srr2
join:2001-12-20
Bethlehem, PA
 ·RCN CABLE
2 edits | reply to Solitude
Re: Computer Noise Unbearable, Watercooling Solution, Help Pleas
I've never been a fan (no pun intended) of installing fans willy-nilly to cool a PC. Too often there's no thought into what the fans are doing, where they're located, where the airflow is, and the location of the heat sources. Most people who throw fans in wholesale don't even know how much power they are dissipating in the case. Do you?
First thing I'd do is measure the actual power that's being dissipated in that case. That will tell you what kind of airflow will be necessary to maintain a given temperature rise, at least in a gross sense. Get yourself a Kill-A-Watt. You can find them for under $25. It'll be among the best $25 you'll spend. Then, you can start to address the problem with some factual foundation, rather than conventional wisdom.
The methodology to follow is to run the computer in its most minimal configuration possible and measure the power consumed with the processor doing nothing. Then load up the processor to 100% utilization and measure that. Try to do the same thing with your video card, take a measurement with it doing nothing, and then with the most demanding load you can create. Next, add in other components one by one and find their contribution to the consumption. Finally, load it up to its maximum load scenario and find the maximum power.
Once you have this, you can make a crude schematic of the heat sources in the case which will suggest where you should try to have air flowing.
You can use this formula to approximate airflow requirements:
CFM = 3.2 * Watts / Temperature rise in Deg F
It's okay for a first-order approximation. Of course, this is the aggregate temperature rise from inlet to outlet, so you need to be mindful of local hotspots, whatever they may be.
I have a 3 year old socket-478 based system. Four HDs (160, 250, 400,400), 3.4GHz Northwood processor, 2GB memory. Couple of add-in cards, and the usual assortment of peripherals. It is barely audible. I used sound-absorbent foam inside to control noise, and also to constrain airflow to where it really matters, for example, there are foam panels that fill the empty space on both sides of the drive cage. Air entering the front of the case has only one way to go, and that's past the drives. Advantage? No need for separate fan for drive cooling. Drives sit at a constant 35C, 24/7. You can't hear them at all, even with your ear right at the front of the case. Then in the back, I've blocked off all unused openings. That leaves me with one opening which is occupied with an 80MM 28CFM Arctic Cooling constant speed fan. This spins at a very low speed (1900 RPM) with steeply pitched blades for low air noise. It also has ceramic sleeve bearings for no bearing noise. Ball-bearing fans make quite a racket, you know. Those bearings are not quiet at all. The ceramic bearings in the Arctic Cooling fan will last forever without lubrication, and run silently the entire time.
On the processor there's a gigantic Thermalright heat-pipe kind of heat sink with another Arctic Cooling low-speed fan mounted on it. This fan is temperature controlled, responding to internal case temperature. The assumption here is that the heat sink is so large and efficient that I'm not going to worry about absolute processor temperature. As long as it stays to safe levels (which I consider as under 55C worst case) then all's well. So, as the temperature in the case rises, that fan speeds up because more airflow will be required to carry off the heat in a higher ambient.
Finally, there's a Seasonic 400W PSU with a temperature-controlled 120MM fan pulling air off the top of the processor. I've yet to see this fan move off its idle speed.
At idle it all consumes 110W and the processor sits at around 40C, at full load (182W input) in a warm room it will reach 53C. Airflow in the case is designed to move in from the lower front and exhaust out the upper rear, partly through the power supply and partly from the 80MM fan. Using a sheet metal nibbler, I removed all restrictions at the fan outlet. In place of the perforated sheet metal I used a simple low-restriction wire fan guard. The idea here is that everywhere you expect air to flow, you remove as many impediments as possible. This decreases back pressure and reduces noise. Also, you take advantage of natural convection as much as possible and arrange for cool air in the bottom and warm air out the top.
Two more thoughts... First, the idea that it's necessary to install push-pull fans is tenuous at best. The only compelling reason for that sort of thing is the need to overcome high static pressures. However, with a little thought put into airflow and some judicious case mods to create openings for airflow in strategic places, you should be able to reduce restrictions to a level that a single stage fan architecture will work. Second, I question the "lower is always better" mantra. In an ideal world, maybe, but if it takes another 25dB of fan noise to lower temperatures another 5C below already-acceptable levels, why do it?
[On edit: Added power numbers, corrected typo, added following paragraph]
The power numbers above should be a cautionary note for those running distributed computing applications. Those clock cycles are NOT free! From idle to 100% load costs an additional 70W in my computer. That's 1.7KWH/day. At $0.10/KWH, $0.17/day or $60/year. It would make a lot more sense to me to send them the $60 contribution to run their application on a high performance number cruncher where that $60 would buy 10X as many FLOPS. |
