CPU Cooling Project: Research, Testing & Analysis

[mbengtson]

Hello,

I am currently engaged in an High School chemistry project. I was wondering if you guys could help me out.

The project itself is not required to be related to chemistry at all; it can be a physics/biology project as well. After some brainstorming, I have narrowed my topic down to CPU cooling. I plan on researching water cooling and doing my project on the different coolants and the proportions of the actual coolants' chemicals to hopefully come up with a valid conclusion. For those of you who aren't familiar with water cooling, it's basically a bunch of tubes inside of your computer that are filled with a coolant. This coolant is what runs across the surface of the CPU and dissipates heat through a radiator.

So far, all of the research has led me to the conclusion that the study of this 'heat transfer' is basically centered on thermodynamics. I, however, have absolutely no idea what some of this physics jargon means. I am in the middle of teaching myself some of it from the basics, but was wondering if some of my questions could be answered here.

One of my questions relates to the actual testing process. I have 5 old pentium 2 300 mhz computers. I am trying to figure out the most efficient/effective way to do the testing. I plan on tesing each of the coolants on each of the (identical) computers twice, yielding 10 data points for each different variable. The computers will be running for 1 hour, and data (regarding CPU temperature) will be collected at the 0,5,10,30,and 60 minute marks. With the cleaning and setup involved, each different coolant will take 3+ hours to test (2 hours testing = 10 data, 1 hour cleaning, setting up). Does this seem viable?

My next question; is there any linear relationship between chemicals that demonstates the heat dissipation properties? I am trying to think of how I will be able to calculate/predict how much heat is going to be dissipated if I change the mixture (ie water + bleach) and/or proportions. I have looked over some thermodynamics literature, and here is a summary of the information that I have gathered;Well there you go! I'm a physicist! Haha thank you for any and all of your help!

I am aware of the danger involved with the CPUs overheating; if I run the water cooling pump for a few minutes before I turn the comp on, it will be fine. Thank You!

 

[Cyrus]

Can you delete your post in the Fluid thread.Go to delete, then highlight delete this post then ok.

Thanks.

 

[3trQN]

The computers will be running for 1 hour, and data (regarding CPU temperature) will be collected at the 0,5,10,30,and 60 minute marks. With the cleaning and setup involved, each different coolant will take 3+ hours to test (2 hours testing = 10 data, 1 hour cleaning, setting up). Does this seem viable?

How will you be measuring the coolant temp/cpu temp, and have you thought about the relationship between CPU load and thermal output?

Sounds like a fun project :)

 

[xw3850]

Some things you may want to keep in mind:

The internal temperature sensor in the CPU die (what the typical software app reads) is useless for this type of thing. It can easily have errors in the 10-15% range, and it's error can be non-linear. It's not really designed for performance measurements. It's more for preventing the CPU from damaging itself. For this application accuracy is not so important.

Although the computers may have the same components part numbers what not, they are not identical. For a part like a CPU the manufacturer only specs on Maximum and typical power, and they can be quite far apart. It's been a long time since I looked at a PII datasheet but I would not be even a little surprised if there was a 10% difference in average power for the same workload among your set of CPUs. Just keep it in mind when you present your results.

Also, as 3trQn mentioned, CPU load is key. I recommend some benchmark running DOS or Linux in single user mode. If you use a multitasking OS it will be difficult to make the amount of power to cool repeatable, which means you'll have to do a lot of extra averaging.

These are the big things. There are some other little things (such as controlling the ambient temperature) but this study does not need to be so professional.

 

[xw3850]

Oh, I don't want to discourage you from trying though. One can learn a lot from this type of experiment.

Since this is a homework forum, I leave figuring out the proper way to measure the CPU die temp to you. It's actually a very interesting problem and different techniques have different drawbacks and advantages.

 

[xw3850]

One more suggestion: Why cool a CPU? Using a fat power resistor or something would make this experiment much easier, and you could draw the same conclusion. Not quite as sexy though. :)

 

[berkeman]

Welcome to the PF, Mikyle. It seems to me that the single most important thing in this experiment will be to figure out how to best conduct the heat from the main heat-producing electronic components into the liquid that you are then running through an external radiator. Normal Personal Computer (PC) construction with planar Printed Circuit Board (PCB) arrangement is not conducive to liquid cooling. You obviously can't run the liquid directly over the PCBs and components, so what are you going to use? Some kind of soft conformal plastic bag things that press up against the top and bottom of each PCB? That will have to be very soft and conformal, given the varied topography of the components on the PCBs. Or are you only going to couple to the main heat-sink components with your liquid cooling arrangement, and still use forced air for the rest of the components?

As for more effective cooling liquid compositions, in high-performance motorcycle applications, we use Engine Ice or Water Wetter -- they make a significant difference in cooling performance:

http://www.cyclelogic.cc/

http://www.redlineoil.com/products_coolant.asp

 

[mbengtson]

If you google "DIY water cooling" there is an article on all of the basic components needed as well as instructions. There is a piece of hardware called a water block that allows the coolant to flow through the cooling system over the surface of the CPU itself.

As for the CPU differentiation in terms of accuracy, I could just do more testing, like maybe obtain 20 data points as opposed to the original 10 proposed. Thank you all very much for your help!

I will be using a program called Prime95 which severely punishes the CPU with ridiculous calculations. This will be running for the whole testing process.

Thanks again!