Were I to go the parallel radiator route, I would take pump output and feed that into a common plenum, then have an outlet from the plenum into each radiator. Getting picky, each outlet from the plenum to the radiators would have to have the same length / restriction to be truly 'balanced' Taking this idea further, have the three radiators' outputs feed into a common plenum, then have each water block be fed from THAT plenum, and their output go into into a final plenum and from there back into pump. So, I would have pump out into radiator in plenum, then radiators and on to the radiator out / water block in plenum, then have a water block out / pump in plenum. From a hydraulic POV, the system is 'balanced' except for the water blocks. Some wee fiddling and measuring, find out all three water blocks true restrictions then add some to the least / second least and walla! paralleled and balanced. But is that the most effective way?

-"What are you doing with your computer that would require all this cooling? Did you modify your CPU to clock faster? – William Hird" -

As a semi-professional animator my PC spends much of its time rendering. You don't know how annoying fan noise can be until you run most of your threads at 100% Given that, I also don't skimp on my components. Two quadros produce a LOT of heat. I would like to cool them as well to keep the TCO down.

• How will you control fluid flow thruogh the three if in parallel? – Solar Mike Jul 16 '20 at 6:30
• I think you should draw a picture to help us understand the two setups as you understand them. Also, is your case big enough to consider either option? – J. Ari Jul 16 '20 at 15:41
• What are you doing with your computer that would require all this cooling? Did you modify your CPU to clock faster? – William Hird Jul 18 '20 at 17:57
• Get A Commercial Solution! Don't try homebrew. Really. All the big theories of efficiency and heat flow and balancing liquid flows are maybe 5% of the actual work. 95% is making sure it never ever leaks, and most homebrew solutions do leak sooner or later, and destroy the PC. The vibrations, the thermal stress, leaving it running for months without check-ups, that all leads to all except specifically dedicated solutions fail sooner or later. And if you get a commercial solution, the guys who designed it made all the calculations for you. – SF. Jul 20 '20 at 11:41
• Sure it can be more expensive than if you make it yourself. Is it more expensive than a replacement PC though? – SF. Jul 20 '20 at 11:43

## 2 Answers

Suppose that any one component has a power demand $$P$$ with a constant fraction $$f_q$$ going to heat output. The component hot temperature $$T_h$$ in air cooling $$T_a$$ will be approximated from the equation

$$f_q P = h_a A(T_h - T_a)$$

where $$h_a$$ is the convection coefficient and $$A$$ is the component area. For a given power load, to bring the temperature from $$T_h$$ to a desired temperature $$T_d$$, you will need to pull out a heat load $$\dot{q}_d$$ to satisfy the revised equation

$$f_q P = h_a A(T_d - T_a) + \dot{q}_d$$

This leads to $$\dot{q}_d = h_a A (T_h - T_d)$$.

When you water cool, that extra heat load is taken by heat exchanger equations

$$\dot{q}_d = U A (T_d - T_{wi}) = \dot{m}_w \tilde{C}_{pw}(T_{wo} - T_{wi})$$

where $$U$$ is an exchanger coefficient, $$\dot{m}$$ is the mass flow of water, $$\tilde{C}_{pw}$$ is the specific heat of water, and $$T_{wj}$$ are the in and out temperatures of the water over the component.

The maximum efficiency of the water cooling is when $$T_{wo} = T_d$$. Substituting back gives this general governing formula for a perfect system

$$h_a A (T_h - T^\star_d) = \dot{m}_w \tilde{C}_{pw}(T^\star_d - T_{wi})$$

You know the values of $$A$$, $$T_h$$, $$T^\star_d$$ (the desired temperature for a perfect system), $$\tilde{C}_{pw}$$, and $$T_{wi}$$. Estimate the value of $$h_a$$ and you can find the water mass flow you need in a perfect case.

Because the heat exchange cannot be perfect, $$T_{wo} < T^\star_d$$ and the actual $$T_d$$ will be higher than the calculated $$T^\star_d$$. The device will not cool as effectively. The trade-off that you might consider to increase $$\dot{m}_w$$ is not as effective as to decrease $$T_{wi}$$.

This is for one component. When you work multiple components in parallel, each component behaves as above. When you work multiple components in series, the $$T_{wi}$$ temperature of an upstream component is the $$T_{wo}$$ temperature of its downstream component.

If you ran the entire system in series, then every heat source would get the same flow, so you would have to design the flow for the heaviest use component.

Running the system in parallel is no different from a central AC system. These are usually balanced via dampers (valves for water) to restrict flow to the lower heat rooms. You would need a way to measure the outlet temp of the water from each heat source. An industrial system might have control valves to control the flow to different loops; basically a cooling tower system.